\\n>

⚙️ Run configuration

\\n> \\n> **Configuration used**: Path: .coderabbit.yaml\\n> \\n> **Review profile**: CHILL\\n> \\n> **Plan**: Pro\\n> \\n> **Run ID**: `1944988f-3de6-4ec3-9555-0b541ae16433`\\n> \\n>

\\n> \\n> You can disable this status message by setting the `reviews.review_status` to `false` in the CodeRabbit configuration file.\\n> \\n> Use the checkbox below for a quick retry:\\n> - [ ] 🔍 Trigger review\\n\\n\\n\\n\\n\\n

\\n

📝 Walkthrough

\\n\\n## Walkthrough\\nAdds a coworking MCP bridge, registry, tools, installers, IPC/preload wiring, renderer sync, and settings UI for per-agent browser/terminal access. Also introduces ephemeral browser partitions, excludes them from persistence/restoration, and adds a gated handler plus UI for clearing embedded browser session data.\\n\\n### Changes\\n\\n**Coworking MCP Server Integration**\\n\\n|Layer / File(s)|Summary|\\n|---|---|\\n|**Shared contracts**
`src/shared/coworkingBrowser.ts`, `src/main/coworking/coworking-types.ts`, `src/main/coworking/installers/types.ts`, `src/types/promise-with-resolvers.d.ts`|Defines shared browser operation and result contracts, coworking ids and bridge RPC shapes, installer interfaces, and the Promise.withResolvers type shim.|\\n|**Registry state**
`src/main/coworking/coworking-registry.ts`, `src/__tests__/main/coworking/coworking-registry.test.ts`, `src/__tests__/main/coworking/coworking-browser.test.ts`|Implements the session-scoped coworking registry for terminal and browser records, public id allocation and resolution, per-session permissions and confirm policy, and change notifications.|\\n|**Tools and session resolution**
`src/main/coworking/coworking-tools.ts`, `src/main/coworking/coworking-session-id.ts`, `src/main/process-manager/ProcessManager.ts`, `src/main/coworking/pid-resolution.ts`, `src/__tests__/main/coworking/coworking-tools.test.ts`, `src/__tests__/main/coworking/coworking-session-id.test.ts`, `src/__tests__/main/coworking/pid-resolution*.test.ts`|Implements coworking terminal and browser tools, resolves owning Maestro session ids from spawn ids and PIDs, and updates ProcessManager spawn env and PID lookup behavior.|\\n|**Bridge and audit**
`src/main/coworking/coworking-bridge.ts`, `src/main/coworking/coworking-audit.ts`, `src/__tests__/main/coworking/coworking-bridge.test.ts`, `src/__tests__/main/coworking/coworking-browser-bridge.test.ts`, `src/__tests__/main/coworking/coworking-audit.test.ts`|Implements the socket/named-pipe coworking bridge, JSON-RPC dispatch, handshake/session binding, browser operation validation, and redacted browser/terminal audit logging.|\\n|**Server script and installers**
`src/main/coworking/coworking-server-script.ts`, `src/main/coworking/coworking-server-paths.ts`, `src/main/coworking/coworking-installer.ts`, `src/main/coworking/installers/*`, `package.json`, tests|Adds the bundled coworking MCP server script and path resolution helpers, the installer orchestrator and per-agent installers, and the comment-json dependency used by installer config management.|\\n|**Main process, preload, and settings wiring**
`src/main/index.ts`, `src/main/ipc/handlers/coworking.ts`, `src/main/ipc/handlers/index.ts`, `src/main/preload/coworking.ts`, `src/main/preload/index.ts`, `src/main/stores/defaults.ts`, `src/shared/settingsMetadata.ts`, `src/renderer/stores/settingsStore.ts`, `src/renderer/hooks/settings/useSettings.ts`, `src/renderer/types/index.ts`, `src/renderer/utils/terminalTabHelpers.ts`, tests|Registers coworking IPC handlers, exposes preload APIs, starts and stops the bridge in app lifecycle, adds coworking settings defaults and metadata, and threads terminal coworking ids through renderer models and ProcessManager.|\\n|**Renderer sync and responders**
`src/renderer/hooks/coworking/*`, `src/renderer/stores/coworkingApprovalStore.ts`, `src/renderer/stores/coworkingBackgroundBrowserStore.ts`, tests|Synchronizes renderer session state to the coworking registry and serves main-process buffer and browser-op requests through renderer-side responders, approval logic, and background browser mounting.|\\n|**Renderer settings and tab UI**
`src/renderer/components/Settings/CoworkingSetup.tsx`, `src/renderer/components/Settings/tabs/EncoreTab/*`, `src/renderer/components/TabBar/*`, `src/renderer/components/coworking/*`, `src/renderer/components/MainPanel/MainPanel*.tsx`, tests|Adds the coworking settings panel and Encore integration, terminal coworking id copy pill, browser incognito badge and new-tab action, and BrowserTabView controls for clear session data and hiding tabs from agents.|\\n\\n**Ephemeral (Incognito) Browser Tabs and Session Data Clearing**\\n\\n|Layer / File(s)|Summary|\\n|---|---|\\n|**Ephemeral partition contract**
`src/renderer/utils/browserTabPersistence.ts`, `src/main/app-lifecycle/window-manager.ts`, `src/renderer/types/index.ts`, tests|Adds the ephemeral partition prefix helpers, the main-process partition allowlist update, and the BrowserTab fields needed to represent ephemeral and hidden tabs.|\\n|**Incognito tab creation and persistence exclusion**
`src/renderer/hooks/tabs/internal/browserTabHelpers.ts`, `src/renderer/hooks/session/useSessionRestoration.ts`, `src/renderer/hooks/utils/useDebouncedPersistence.ts`, tests|Threads an ephemeral option through browser tab creation and handler signatures, and filters ephemeral tabs out of restoration and disk persistence.|\\n|**Incognito UI and clear-session flow**
`src/renderer/components/TabBar/*`, `src/renderer/components/MainPanel/BrowserTabView.tsx`, `src/renderer/components/MainPanel/types.ts`, `src/renderer/hooks/props/useMainPanelProps.ts`, `src/renderer/hooks/tabs/internal/*`, `src/main/ipc/handlers/browser-session.ts`, `src/main/preload/browserSession.ts`, `src/renderer/global.d.ts`, tests|Adds incognito badges and a new incognito browser action, updates BrowserTabView and prop signatures for ephemeral tabs, and adds the clear-session-data control backed by the browser-session API.|\\n\\n**Estimated code review effort:** 5 (Critical) | ~120 minutes\\n\\n### Sequence Diagram(s)\\n\\n```mermaid\\nsequenceDiagram\\n participant MCPServerScript\\n participant CoworkingBridge\\n participant CoworkingRegistry\\n participant RendererResponder\\n MCPServerScript->>CoworkingBridge: hello (sessionId or ppid)\\n CoworkingBridge->>CoworkingBridge: resolveSessionFromPidWalk (if needed)\\n CoworkingBridge-->>MCPServerScript: hello ack\\n MCPServerScript->>CoworkingBridge: browserInteract(op)\\n CoworkingBridge->>CoworkingRegistry: resolve tab for sessionId\\n CoworkingBridge->>RendererResponder: requestBrowserOp via IPC\\n RendererResponder->>RendererResponder: approval gate + mount/focus tab\\n RendererResponder-->>CoworkingBridge: BrowserOpResult\\n CoworkingBridge-->>MCPServerScript: JSON-RPC response\\n```\\n\\n```mermaid\\nsequenceDiagram\\n participant User\\n participant BrowserTabView\\n participant BrowserSessionIpc\\n participant ElectronSession\\n User->>BrowserTabView: click clear session data (armed)\\n User->>BrowserTabView: click again (confirm)\\n BrowserTabView->>BrowserSessionIpc: clearSessionData(partition)\\n BrowserSessionIpc->>BrowserSessionIpc: validate sender + partition allowlist\\n BrowserSessionIpc->>ElectronSession: clearStorageData + clearCache\\n ElectronSession-->>BrowserSessionIpc: ok\\n BrowserSessionIpc-->>BrowserTabView: { ok: true }\\n BrowserTabView->>BrowserTabView: reload webview\\n```\\n\\n**Possibly related PRs**\\n- [RunMaestro/Maestro#710](https://github.com/RunMaestro/Maestro/pull/710): Both PRs touch `MainPanel`/`MainPanelContent` wiring, so this coworking integration overlaps with that area of the renderer tree.\\n- [RunMaestro/Maestro#785](https://github.com/RunMaestro/Maestro/pull/785): Both PRs modify browser partition allowlist behavior in `window-manager.ts`, including support for the new ephemeral partition prefix.\\n\\n

\\n\\n\\n\\n\\n

\\n

🚥 Pre-merge checks | ✅ 5

\\n\\n

\\n

✅ Passed checks (5 passed)

\\n\\n

\\n\\n\\n\\n\\n

\\n

✨ Finishing Touches

\\n\\n

\\n

🧪 Generate unit tests (beta)

\\n\\n- [ ] Create PR with unit tests\\n\\n

\\n\\n

\\n\\n\\n\\n\\n---\\n\\nThanks for using [CodeRabbit](https://coderabbit.ai?utm_source=oss&utm_medium=github&utm_campaign=RunMaestro/Maestro&utm_content=1160)! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.\\n\\n

\\n

❤️ Share

\\n\\n- [X](https://twitter.com/intent/tweet?text=I%20just%20used%20%40coderabbitai%20for%20my%20code%20review%2C%20and%20it%27s%20fantastic%21%20It%27s%20free%20for%20OSS%20and%20offers%20a%20free%20trial%20for%20the%20proprietary%20code.%20Check%20it%20out%3A&url=https%3A//coderabbit.ai)\\n- [Mastodon](https://mastodon.social/share?text=I%20just%20used%20%40coderabbitai%20for%20my%20code%20review%2C%20and%20it%27s%20fantastic%21%20It%27s%20free%20for%20OSS%20and%20offers%20a%20free%20trial%20for%20the%20proprietary%20code.%20Check%20it%20out%3A%20https%3A%2F%2Fcoderabbit.ai)\\n- [Reddit](https://www.reddit.com/submit?title=Great%20tool%20for%20code%20review%20-%20CodeRabbit&text=I%20just%20used%20CodeRabbit%20for%20my%20code%20review%2C%20and%20it%27s%20fantastic%21%20It%27s%20free%20for%20OSS%20and%20offers%20a%20free%20trial%20for%20proprietary%20code.%20Check%20it%20out%3A%20https%3A//coderabbit.ai)\\n- [LinkedIn](https://www.linkedin.com/sharing/share-offsite/?url=https%3A%2F%2Fcoderabbit.ai&mini=true&title=Great%20tool%20for%20code%20review%20-%20CodeRabbit&summary=I%20just%20used%20CodeRabbit%20for%20my%20code%20review%2C%20and%20it%27s%20fantastic%21%20It%27s%20free%20for%20OSS%20and%20offers%20a%20free%20trial%20for%20proprietary%20code)\\n\\n

\\n\\n\\n_{Comment `@coderabbitai help` to get the list of available commands.}\\n\\n\",\"path\":null,\"side\":null,\"line\":null}{\"type\":\"pull_request_comment\",\"body\":\"

Greptile Summary

\\n\\nThis PR introduces the full \\\"coworking\\\" subsystem: a session-bound MCP server that lets AI agents (Claude Code, Codex, OpenCode, factory-droid) read terminal scrollback and drive in-app browser tabs via a private Unix socket / named-pipe bridge. It also adds ephemeral (incognito) browser tabs, background webview mounts for cross-session access, per-agent browser-interaction permissions with a per-call approval UI, and a JSONL audit trail.\\n\\n- **Bridge & session binding**: a newline-delimited JSON-RPC server in the main process authenticates each MCP subprocess via a `hello` handshake (env-injected `sessionId` for Claude Code / OpenCode, PID-tree walk for Codex). Fail-closed semantics ensure agents can never read another session's terminals.\\n- **Browser tooling**: 15 new MCP tools (read, navigate, click, type, eval, screenshot, wait, open/close tab) are routed through the bridge to live renderer webviews, with approval gating, audit logging, and an opt-in background webview host for non-focused agents.\\n- **Persistence & settings**: ephemeral tab partitions, session restoration, and five new settings keys with a new Settings UI panel.\\n\\n

Confidence Score: 3/5

\\n\\nThe core session-isolation design is sound, but two synchronous blocking patterns in the main process will cause Electron event-loop stalls in normal use.\\n\\nThe audit sink calls fs.appendFileSync on the Electron main thread once per MCP tool invocation; under any sustained agent workload this blocks all IPC and rendering. The Codex PID-resolution path calls execFileSync synchronously during each hello handshake with timeouts up to 3 seconds. Both patterns are in the hot path of a feature expected to run continuously during agent sessions.\\n\\nsrc/main/coworking/coworking-audit.ts and src/main/coworking/pid-resolution.ts are the primary files to address before merge. src/main/app-lifecycle/window-manager.ts should also be updated to tighten the ephemeral-partition regex.\\n\\n

Security Review

\\n\\n- Partition validation inconsistency in `src/main/app-lifecycle/window-manager.ts`: the `will-attach-webview` gate uses a loose `startsWith` prefix while `clearSessionData` enforces a full regex.\\n- `browser_eval` is correctly gated behind per-agent interaction permission and per-call approval.\\n- Audit log redaction strips URL query strings/fragments and reduces eval code to lengths.\\n- Session isolation via fail-closed hello handshake prevents cross-agent terminal/browser leakage.\\n

\\n\\n\\n

Important Files Changed

\\n\\n\\n\\n\\n| Filename | Overview |\\n|----------|----------|\\n| src/main/coworking/coworking-bridge.ts | New IPC bridge server with fail-closed hello handshake; per-connection WeakMap state is correctly managed. |\\n| src/main/coworking/coworking-audit.ts | Uses fs.appendFileSync in the hot path of every bridge dispatch, blocking the main-process event loop on each write. |\\n| src/main/coworking/pid-resolution.ts | PID-tree walker using execFileSync synchronously on the Electron main thread during every Codex connection handshake. |\\n| src/main/coworking/coworking-server-script.ts | Bundled MCP stdio server; handshake and browser op routing are well-structured with timeout caps. |\\n| src/main/coworking/coworking-registry.ts | Per-session registry with stable browser:N id assignment and hidden-from-agent enforcement; looks correct. |\\n| src/main/ipc/handlers/browser-session.ts | Defensive clearSessionData handler with sender-type and full-regex partition validation. |\\n| src/main/app-lifecycle/window-manager.ts | Adds ephemeral partition support with a looser startsWith check inconsistent with the precise regex used in browser-session.ts. |\\n| src/renderer/hooks/coworking/useCoworkingBrowserResponder.ts | Renderer-side browser op executor with approval gating; waitFor loop could be hardened against mid-poll navigation. |\\n| src/shared/coworkingBrowser.ts | Shared BrowserOp contract types; ALWAYS_CONFIRM_KINDS correctly includes newTab and closeTab. |\\n\\n\\n\\n\\n_{Reviews (1): Last reviewed commit: [\\\"test(coworking): add VenetianMask to Tab...\\\"](https://github.com/runmaestro/maestro/commit/98eae338f23ee2b05ae94a13aeda78d397f83ce9) | [Re-trigger Greptile](https://app.greptile.com/api/retrigger?id=41434169)}\",\"path\":null,\"side\":null,\"line\":null}{\"type\":\"pull_request_comment\",\"body\":\" $\\\"P1\\\"$ **Synchronous filesystem write blocks the Electron main-process event loop**\\n\\n`fs.appendFileSync` is called in the bridge's hot path once per coworking tool call. On the Electron main process this blocks the event loop until the OS flushes the write, which pauses all IPC, window rendering, and other main-process work for that duration. Under any sustained agent workload (e.g., an agent polling `read_terminal` in a loop) this will produce measurable UI jank.\\n\\nReplace with `fs.promises.appendFile` wrapped in a fire-and-forget (errors are already captured via `captureException`); or batch writes through a queued async writer so individual calls never block.\",\"path\":\"src/main/coworking/coworking-audit.ts\",\"side\":\"RIGHT\",\"line\":null}{\"type\":\"pull_request_comment\",\"body\":\" $\\\"P1\\\"$ **Synchronous subprocess spawns block the Electron main-process event loop**\\n\\n`execFileSync('ps', …)` on macOS (1 s timeout) and `execFileSync('wmic', …)` / `execFileSync('powershell', …)` on Windows (2–3 s timeout) are called synchronously from the bridge's `dispatch` function during the `hello` handshake. This happens on the Electron main-process thread, which handles all IPC, events, and window repaints. A Codex agent connecting while the `powershell` probe is in flight (cold-start ~200 ms, worst-case timeout 3 s) stalls the entire app.\\n\\nConsider `execFile` (async) or moving PID resolution to a worker thread / `child_process.fork` so the main event loop is never blocked.\",\"path\":\"src/main/coworking/pid-resolution.ts\",\"side\":\"RIGHT\",\"line\":null}{\"type\":\"pull_request_comment\",\"body\":\" $\\\"P2\\\"$ $\\\"security\\\"$ **Ephemeral-partition validation in the webview gate is weaker than in the `clearSessionData` handler**\\n\\n`isAllowedBrowserTabPartition` accepts any string starting with `maestro-ephemeral-` (`startsWith` prefix match). The `clearSessionData` IPC handler in `browser-session.ts` enforces the full regex `^maestro-ephemeral-[A-Za-z0-9_-]+-[a-z0-9]{8}$`. Consider extracting the regex into a shared constant in `browserTabPersistence.ts` and importing it in both files to keep the two gates consistent.\\n\\nNote: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!\",\"path\":\"src/main/app-lifecycle/window-manager.ts\",\"side\":\"RIGHT\",\"line\":null}{\"type\":\"pull_request_comment\",\"body\":\" $\\\"P2\\\"$ **`waitFor` polling loop has no handle-validity check inside the loop body**\\n\\nThe loop polls `handle.executeJavaScript` every 150 ms for up to 30 s. If the tab navigates during the wait, `executeJavaScript` returns `false` (selector gone) and the loop runs to timeout rather than surfacing the navigation. Adding a URL-staleness check or `handle.getTabId()` identity check inside the loop would surface mid-poll navigations faster.\",\"path\":\"src/renderer/hooks/coworking/useCoworkingBrowserResponder.ts\",\"side\":\"RIGHT\",\"line\":358}{\"type\":\"pull_request_comment\",\"body\":\"**_{_{![P2 Badge](https://img.shields.io/badge/P2-yellow?style=flat)}} Clear stale active fields for new browser tabs**\\n\\nWhen an agent calls `browser_new_tab` while its agent is currently on a file tab or terminal, this state update only sets `activeBrowserTabId`. The old `activeFileTabId` or `activeTerminalTabId` and `inputMode` remain, so the tab strip can show both file and browser tabs as active, or terminal mode can keep the new browser hidden even though the tool reported it was selected. Match the normal new-browser-tab path by clearing the competing active ids and setting `inputMode: 'ai'`.\\n\\nUseful? React with 👍 / 👎.\",\"path\":\"src/renderer/hooks/coworking/useCoworkingBrowserResponder.ts\",\"side\":\"RIGHT\",\"line\":161}{\"type\":\"pull_request_comment\",\"body\":\"**_{_{![P2 Badge](https://img.shields.io/badge/P2-yellow?style=flat)}} Drop active-session background browser mounts**\\n\\nWhen background browsing is enabled, a tab that was mounted off-screen stays in `mounts` even after the user focuses that same agent. `MainPanelContent` also mounts the active agent's browser tabs, so this loop renders a second hidden `` for the same tab; its navigation events still call `handleBackgroundTabUpdate`, which can race with and overwrite the visible tab's URL/title/state while also duplicating page side effects. Filter out the active session or clear those mounts on focus.\\n\\nUseful? React with 👍 / 👎.\",\"path\":\"src/renderer/components/coworking/CoworkingBackgroundBrowsers.tsx\",\"side\":\"RIGHT\",\"line\":null}{\"type\":\"pull_request_comment\",\"body\":\"**_{_{![P2 Badge](https://img.shields.io/badge/P2-yellow?style=flat)}} Replace disallowed dash characters**\\n\\nAGENTS.md for `/workspace/Maestro` forbids U+2014 and U+2013 anywhere, including UI copy. This new Settings description renders a U+2014 dash, and the commit adds many similar characters in comments and strings, so it violates the repo-wide rule and should be replaced with a comma, colon, or plain hyphen.\\n\\nUseful? React with 👍 / 👎.\",\"path\":\"src/renderer/components/Settings/CoworkingSetup.tsx\",\"side\":\"RIGHT\",\"line\":null}{\"type\":\"pull_request_comment\",\"body\":\"\\n### 💡 Codex Review\\n\\nHere are some automated review suggestions for this pull request.\\n\\n**Reviewed commit:** `98eae338f2`\\n \\n\\n

ℹ️ About Codex in GitHub

\\n
\\n\\n[Your team has set up Codex to review pull requests in this repo](https://chatgpt.com/codex/cloud/settings/general). Reviews are triggered when you\\n- Open a pull request for review\\n- Mark a draft as ready\\n- Comment \\\"@codex review\\\".\\n\\nIf Codex has suggestions, it will comment; otherwise it will react with 👍.\\n\\n\\n\\n\\nCodex can also answer questions or update the PR. Try commenting \\\"@codex address that feedback\\\".\\n \\n

\",\"path\":null,\"side\":null,\"line\":null}{\"type\":\"pull_request_comment\",\"body\":\"**_{_{![P2 Badge](https://img.shields.io/badge/P2-yellow?style=flat)}} Restore the previous non-browser surface after mounting**\\n\\nIf the focused agent is currently on an AI, file, or terminal tab and an agent calls `read_browser` or `browser_wait_for` for an unmounted browser tab, this fallback activates the browser to mount it but only remembers a previous browser tab. When `prevActiveBrowserTabId` is null, the `finally` block leaves the user switched to the agent-driven browser tab after the tool call; capture and restore the full prior focus state, not just a browser id.\\n\\nUseful? React with 👍 / 👎.\",\"path\":\"src/renderer/hooks/coworking/useCoworkingBrowserResponder.ts\",\"side\":\"RIGHT\",\"line\":null}{\"type\":\"pull_request_comment\",\"body\":\"**_{_{![P2 Badge](https://img.shields.io/badge/P2-yellow?style=flat)}} Mark navigations as loading before later browser reads**\\n\\nAfter `browser_navigate`, `isDomReadyRef` can still be true from the previous page because this path only assigns `webview.src`. A follow-up `read_browser` or `browser_wait_for` issued immediately can then run JavaScript against the old page instead of waiting for the new page's `dom-ready`, returning stale content or a false selector match. Clear the ready flag or wait for navigation completion when changing `src`.\\n\\nUseful? React with 👍 / 👎.\",\"path\":\"src/renderer/components/MainPanel/BrowserTabView.tsx\",\"side\":\"RIGHT\",\"line\":null}{\"type\":\"pull_request_comment\",\"body\":\"**_{_{![P2 Badge](https://img.shields.io/badge/P2-yellow?style=flat)}} Use a reliable runtime for the bundled MCP server**\\n\\nOn machines where `which node` or `where node` fails, or where a GUI launch cannot see a version-manager PATH, installs write `command: \\\"node\\\"` into the agent's MCP config. Codex and Factory Droid can be installed without a standalone Node binary, so the feature will report as installed but the MCP subprocess cannot start. Use the bundled Electron runtime in Node mode or fail the install with an actionable error instead of writing a command that may not exist.\\n\\nUseful? React with 👍 / 👎.\",\"path\":\"src/main/coworking/coworking-server-paths.ts\",\"side\":\"RIGHT\",\"line\":76}{\"type\":\"pull_request_comment\",\"body\":\"**_{_{![P2 Badge](https://img.shields.io/badge/P2-yellow?style=flat)}} Inject the coworking session id before SSH wrapping**\\n\\nFor SSH-backed agents, `process:spawn` builds the remote stdin script with `mergedSshEnvVars` and then clears `customEnvVarsToPass` before calling `ProcessManager.spawn`; this injection therefore lands only in the local `ssh` process environment, not in the remote agent process. The remote MCP subprocess will not have `MAESTRO_COWORKING_SESSION_ID`, and its `ppid` is a remote PID the local resolver cannot match, so coworking tools fail for SSH agents. Add this env var before the SSH command is built.\\n\\nUseful? React with 👍 / 👎.\",\"path\":\"src/main/process-manager/ProcessManager.ts\",\"side\":\"RIGHT\",\"line\":90}{\"type\":\"pull_request_comment\",\"body\":\"**_{_{![P1 Badge](https://img.shields.io/badge/P1-orange?style=flat)}} Verify explicit session ids against the caller PID**\\n\\nWhen a client supplies `sessionId`, the bridge binds to it immediately and skips the PID resolver even though the bundled server also sends `ppid`. A hand-written client, or an agent that can run a local command and read another Maestro session id from app data, can connect to the userData socket and claim that other id, bypassing the per-agent scoping this bridge is meant to enforce for `read_terminal` and browser reads. When `ppid` is present, verify it resolves to the same session before trusting the explicit id.\\n\\nUseful? React with 👍 / 👎.\",\"path\":\"src/main/coworking/coworking-bridge.ts\",\"side\":\"RIGHT\",\"line\":324}{\"type\":\"pull_request_comment\",\"body\":\"_🔒 Security & Privacy_ | _🟠 Major_ | _⚡ Quick win_\\n\\n**Derive the Windows pipe slug from the full `userData` path** \\n`path.basename(app.getPath('userData'))` collapses to the same app-name folder for every OS user, so the named pipe name is not actually per-user on shared Windows machines. Hash the full `userData` path (or otherwise include the user-specific portion) to avoid cross-user bridge collisions.\\n\\n

\\n

🤖 Prompt for AI Agents

\\n\\n```\\nVerify each finding against current code. Fix only still-valid issues, skip the\\nrest with a brief reason, keep changes minimal, and validate.\\n\\nIn `@src/main/coworking/coworking-bridge.ts` around lines 95 - 102, The Windows\\npipe name in getBridgeSocketPath is not truly per-user because\\npath.basename(app.getPath('userData')) only uses the shared app folder name.\\nUpdate the Windows branch in getBridgeSocketPath to derive the slug from the\\nfull app.getPath('userData') value, such as hashing it or otherwise encoding the\\nuser-specific path, while keeping the existing pipe format in\\ncoworking-bridge.ts.\\n```\\n\\n

\\n\\n\\n\\n\\n\\n\\n\\n\",\"path\":\"src/main/coworking/coworking-bridge.ts\",\"side\":\"RIGHT\",\"line\":106}{\"type\":\"pull_request_comment\",\"body\":\"_🗄️ Data Integrity & Integration_ | _🟡 Minor_ | _⚡ Quick win_\\n\\n**Audit records \\\"ok\\\" before the call actually runs.**\\n\\nEvery other branch (`readTerminal`, `listBrowsers`, `getBrowserUrl`, `readBrowser`, `browserInteract`) wraps the tool call in `auditedBrowserCall`, which records success/error based on the actual outcome. `listTerminals` instead records `status: 'ok'` unconditionally *before* calling `listTerminals(sessionId)` (line 345) — if the call throws, the audit trail will still claim success.\\n\\n\\n\\n\\n

\\n

🩹 Suggested fix — use the same audited pattern as the other methods

\\n\\n```diff\\n if (method === 'listTerminals') {\\n-\\t\\t\\trecordBrowserAudit({\\n-\\t\\t\\t\\tts: Date.now(),\\n-\\t\\t\\t\\tsessionId,\\n-\\t\\t\\t\\tagentType: coworkingRegistry.getAgentType(sessionId),\\n-\\t\\t\\t\\ttool: 'list_terminals',\\n-\\t\\t\\t\\tstatus: 'ok',\\n-\\t\\t\\t});\\n-\\t\\t\\treturn { id: req.id, result: listTerminals(sessionId) };\\n+\\t\\t\\tconst result = await auditedBrowserCall(\\n+\\t\\t\\t\\t{ sessionId, agentType: coworkingRegistry.getAgentType(sessionId), tool: 'list_terminals' },\\n+\\t\\t\\t\\t() => listTerminals(sessionId)\\n+\\t\\t\\t);\\n+\\t\\t\\treturn { id: req.id, result };\\n \\t\\t}\\n```\\n

\\n\\n\\n\\n

\\n

📝 Committable suggestion

\\n\\n\\n\\n

\\n

🤖 Prompt for AI Agents

\\n\\n```\\nVerify each finding against current code. Fix only still-valid issues, skip the\\nrest with a brief reason, keep changes minimal, and validate.\\n\\nIn `@src/main/coworking/coworking-bridge.ts` around lines 337 - 346, The\\nlistTerminals branch in coworking-bridge.ts records an ok audit before the tool\\nresult is known, unlike readTerminal, listBrowsers, getBrowserUrl, readBrowser,\\nand browserInteract which use auditedBrowserCall. Update the listTerminals\\nhandling to wrap listTerminals(sessionId) in auditedBrowserCall so success/error\\nis recorded based on the actual outcome, and remove the unconditional pre-call\\nrecordBrowserAudit call.\\n```\\n\\n

\\n\\n\\n\\n\\n\\n\\n\\n\",\"path\":\"src/main/coworking/coworking-bridge.ts\",\"side\":\"RIGHT\",\"line\":414}{\"type\":\"pull_request_comment\",\"body\":\"_🚀 Performance & Scalability_ | _🟡 Minor_ | _⚡ Quick win_\\n\\n**`maxChars` has no upper bound.**\\n\\n`readBrowser`'s `maxChars` is only checked for `> 0`; nothing caps the maximum. An agent (or a misbehaving prompt) could request an arbitrarily large value, forcing a very large in-memory string / IPC payload for a single `read_browser` call.\\n\\n\\n\\n

\\n

🛡️ Suggested fix — clamp to a sane ceiling

\\n\\n```diff\\n \\t\\t\\tif (\\n \\t\\t\\t\\tmaxChars !== undefined &&\\n-\\t\\t\\t\\t(typeof maxChars !== 'number' || !Number.isInteger(maxChars) || maxChars <= 0)\\n+\\t\\t\\t\\t(typeof maxChars !== 'number' ||\\n+\\t\\t\\t\\t\\t!Number.isInteger(maxChars) ||\\n+\\t\\t\\t\\t\\tmaxChars <= 0 ||\\n+\\t\\t\\t\\t\\tmaxChars > 2_000_000)\\n \\t\\t\\t) {\\n \\t\\t\\t\\treturn {\\n \\t\\t\\t\\t\\tid: req.id,\\n-\\t\\t\\t\\t\\terror: { code: -32602, message: '`maxChars` must be a positive integer' },\\n+\\t\\t\\t\\t\\terror: { code: -32602, message: '`maxChars` must be a positive integer <= 2,000,000' },\\n \\t\\t\\t\\t};\\n \\t\\t\\t}\\n```\\n

\\n\\n\\n\\n

\\n

📝 Committable suggestion

\\n\\n\\n\\n

\\n

🤖 Prompt for AI Agents

\\n\\n```\\nVerify each finding against current code. Fix only still-valid issues, skip the\\nrest with a brief reason, keep changes minimal, and validate.\\n\\nIn `@src/main/coworking/coworking-bridge.ts` around lines 404 - 413, `readBrowser`\\ncurrently validates `maxChars` in `coworking-bridge.ts` only for being a\\npositive integer, so it can still accept unbounded values and create oversized\\nin-memory/IPC payloads. Update the `maxChars` validation in the request handling\\nblock for `params.maxChars` to enforce a reasonable upper limit (or clamp it)\\nalongside the existing positive-integer check, and return the same `-32602`\\nvalidation error when the value exceeds that ceiling.\\n```\\n\\n

\\n\\n\\n\\n\\n\\n\\n\\n\",\"path\":\"src/main/coworking/coworking-bridge.ts\",\"side\":\"RIGHT\",\"line\":487}{\"type\":\"pull_request_comment\",\"body\":\"_🗄️ Data Integrity & Integration_ | _🟠 Major_ | _⚡ Quick win_\\n\\n**Install/uninstall separator math isn't reversible for two realistic inputs.**\\n\\nTracing the newline bookkeeping by hand:\\n\\n- If the existing `config.toml` has **no trailing newline**, `install` uses `sep = '\\\\n\\\\n'` (2 chars added). On `uninstall`, the trim logic only ever removes **one** of the two newlines it detects (Line 111: `startTrim -= 1`), leaving a spurious trailing newline that was never in the original file.\\n- If the existing file already ends with a blank line (`\\\\n\\\\n`), `install` uses `sep = ''` (adds nothing). On `uninstall`, the same lookbehind still sees two consecutive `\\\\n`s immediately before the block and removes one — deleting a blank line that belonged to the user's own content, not something this installer added.\\n\\nBoth cases stem from trying to infer, purely from the merged file, how many characters were inserted at install time — that information isn't recoverable without extra bookkeeping. The current test suite (`codex.test.ts`) only covers the case where `existing` already ends in exactly one `\\\\n`, so these regressions aren't caught.\\n\\n\\n\\n\\n

\\n

♻️ Suggested direction — normalize instead of guessing

\\n\\nRather than trying to precisely restore original whitespace, normalize trailing newlines deterministically on both sides so the transform is provably reversible (byte-exact preservation of blank-line *count* is sacrificed, but TOML validity and idempotency are guaranteed):\\n\\n```diff\\n \\tasync install(spec: CoworkingMcpServerSpec) {\\n \\t\\tconst block = buildBlock(spec);\\n \\t\\tconst existing = await readConfig();\\n \\t\\tif (existing == null) {\\n \\t\\t\\tawait writeConfig(block + '\\\\n');\\n \\t\\t\\treturn;\\n \\t\\t}\\n \\t\\tconst found = findBlock(existing);\\n \\t\\tif (found) {\\n \\t\\t\\tconst replaced = existing.slice(0, found.start) + block + existing.slice(found.end);\\n \\t\\t\\tawait writeConfig(replaced);\\n \\t\\t\\treturn;\\n \\t\\t}\\n-\\t\\t// Append, ensuring exactly one blank line between user content and our block.\\n-\\t\\tconst sep = existing.endsWith('\\\\n\\\\n') ? '' : existing.endsWith('\\\\n') ? '\\\\n' : '\\\\n\\\\n';\\n-\\t\\tawait writeConfig(existing + sep + block + '\\\\n');\\n+\\t\\t// Normalize trailing whitespace so uninstall never has to guess how much\\n+\\t\\t// separator we added.\\n+\\t\\tconst trimmed = existing.replace(/\\\\n+$/, '');\\n+\\t\\tconst sep = trimmed.length > 0 ? '\\\\n\\\\n' : '';\\n+\\t\\tawait writeConfig(trimmed + sep + block + '\\\\n');\\n \\t},\\n \\n \\tasync uninstall() {\\n \\t\\tconst existing = await readConfig();\\n \\t\\tif (!existing) return;\\n \\t\\tconst found = findBlock(existing);\\n \\t\\tif (!found) return;\\n-\\t\\t// Trim a single trailing newline (and at most one blank-line separator we may have added).\\n-\\t\\tlet endTrim = found.end;\\n-\\t\\tif (existing[endTrim] === '\\\\n') endTrim += 1;\\n-\\t\\t// If a blank line precedes our block, drop one (mirror the install spacing).\\n-\\t\\tlet startTrim = found.start;\\n-\\t\\tif (startTrim >= 2 && existing[startTrim - 1] === '\\\\n' && existing[startTrim - 2] === '\\\\n') {\\n-\\t\\t\\tstartTrim -= 1;\\n-\\t\\t}\\n-\\t\\tconst updated = existing.slice(0, startTrim) + existing.slice(endTrim);\\n+\\t\\tconst before = existing.slice(0, found.start).replace(/\\\\n+$/, '');\\n+\\t\\tconst after = existing.slice(found.end).replace(/^\\\\n+/, '');\\n+\\t\\tconst updated = before + (before && after ? '\\\\n\\\\n' : after ? '' : '') + after;\\n \\t\\tawait writeConfig(updated);\\n \\t},\\n```\\n\\nPlease add regression tests for a config lacking a trailing newline and one with a pre-existing blank line before the block, to lock in the desired behavior.\\n

\\n\\n\\n\\n

\\n

📝 Committable suggestion

\\n\\n\\n\\n

\\n

🤖 Prompt for AI Agents

\\n\\n```\\nVerify each finding against current code. Fix only still-valid issues, skip the\\nrest with a brief reason, keep changes minimal, and validate.\\n\\nIn `@src/main/coworking/installers/codex.ts` around lines 96 - 113, The\\ninstall/uninstall newline handling in codex.ts is not reversible for inputs with\\nno trailing newline or with an existing blank line before the injected block.\\nUpdate the logic in install() and uninstall() to stop guessing original spacing\\nfrom the merged file; instead normalize trailing newlines deterministically so\\nremove/add operations are idempotent. Add regression tests in codex.test.ts for\\na config without a trailing newline and one that already ends with a blank line\\nbefore the block.\\n```\\n\\n

\\n\\n\\n\\n\\n\\n\\n\\n\",\"path\":\"src/main/coworking/installers/codex.ts\",\"side\":\"RIGHT\",\"line\":115}{\"type\":\"pull_request_comment\",\"body\":\"_🎯 Functional Correctness_ | _🟠 Major_ | _⚡ Quick win_\\n\\n**`clearArmed` confirmation state isn't reset when the active tab changes.**\\n\\nThis component instance is reused across tab switches (see the existing `addressValue` reset in the `[tab.id, tab.url]` effect and `isDomReadyRef` reset in the `[tab.id]`-keyed listener effect further down), but `clearArmed`/`clearDisarmTimerRef` have no equivalent per-tab reset. If a user arms \\\"Clear session data\\\" on one tab and switches to another tab before confirming or before the 4s auto-disarm fires, the next tab's clear button renders pre-armed, and a single click on it will immediately wipe that tab's session data — silently skipping the two-step confirmation this feature exists to enforce.\\n\\n

\\n

🛠️ Proposed fix: reset arm state on tab change

\\n\\n```diff\\n \\t\\tconst [clearArmed, setClearArmed] = useState(false);\\n \\t\\tconst clearDisarmTimerRef = useRef(null);\\n \\n \\t\\tuseEffect(() => {\\n \\t\\t\\treturn () => {\\n \\t\\t\\t\\tif (clearDisarmTimerRef.current !== null) {\\n \\t\\t\\t\\t\\twindow.clearTimeout(clearDisarmTimerRef.current);\\n \\t\\t\\t\\t}\\n \\t\\t\\t};\\n \\t\\t}, []);\\n+\\n+\\t\\t// Reset the destructive-action confirmation whenever the active tab\\n+\\t\\t// changes, so an armed state on one tab can never be consumed by a\\n+\\t\\t// click on a different tab.\\n+\\t\\tuseEffect(() => {\\n+\\t\\t\\tsetClearArmed(false);\\n+\\t\\t\\tif (clearDisarmTimerRef.current !== null) {\\n+\\t\\t\\t\\twindow.clearTimeout(clearDisarmTimerRef.current);\\n+\\t\\t\\t\\tclearDisarmTimerRef.current = null;\\n+\\t\\t\\t}\\n+\\t\\t}, [tab.id]);\\n```\\n

\\n\\n\\n\\n\\n\\nAlso applies to: 740-780\\n\\n

\\n

🤖 Prompt for AI Agents

\\n\\n```\\nVerify each finding against current code. Fix only still-valid issues, skip the\\nrest with a brief reason, keep changes minimal, and validate.\\n\\nIn `@src/renderer/components/MainPanel/BrowserTabView.tsx` around lines 188 - 199,\\nThe two-step clear-browsing-data guard in BrowserTabView needs to be reset when\\nthe active tab changes, because clearArmed and clearDisarmTimerRef currently\\npersist across reused component instances. Update the state reset logic\\nalongside the existing tab-scoped effects that handle addressValue and\\nisDomReadyRef so that switching tabs clears any armed state and cancels the\\npending timeout. Use the BrowserTabView tab-id/url effects and the\\nclearDisarmTimerRef cleanup to locate the right place for the fix.\\n```\\n\\n

\\n\\n\\n\\n\\n\\n\\n\\n\",\"path\":\"src/renderer/components/MainPanel/BrowserTabView.tsx\",\"side\":\"RIGHT\",\"line\":199}{\"type\":\"pull_request_comment\",\"body\":\"_🩺 Stability & Availability_ | _🟠 Major_ | _⚡ Quick win_\\n\\n**Approval/background-browser hosts are unreachable from several early-return branches.**\\n\\n`CoworkingApprovalHost` and `CoworkingBackgroundBrowsers` are only mounted in the final \\\"normal session\\\" return. The `logViewerOpen`, `agentSessionsOpen`, `memoryViewerOpen`, and `!activeSession` branches (Lines 735-803) all `return` before reaching them. Any coworking approval requested while one of those views is open has no UI to resolve it, so the agent's awaiting tool call can hang indefinitely until the user happens to navigate back. Background-browser keep-alive is likewise paused while those views are shown.\\n\\nHoist both components above the conditional returns (or render them from a shared wrapper) so they're always mounted regardless of which panel is showing.\\n\\n

\\n

🐛 Suggested fix sketch

\\n\\n```diff\\n+\\t\\treturn (\\n+\\t\\t\\t<>\\n+\\t\\t\\t\\t{(() => {\\n+\\t\\t\\t\\t\\tif (logViewerOpen) {\\n+\\t\\t\\t\\t\\t\\treturn ( /* existing log viewer JSX */ );\\n+\\t\\t\\t\\t\\t}\\n+\\t\\t\\t\\t\\tif (agentSessionsOpen && hasCapability('supportsSessionStorage')) {\\n+\\t\\t\\t\\t\\t\\treturn ( /* existing agent sessions JSX */ );\\n+\\t\\t\\t\\t\\t}\\n+\\t\\t\\t\\t\\tif (memoryViewerOpen && hasCapability('supportsProjectMemory')) {\\n+\\t\\t\\t\\t\\t\\treturn ( /* existing memory viewer JSX */ );\\n+\\t\\t\\t\\t\\t}\\n+\\t\\t\\t\\t\\tif (!activeSession) {\\n+\\t\\t\\t\\t\\t\\treturn ( /* existing empty-state JSX */ );\\n+\\t\\t\\t\\t\\t}\\n+\\t\\t\\t\\t\\treturn ( /* existing normal session JSX (without the two coworking hosts) */ );\\n+\\t\\t\\t\\t})()}\\n+\\t\\t\\t\\t\\n+\\t\\t\\t\\t\\n+\\t\\t\\t\\n+\\t\\t);\\n```\\n

\\n\\n

\\n

🤖 Prompt for AI Agents

\\n\\n```\\nVerify each finding against current code. Fix only still-valid issues, skip the\\nrest with a brief reason, keep changes minimal, and validate.\\n\\nIn `@src/renderer/components/MainPanel/MainPanel.tsx` around lines 1097 - 1098,\\nThe coworking host components are only rendered in the final normal-session\\nbranch, so early returns like logViewerOpen, agentSessionsOpen,\\nmemoryViewerOpen, and !activeSession bypass them. Update MainPanel so\\nCoworkingApprovalHost and CoworkingBackgroundBrowsers are mounted from a shared\\ntop-level wrapper or hoisted above the conditional return logic, ensuring they\\nstay active regardless of which view is displayed.\\n```\\n\\n

\\n\\n\\n\\n\\n\\n\\n\\n\",\"path\":\"src/renderer/components/MainPanel/MainPanel.tsx\",\"side\":\"RIGHT\",\"line\":1106}{\"type\":\"pull_request_comment\",\"body\":\"_🗄️ Data Integrity & Integration_ | _🟠 Major_ | _⚡ Quick win_\\n\\n**`createTabForSession` reports success even when the session no longer exists.**\\n\\nUnlike `closeTabForSession` below it (which tracks a `closed` flag), this function never checks whether `setSessions`'s updater actually matched `sessionId`. Since the session-existence check in `resolveAndRun` happens *before* the (potentially long) approval await, a session/tab closed while a human is still responding to the confirm dialog will cause the new tab to be silently dropped while the agent is told it succeeded.\\n\\n\\n\\n\\n

\\n

🐛 Proposed fix: track whether the session was found

\\n\\n```diff\\n \\tconst { setSessions } = useSessionStore.getState();\\n \\tconst created = createBrowserTab(sessionId, targetUrl, {\\n \\t\\tephemeral: op.ephemeral === true,\\n \\t});\\n+\\tlet sessionFound = false;\\n \\tsetSessions((prev) =>\\n \\t\\tprev.map((s) => {\\n \\t\\t\\tif (s.id !== sessionId) return s;\\n+\\t\\t\\tsessionFound = true;\\n \\t\\t\\treturn {\\n \\t\\t\\t\\t...s,\\n \\t\\t\\t\\tbrowserTabs: [...(s.browserTabs || []), created],\\n \\t\\t\\t\\tactiveBrowserTabId: created.id,\\n \\t\\t\\t\\tunifiedTabOrder: insertAfterActiveInUnifiedTabOrder(s, {\\n \\t\\t\\t\\t\\ttype: 'browser',\\n \\t\\t\\t\\t\\tid: created.id,\\n \\t\\t\\t\\t}),\\n \\t\\t\\t};\\n \\t\\t})\\n \\t);\\n+\\tif (!sessionFound) {\\n+\\t\\treturn { ok: false, content: 'Session no longer exists; the tab could not be created.' };\\n+\\t}\\n \\treturn {\\n \\t\\tok: true,\\n \\t\\turl: created.url,\\n \\t\\tcontent: `Opened a new${op.ephemeral ? ' incognito' : ''} browser tab at ${created.url}. Call list_browsers to get its id.`,\\n \\t};\\n```\\n

\\n\\n\\n\\n

\\n

📝 Committable suggestion

\\n\\n\\n\\n

\\n

🤖 Prompt for AI Agents

\\n\\n```\\nVerify each finding against current code. Fix only still-valid issues, skip the\\nrest with a brief reason, keep changes minimal, and validate.\\n\\nIn `@src/renderer/hooks/coworking/useCoworkingBrowserResponder.ts` around lines\\n123 - 161, createTabForSession can return a success result even when the target\\nsession was removed before the state update runs. Update createTabForSession to\\ntrack whether setSessions actually matched sessionId, similar to the closed flag\\npattern in closeTabForSession, and if no session was updated return ok: false\\nwith an appropriate message instead of reporting success. Use the existing\\ncreateTabForSession, setSessions, and useSessionStore flow to locate the fix.\\n```\\n\\n

\\n\\n\\n\\n\\n\\n\\n\\n\",\"path\":\"src/renderer/hooks/coworking/useCoworkingBrowserResponder.ts\",\"side\":\"RIGHT\",\"line\":179}{\"type\":\"pull_request_comment\",\"body\":\"_🗄️ Data Integrity & Integration_ | _🟠 Major_ | _⚡ Quick win_\\n\\n**Race between overlapping async sync batches can clobber newer state.**\\n\\n`lastPayloadRef`/`lastSessionIdsRef` are optimistically written synchronously before each async IIFE starts, but the failure-path rollback (Lines 203-204) unconditionally overwrites them with `previousSessionIds` captured at that IIFE's start — with no generation guard. If an older, slower batch fails *after* a newer batch has already started (or succeeded), its rollback stomps on the newer batch's state: `lastSessionIdsRef` reverts to a stale snapshot, `removed` is computed incorrectly on the next run, and the main-process registry can drift out of sync with what was actually pushed (or previously-cleared sessions can reappear as \\\"removed\\\" against the wrong baseline). A related variant: the disable-branch's fire-and-forget `removeSession` loop (Lines 118-140) has no ordering guarantee against a subsequent quick re-enable's resync, so a slow `removeSession` can evict a session's freshly-synced state.\\n\\nGuard the rollback with a monotonic run id so only the *current* run can mutate the shared refs on failure.\\n\\n

\\n

🔒 Proposed fix — generation-guarded rollback

\\n\\n```diff\\n+\\tconst runIdRef = useRef(0);\\n \\tuseEffect(() => {\\n \\t\\tconst bridge = window.maestro?.coworking;\\n \\t\\tif (!bridge) {\\n \\t\\t\\tlastPayloadRef.current = '';\\n \\t\\t\\tlastSessionIdsRef.current = new Set();\\n \\t\\t\\treturn;\\n \\t\\t}\\n+\\t\\tconst runId = ++runIdRef.current;\\n \\n \\t\\tif (!enabled) {\\n \\t\\t\\t...\\n \\t\\t}\\n \\t\\t...\\n \\t\\t(async () => {\\n \\t\\t\\ttry {\\n \\t\\t\\t\\tfor (const sid of removed) {\\n \\t\\t\\t\\t\\tawait bridge.removeSession(sid);\\n \\t\\t\\t\\t}\\n \\t\\t\\t\\tfor (const { sessionId, records } of perSession) {\\n \\t\\t\\t\\t\\tawait bridge.syncSessionTerminals(sessionId, records);\\n \\t\\t\\t\\t}\\n \\t\\t\\t\\tfor (const { sessionId, inputs, interactionEnabled, agentType, confirmPolicy } of perSessionBrowsers) {\\n \\t\\t\\t\\t\\tawait bridge.syncSessionBrowsers(sessionId, inputs, interactionEnabled, agentType, confirmPolicy);\\n \\t\\t\\t\\t}\\n \\t\\t\\t} catch (err) {\\n-\\t\\t\\t\\tlastPayloadRef.current = '';\\n-\\t\\t\\t\\tlastSessionIdsRef.current = previousSessionIds;\\n-\\t\\t\\t\\treportIfUnexpected(err, 'sync');\\n+\\t\\t\\t\\tif (runIdRef.current === runId) {\\n+\\t\\t\\t\\t\\t// Only roll back if no newer run has superseded this one.\\n+\\t\\t\\t\\t\\tlastPayloadRef.current = '';\\n+\\t\\t\\t\\t\\tlastSessionIdsRef.current = previousSessionIds;\\n+\\t\\t\\t\\t}\\n+\\t\\t\\t\\treportIfUnexpected(err, 'sync');\\n \\t\\t\\t}\\n \\t\\t})();\\n \\t}, [enabled, sessions, browserInteractionAgents, browserConfirmPolicies]);\\n```\\n

\\n\\n\\n\\n

\\n

📝 Committable suggestion

\\n\\n> ‼️ **IMPORTANT**\\n> Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.\\n\\n```suggestion\\n\\tconst runIdRef = useRef(0);\\n\\tuseEffect(() => {\\n\\t\\t// Bail out cleanly when the coworking bridge isn't exposed (e.g. in test\\n\\t\\t// harnesses that mock `window.maestro` without the namespace, or in older\\n\\t\\t// preload bundles before this PR shipped). The hook is best-effort —\\n\\t\\t// without the bridge there's nothing to sync to.\\n\\t\\tconst bridge = window.maestro?.coworking;\\n\\t\\tif (!bridge) {\\n\\t\\t\\tlastPayloadRef.current = '';\\n\\t\\t\\tlastSessionIdsRef.current = new Set();\\n\\t\\t\\treturn;\\n\\t\\t}\\n\\t\\tconst runId = ++runIdRef.current;\\n\\n\\t\\tif (!enabled) {\\n\\t\\t\\t// Drop everything we previously pushed. We don't need to emit one big\\n\\t\\t\\t// \\\"clear\\\" — calling removeSession per known session is what keeps the\\n\\t\\t\\t// registry tidy without us having to add a new RPC.\\n\\t\\t\\tconst prev = lastSessionIdsRef.current;\\n\\t\\t\\tif (prev.size > 0) {\\n\\t\\t\\t\\t(async () => {\\n\\t\\t\\t\\t\\tfor (const sid of prev) {\\n\\t\\t\\t\\t\\t\\ttry {\\n\\t\\t\\t\\t\\t\\t\\tawait bridge.removeSession(sid);\\n\\t\\t\\t\\t\\t\\t} catch (err) {\\n\\t\\t\\t\\t\\t\\t\\t// Best-effort teardown while disabling the feature: capture\\n\\t\\t\\t\\t\\t\\t\\t// unexpected failures but keep clearing the remaining sessions\\n\\t\\t\\t\\t\\t\\t\\t// (don't abort the loop on one bad removeSession).\\n\\t\\t\\t\\t\\t\\t\\tif (!isExpectedTeardownIpcError(err)) {\\n\\t\\t\\t\\t\\t\\t\\t\\tvoid captureException(err instanceof Error ? err : new Error(String(err)), {\\n\\t\\t\\t\\t\\t\\t\\t\\t\\textra: { scope: 'useCoworkingRegistrySync:disable-clear' },\\n\\t\\t\\t\\t\\t\\t\\t\\t});\\n\\t\\t\\t\\t\\t\\t\\t}\\n\\t\\t\\t\\t\\t\\t}\\n\\t\\t\\t\\t\\t}\\n\\t\\t\\t\\t})();\\n\\t\\t\\t}\\n\\t\\t\\tlastPayloadRef.current = '';\\n\\t\\t\\tlastSessionIdsRef.current = new Set();\\n\\t\\t\\treturn;\\n\\t\\t}\\n\\n\\t\\t// Build the cross-session payload.\\n\\t\\tconst perSession = sessions.map((s) => ({ sessionId: s.id, records: buildRecords(s) }));\\n\\t\\tconst perSessionBrowsers = sessions.map((s) => ({\\n\\t\\t\\tsessionId: s.id,\\n\\t\\t\\tinputs: buildBrowserInputs(s),\\n\\t\\t\\tinteractionEnabled: browserInteractionAgents.includes(s.toolType),\\n\\t\\t\\tagentType: s.toolType,\\n\\t\\t\\tconfirmPolicy: browserConfirmPolicies[s.toolType] ?? DEFAULT_BROWSER_CONFIRM_POLICY,\\n\\t\\t}));\\n\\t\\tconst currentIds = new Set(perSession.map((p) => p.sessionId));\\n\\n\\t\\t// Detect sessions that disappeared since the previous run so we can clear\\n\\t\\t// them from the registry instead of leaving stale records.\\n\\t\\tconst removed: string[] = [];\\n\\t\\tfor (const id of lastSessionIdsRef.current) {\\n\\t\\t\\tif (!currentIds.has(id)) removed.push(id);\\n\\t\\t}\\n\\n\\t\\t// Skip the IPC if nothing has changed since last sync (cheap stringify of small payload).\\n\\t\\tconst payload = JSON.stringify({ perSession, perSessionBrowsers, removed });\\n\\t\\tif (payload === lastPayloadRef.current) return;\\n\\t\\tlastPayloadRef.current = payload;\\n\\t\\t// Snapshot the session-id set so we can restore it if the async sync batch\\n\\t\\t// below fails. Otherwise a failed removeSession drops the removed session\\n\\t\\t// from our cache, so the next run won't re-detect it as `removed` and the\\n\\t\\t// main registry keeps stale tabs forever.\\n\\t\\tconst previousSessionIds = lastSessionIdsRef.current;\\n\\t\\tlastSessionIdsRef.current = currentIds;\\n\\n\\t\\t(async () => {\\n\\t\\t\\ttry {\\n\\t\\t\\t\\tfor (const sid of removed) {\\n\\t\\t\\t\\t\\tawait bridge.removeSession(sid);\\n\\t\\t\\t\\t}\\n\\t\\t\\t\\tfor (const { sessionId, records } of perSession) {\\n\\t\\t\\t\\t\\tawait bridge.syncSessionTerminals(sessionId, records);\\n\\t\\t\\t\\t}\\n\\t\\t\\t\\tfor (const {\\n\\t\\t\\t\\t\\tsessionId,\\n\\t\\t\\t\\t\\tinputs,\\n\\t\\t\\t\\t\\tinteractionEnabled,\\n\\t\\t\\t\\t\\tagentType,\\n\\t\\t\\t\\t\\tconfirmPolicy,\\n\\t\\t\\t\\t} of perSessionBrowsers) {\\n\\t\\t\\t\\t\\tawait bridge.syncSessionBrowsers(\\n\\t\\t\\t\\t\\t\\tsessionId,\\n\\t\\t\\t\\t\\t\\tinputs,\\n\\t\\t\\t\\t\\t\\tinteractionEnabled,\\n\\t\\t\\t\\t\\t\\tagentType,\\n\\t\\t\\t\\t\\t\\tconfirmPolicy\\n\\t\\t\\t\\t\\t);\\n\\t\\t\\t\\t}\\n\\t\\t\\t} catch (err) {\\n\\t\\t\\t\\t// Roll back the optimistic payload-cache write FIRST so the next effect\\n\\t\\t\\t\\t// run retries instead of treating the same payload as already-synced and\\n\\t\\t\\t\\t// leaving the main-process registry stale. Then surface the failure\\n\\t\\t\\t\\t// (teardown IPC errors stay quiet; anything else re-throws).\\n\\t\\t\\t\\tif (runIdRef.current === runId) {\\n\\t\\t\\t\\t\\tlastPayloadRef.current = '';\\n\\t\\t\\t\\t\\tlastSessionIdsRef.current = previousSessionIds;\\n\\t\\t\\t\\t}\\n\\t\\t\\t\\treportIfUnexpected(err, 'sync');\\n\\t\\t\\t}\\n\\t\\t})();\\n\\t}, [enabled, sessions, browserInteractionAgents, browserConfirmPolicies]);\\n```\\n\\n

\\n\\n\\n\\n

\\n

🤖 Prompt for AI Agents

\\n\\n```\\nVerify each finding against current code. Fix only still-valid issues, skip the\\nrest with a brief reason, keep changes minimal, and validate.\\n\\nIn `@src/renderer/hooks/coworking/useCoworkingRegistrySync.ts` around lines 106 -\\n208, The async sync in useCoworkingRegistrySync can roll back shared refs from\\nan older batch after a newer batch has already started or completed. Add a\\nmonotonic run/generation identifier inside the useEffect sync path and capture\\nit before starting the async IIFE, then only apply the failure rollback to\\nlastPayloadRef and lastSessionIdsRef when the failing run is still the latest\\none. Make sure the same guard is applied to the disable/clear flow around\\nbridge.removeSession so stale teardown work cannot overwrite newer sync state.\\n```\\n\\n

\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n✅ Addressed in commits 74e322f to af7dc7b\",\"path\":\"src/renderer/hooks/coworking/useCoworkingRegistrySync.ts\",\"side\":\"RIGHT\",\"line\":230}{\"type\":\"pull_request_comment\",\"body\":\"**Actionable comments posted: 8**\\n\\n

\\n

🧹 Nitpick comments (14)

\\n\\n
\\n
src/shared/coworkingBrowser.ts (1)
\\n\\n`77-95`: _📐 Maintainability & Code Quality_ | _🔵 Trivial_ | _💤 Low value_\\n\\n**Stale doc comment for 'dangerous' policy.**\\n\\nThe docstring says \\\"'dangerous' : approve only the sharp-edge ops (navigate, eval). Default.\\\" but `ALWAYS_CONFIRM_KINDS` also includes `newTab` and `closeTab` (per the comment directly above it, \\\"`newTab` is navigation-equivalent... and `closeTab` is destructive, so both sit alongside navigate/eval\\\"). Update the policy docstring to mention all four ops so readers don't undercount what \\\"dangerous\\\" actually gates.\\n\\n
\\n
📝 Proposed fix
\\n\\n```diff\\n * - 'off' : no per-call approval (the interaction permission alone gates).\\n- * - 'dangerous' : approve only the sharp-edge ops (navigate, eval). Default.\\n+ * - 'dangerous' : approve only the sharp-edge ops (navigate, eval, newTab, closeTab). Default.\\n * - 'all' : approve every interaction op. */\\n```\\n
\\n\\n
\\n
🤖 Prompt for AI Agents
\\n\\n```\\nVerify each finding against current code. Fix only still-valid issues, skip the\\nrest with a brief reason, keep changes minimal, and validate.\\n\\nIn `@src/shared/coworkingBrowser.ts` around lines 77 - 95, The\\nBrowserConfirmPolicy doc comment is stale because the 'dangerous' policy\\ndescription in coworkingBrowser no longer matches the actual always-confirm set.\\nUpdate the comment above BrowserConfirmPolicy and/or\\nDEFAULT_BROWSER_CONFIRM_POLICY so it explicitly lists all ops covered by\\n'dangerous'—navigate, eval, newTab, and closeTab—matching ALWAYS_CONFIRM_KINDS\\nand keeping the policy wording aligned with the implementation.\\n```\\n\\n
\\n\\n\\n\\n
\\n
\\n
src/__tests__/renderer/hooks/coworking/coworkingApproval.test.ts (1)
\\n\\n`61-69`: _🎯 Functional Correctness_ | _🔵 Trivial_ | _⚡ Quick win_\\n\\n**Missing coverage for `newTab`/`closeTab`/`waitFor` under 'dangerous' policy.**\\n\\n`ALWAYS_CONFIRM_KINDS` in `src/shared/coworkingBrowser.ts` includes `newTab` and `closeTab`, but this test only asserts `eval`/`navigate` return `true` and `click`/`type`/`reload`/`back`/`screenshot` return `false` for the `'dangerous'` policy. Add assertions for `newTab`, `closeTab` (expected `true`) and `waitFor` (expected `false`) to lock in the intended confirm-policy contract and guard against regressions when the always-confirm list changes.\\n\\n
\\n
✅ Proposed additional assertions
\\n\\n```diff\\n \\tit('dangerous confirms only the sharp-edge ops', () => {\\n \\t\\texpect(browserOpNeedsConfirm('dangerous', 'eval')).toBe(true);\\n \\t\\texpect(browserOpNeedsConfirm('dangerous', 'navigate')).toBe(true);\\n+\\t\\texpect(browserOpNeedsConfirm('dangerous', 'newTab')).toBe(true);\\n+\\t\\texpect(browserOpNeedsConfirm('dangerous', 'closeTab')).toBe(true);\\n+\\t\\texpect(browserOpNeedsConfirm('dangerous', 'waitFor')).toBe(false);\\n \\t\\texpect(browserOpNeedsConfirm('dangerous', 'click')).toBe(false);\\n \\t\\texpect(browserOpNeedsConfirm('dangerous', 'type')).toBe(false);\\n \\t\\texpect(browserOpNeedsConfirm('dangerous', 'reload')).toBe(false);\\n \\t\\texpect(browserOpNeedsConfirm('dangerous', 'back')).toBe(false);\\n \\t\\texpect(browserOpNeedsConfirm('dangerous', 'screenshot')).toBe(false);\\n \\t});\\n```\\n
\\n\\n
\\n
🤖 Prompt for AI Agents
\\n\\n```\\nVerify each finding against current code. Fix only still-valid issues, skip the\\nrest with a brief reason, keep changes minimal, and validate.\\n\\nIn `@src/__tests__/renderer/hooks/coworking/coworkingApproval.test.ts` around\\nlines 61 - 69, The `browserOpNeedsConfirm` coverage for the `'dangerous'` policy\\nis incomplete in `coworkingApproval.test.ts`. Update the existing `dangerous\\nconfirms only the sharp-edge ops` test to also assert that `newTab` and\\n`closeTab` return true, and that `waitFor` returns false, matching the intent of\\n`ALWAYS_CONFIRM_KINDS` in `coworkingBrowser.ts`. Keep the assertions grouped\\nwith the other `browserOpNeedsConfirm` checks so this contract stays protected\\nif the confirm list changes.\\n```\\n\\n
\\n\\n\\n\\n
\\n
\\n
src/main/coworking/coworking-registry.ts (1)
\\n\\n`129-170`: _📐 Maintainability & Code Quality_ | _🔵 Trivial_ | _💤 Low value_\\n\\n**`syncSessionBrowsers` accumulates several positional params.**\\n\\nThe signature now takes 5 params (2 optional trailing) — `sessionId, inputs, interactionEnabled, agentType?, confirmPolicy?`. Logic itself is correct and well-tested (stable id assignment, hidden-tab handling verified in `coworking-browser.test.ts`), but as more per-session settings get mirrored here this call site risks parameter-order mistakes. Consider bundling the trailing settings into an options object at the next call-site change.\\n\\n
\\n
🤖 Prompt for AI Agents
\\n\\n```\\nVerify each finding against current code. Fix only still-valid issues, skip the\\nrest with a brief reason, keep changes minimal, and validate.\\n\\nIn `@src/main/coworking/coworking-registry.ts` around lines 129 - 170,\\nsyncSessionBrowsers in CoworkingRegistry now takes several positional session\\nsettings, which makes call-site mistakes more likely as the mirrored options\\ngrow. Keep the existing behavior, but when you touch the related call sites,\\nrefactor the trailing parameters after interactionEnabled into a single options\\nobject and update syncSessionBrowsers to read agentType and confirmPolicy from\\nthat object. Use the syncSessionBrowsers method and its current sessionId,\\ninputs, interactionEnabled, agentType, and confirmPolicy parameters as the\\nanchor points.\\n```\\n\\n
\\n\\n\\n\\n
\\n
\\n
src/main/app-lifecycle/window-manager.ts (1)
\\n\\n`76-81`: _🔒 Security & Privacy_ | _🔵 Trivial_ | _⚡ Quick win_\\n\\n**Duplicate partition allowlist across main-process files.**\\n\\nThis exact `isAllowedBrowserTabPartition` implementation is also present in `src/main/ipc/handlers/browser-session.ts` (per supplied graph context). Two independent copies of a security-relevant allowlist function invite drift — if one is updated (new prefix, stricter check) and the other isn't, webview attachment and session-clear gating could silently diverge.\\n\\nConsider extracting this to a shared module (e.g. a `browserPartitions.ts` util) imported by both files.\\n\\n\\n\\n
\\n
♻️ Suggested extraction
\\n\\n```diff\\n+// src/main/app-lifecycle/browserPartitionAllowlist.ts\\n+export const BROWSER_TAB_PARTITION_PREFIX = 'persist:maestro-browser-session-';\\n+export const EPHEMERAL_BROWSER_TAB_PARTITION_PREFIX = 'maestro-ephemeral-';\\n+export function isAllowedBrowserTabPartition(partition: string): boolean {\\n+ return (\\n+ partition.startsWith(BROWSER_TAB_PARTITION_PREFIX) ||\\n+ partition.startsWith(EPHEMERAL_BROWSER_TAB_PARTITION_PREFIX)\\n+ );\\n+}\\n```\\n\\nThen import this in both `window-manager.ts` and `browser-session.ts` instead of maintaining two copies.\\n
\\n\\n
\\n
🤖 Prompt for AI Agents
\\n\\n```\\nVerify each finding against current code. Fix only still-valid issues, skip the\\nrest with a brief reason, keep changes minimal, and validate.\\n\\nIn `@src/main/app-lifecycle/window-manager.ts` around lines 76 - 81, The allowlist\\nlogic in isAllowedBrowserTabPartition is duplicated in another main-process\\nmodule, so extract this security-sensitive check into a shared helper (for\\nexample a browserPartitions util) and import it from both window-manager and\\nbrowser-session. Keep the existing function name or replace both call sites with\\nthe shared helper so future prefix or validation changes stay in sync.\\n```\\n\\n
\\n\\n\\n\\n
\\n
\\n
src/renderer/hooks/session/useSessionRestoration.ts (1)
\\n\\n`23-23`: _📐 Maintainability & Code Quality_ | _🔵 Trivial_ | _⚡ Quick win_\\n\\n**Ephemeral-tab filtering is correct and matches the persistence-side contract.**\\n\\nFilter-then-rehydrate correctly drops leaked ephemeral tabs and downstream `activeBrowserTabId`/`unifiedTabOrder` repair handles the removal consistently with the test expectations.\\n\\nThe same filter+map pattern (`filter(!isEphemeralBrowserTab).map(rehydrate/sanitize)`) is now duplicated here and in `useDebouncedPersistence.ts`'s `cleanedBrowserTabs`. Consider extracting a shared helper (e.g. `filterPersistableBrowserTabs(tabs, sessionId, transform)`) in `browserTabPersistence.ts` to keep the two call sites from drifting.\\n\\n\\n\\n\\n\\nAlso applies to: 414-419\\n\\n
\\n
🤖 Prompt for AI Agents
\\n\\n```\\nVerify each finding against current code. Fix only still-valid issues, skip the\\nrest with a brief reason, keep changes minimal, and validate.\\n\\nIn `@src/renderer/hooks/session/useSessionRestoration.ts` at line 23, The\\nephemeral-tab filtering logic is duplicated between useSessionRestoration and\\nuseDebouncedPersistence, so extract the shared filter-then-transform behavior\\ninto a helper in browserTabPersistence.ts (for example, a\\nfilterPersistableBrowserTabs function that applies isEphemeralBrowserTab plus\\nrehydrateBrowserTab/sanitization) and update both call sites to use it. Keep the\\nexisting behavior for activeBrowserTabId and unifiedTabOrder repairs unchanged\\nwhile centralizing the browser tab persistence contract.\\n```\\n\\n
\\n\\n\\n\\n
\\n
\\n
src/__tests__/renderer/components/MainPanel/BrowserTabView.test.tsx (1)
\\n\\n`746-846`: _🎯 Functional Correctness_ | _🔵 Trivial_ | _⚡ Quick win_\\n\\n**Solid coverage for the two-step confirm flow; consider adding a tab-switch case.**\\n\\nThese tests exercise arm/confirm/disarm/error paths well. None of them cover switching `tab` (e.g. via `rerender` with a different `tab.id`) while `clearArmed` is true, which is exactly the scenario in the sibling comment on `BrowserTabView.tsx` (arm state carrying over to a different tab). Once that fix lands, a regression test here would lock in the corrected behavior.\\n\\n
\\n
🤖 Prompt for AI Agents
\\n\\n```\\nVerify each finding against current code. Fix only still-valid issues, skip the\\nrest with a brief reason, keep changes minimal, and validate.\\n\\nIn `@src/__tests__/renderer/components/MainPanel/BrowserTabView.test.tsx` around\\nlines 746 - 846, Add a regression test in BrowserTabView.test.tsx for switching\\ntabs while the clear-session confirm state is armed. In the clear browsing data\\n+ incognito badge suite, use BrowserTabView and rerender from one tab.id to a\\ndifferent tab.id after the first click on browser-tab-clear-session-data, then\\nassert the armed state does not carry over and the second click on the new tab\\nre-arms instead of calling browserSession.clearSessionData. This should cover\\nthe clearArmed state reset behavior tied to BrowserTabView and the clear-session\\nbutton flow.\\n```\\n\\n
\\n\\n\\n\\n
\\n
\\n
src/main/preload/coworking.ts (1)
\\n\\n`18-33`: _📐 Maintainability & Code Quality_ | _🔵 Trivial_ | _⚡ Quick win_\\n\\n**Duplicated type definitions across main/preload/renderer.**\\n\\n`CoworkingTerminalEntry`, `CoworkingTerminalRecord`, and `CoworkingInstallStatus` are independently redefined here, in `src/main/coworking/coworking-types.ts`, and a third time inline in `src/renderer/global.d.ts`. A future field change to any one of them (e.g. adding a field to the registry record) can silently drift out of sync with the other two since nothing enforces structural equality across files.\\n\\nConsider importing these as `import type` from a single shared source (e.g. `coworking-types.ts`, since preload-side `import type` has no runtime bundling cost) to keep the contract in one place.\\n\\n\\n\\n\\n
\\n
♻️ Example consolidation
\\n\\n```diff\\n-export interface CoworkingTerminalEntry {\\n-\\tid: string;\\n-\\tcwd: string;\\n-\\ttitle: string;\\n-}\\n-\\n-export interface CoworkingTerminalRecord extends CoworkingTerminalEntry {\\n-\\ttabUuid: string;\\n-\\tsessionId: string;\\n-}\\n-\\n-export interface CoworkingInstallStatus {\\n-\\tagentId: string;\\n-\\tconfigPath: string;\\n-\\tinstalled: boolean;\\n-}\\n+import type {\\n+\\tCoworkingTerminalEntry,\\n+\\tCoworkingTerminalRecord,\\n+\\tCoworkingInstallStatus,\\n+} from '../coworking/coworking-types';\\n```\\n
\\n\\n
\\n
🤖 Prompt for AI Agents
\\n\\n```\\nVerify each finding against current code. Fix only still-valid issues, skip the\\nrest with a brief reason, keep changes minimal, and validate.\\n\\nIn `@src/main/preload/coworking.ts` around lines 18 - 33, The coworking type\\ninterfaces are duplicated across preload, main, and renderer, so update the\\npreload definitions in CoworkingTerminalEntry, CoworkingTerminalRecord, and\\nCoworkingInstallStatus to come from a single shared source instead of being\\nredefined locally. Move or reuse the canonical types from coworking-types.ts and\\nswitch the preload-side usage to import type so there is no runtime cost, then\\nupdate src/renderer/global.d.ts and any main-side declarations to reference the\\nsame shared interfaces and keep the contract consistent.\\n```\\n\\n
\\n\\n\\n\\n
\\n
\\n
src/renderer/global.d.ts (1)
\\n\\n`3746-3826`: _📐 Maintainability & Code Quality_ | _🔵 Trivial_ | _⚡ Quick win_\\n\\n**Hand-duplicated `BrowserOp` union and terminal/browser record shapes.**\\n\\nThe `op` union in `onRequestBrowserOp` (lines 3798-3811) and the terminal/browser record shapes (lines 3757-3789) are manually retyped copies of `BrowserOp`/`CoworkingBrowserInput` from `src/shared/coworkingBrowser.ts`. This file already uses inline `import('../shared/...')` type expressions elsewhere (e.g. `DirectorNotesNarrative` at line 3558), so the same pattern could be used here to keep this ambient declaration in sync with the shared source of truth instead of risking drift when a new `BrowserOp` kind is added.\\n\\n\\n\\n\\n
\\n
♻️ Example using inline shared-type import
\\n\\n```diff\\n \\t\\tonRequestBrowserOp: (\\n \\t\\t\\tcallback: (\\n \\t\\t\\t\\ttabUuid: string,\\n \\t\\t\\t\\tsessionId: string,\\n-\\t\\t\\t\\top:\\n-\\t\\t\\t\\t\\t| { kind: 'read'; format: 'text' | 'innerText' | 'html'; selector?: string }\\n-\\t\\t\\t\\t\\t| { kind: 'navigate'; url: string }\\n-\\t\\t\\t\\t\\t| { kind: 'back' }\\n-\\t\\t\\t\\t\\t| { kind: 'forward' }\\n-\\t\\t\\t\\t\\t| { kind: 'reload' }\\n-\\t\\t\\t\\t\\t| { kind: 'stop' }\\n-\\t\\t\\t\\t\\t| { kind: 'click'; selector: string }\\n-\\t\\t\\t\\t\\t| { kind: 'type'; selector: string; text: string }\\n-\\t\\t\\t\\t\\t| { kind: 'eval'; code: string }\\n-\\t\\t\\t\\t\\t| { kind: 'screenshot' }\\n-\\t\\t\\t\\t\\t| { kind: 'waitFor'; selector: string; timeoutMs?: number }\\n-\\t\\t\\t\\t\\t| { kind: 'newTab'; url?: string; ephemeral?: boolean }\\n-\\t\\t\\t\\t\\t| { kind: 'closeTab' },\\n+\\t\\t\\t\\top: import('../shared/coworkingBrowser').BrowserOp,\\n \\t\\t\\t\\tresponseChannel: string,\\n \\t\\t\\t\\tneedsConfirm?: boolean\\n \\t\\t\\t) => void\\n \\t\\t) => () => void;\\n```\\n
\\n\\n
\\n
🤖 Prompt for AI Agents
\\n\\n```\\nVerify each finding against current code. Fix only still-valid issues, skip the\\nrest with a brief reason, keep changes minimal, and validate.\\n\\nIn `@src/renderer/global.d.ts` around lines 3746 - 3826, The coworking API types\\nin the ambient declaration are duplicated by hand, which can drift from the\\nshared source types. Replace the inline record and browser-op shapes in\\ncoworking with the corresponding shared type references from\\nsrc/shared/coworkingBrowser.ts, following the existing inline import pattern\\nused elsewhere in this file (for example the DirectorNotesNarrative type).\\nUpdate the affected symbols on coworking.getInstallStatus, syncSessionTerminals,\\nupsertTerminal, syncSessionBrowsers, and onRequestBrowserOp so the declaration\\nstays aligned with the source of truth.\\n```\\n\\n
\\n\\n\\n\\n
\\n
\\n
src/renderer/components/Settings/CoworkingSetup.tsx (1)
\\n\\n`20-20`: _📐 Maintainability & Code Quality_ | _🔵 Trivial_ | _⚡ Quick win_\\n\\n**Hardcoded default policy duplicates `DEFAULT_BROWSER_CONFIRM_POLICY`.**\\n\\n`browserConfirm[s.agentId] ?? 'dangerous'` re-implements the default that `useCoworkingRegistrySync.ts` sources from the shared `DEFAULT_BROWSER_CONFIRM_POLICY` constant. If that constant ever changes, this UI would silently show the wrong policy as selected while the actual enforced default diverges.\\n\\n
\\n
♻️ Import the shared constant instead of hardcoding
\\n\\n```diff\\n-import type { BrowserConfirmPolicy } from '../../../shared/coworkingBrowser';\\n+import { DEFAULT_BROWSER_CONFIRM_POLICY, type BrowserConfirmPolicy } from '../../../shared/coworkingBrowser';\\n```\\n\\n```diff\\n-\\t\\t\\t\\t\\tconst confirmPolicy = browserConfirm[s.agentId] ?? 'dangerous';\\n+\\t\\t\\t\\t\\tconst confirmPolicy = browserConfirm[s.agentId] ?? DEFAULT_BROWSER_CONFIRM_POLICY;\\n```\\n
\\n\\n\\n\\n\\n\\nAlso applies to: 280-280\\n\\n
\\n
🤖 Prompt for AI Agents
\\n\\n```\\nVerify each finding against current code. Fix only still-valid issues, skip the\\nrest with a brief reason, keep changes minimal, and validate.\\n\\nIn `@src/renderer/components/Settings/CoworkingSetup.tsx` at line 20, The\\ncoworking setup UI is hardcoding the fallback browser confirm policy instead of\\nusing the shared default. Update CoworkingSetup.tsx to import and use\\nDEFAULT_BROWSER_CONFIRM_POLICY from the shared coworkingBrowser module wherever\\nbrowserConfirm[s.agentId] falls back, so the selected policy stays in sync with\\nuseCoworkingRegistrySync and any future default changes.\\n```\\n\\n
\\n\\n\\n\\n
\\n
\\n
src/renderer/components/coworking/CoworkingApprovalHost.tsx (1)
\\n\\n`16-24`: _🎯 Functional Correctness_ | _🔵 Trivial_ | _⚡ Quick win_\\n\\n**Approval outcome relies on ConfirmModal's internal callback order.**\\n\\n`onConfirm` and `onClose` both call `settle` with opposite booleans. This only resolves correctly today because `ConfirmModal.handleConfirm` calls `onConfirm()` before `onClose()`, so the store's `if (!request) return` guard makes the second call a no-op. If that ordering in `ConfirmModal` ever changes, clicking \\\"Allow\\\" would silently resolve the coworking approval as denied instead of approved.\\n\\nConsider guarding against double-settlement explicitly rather than relying on this coupling.\\n\\n
\\n
♻️ Suggested guard
\\n\\n```diff\\n+import { useRef } from 'react';\\n import { ConfirmModal } from '../ConfirmModal';\\n import type { Theme } from '../../types';\\n import { useCoworkingApprovalStore } from '../../stores/coworkingApprovalStore';\\n\\n export function CoworkingApprovalHost({ theme }: { theme: Theme }) {\\n \\tconst front = useCoworkingApprovalStore((s) => s.queue[0]);\\n \\tconst settle = useCoworkingApprovalStore((s) => s.settle);\\n+\\tconst settledRef = useRef(false);\\n \\tif (!front) return null;\\n+\\tsettledRef.current = false;\\n+\\tconst settleOnce = (approved: boolean) => {\\n+\\t\\tif (settledRef.current) return;\\n+\\t\\tsettledRef.current = true;\\n+\\t\\tsettle(front.id, approved);\\n+\\t};\\n \\treturn (\\n \\t\\t settle(front.id, true)}\\n-\\t\\t\\tonClose={() => settle(front.id, false)}\\n+\\t\\t\\tonConfirm={() => settleOnce(true)}\\n+\\t\\t\\tonClose={() => settleOnce(false)}\\n \\t\\t/>\\n \\t);\\n }\\n```\\n
\\n\\n
\\n
🤖 Prompt for AI Agents
\\n\\n```\\nVerify each finding against current code. Fix only still-valid issues, skip the\\nrest with a brief reason, keep changes minimal, and validate.\\n\\nIn `@src/renderer/components/coworking/CoworkingApprovalHost.tsx` around lines 16\\n- 24, The coworking approval flow is coupled to ConfirmModal’s internal callback\\norder because both onConfirm and onClose call settle with opposite values.\\nUpdate CoworkingApprovalHost so the approval result is resolved explicitly once\\nper action, and do not rely on ConfirmModal.handleConfirm invoking callbacks in\\na specific order. Use the existing settle logic in CoworkingApprovalHost and, if\\nneeded, add a guard to prevent a second resolution from overwriting the first.\\n```\\n\\n
\\n\\n\\n\\n
\\n
\\n
src/main/coworking/coworking-audit.ts (1)
\\n\\n`91-108`: _🚀 Performance & Scalability_ | _🔵 Trivial_ | _⚡ Quick win_\\n\\n**Audit log has no size cap or rotation.**\\n\\n`coworking-browser-audit.jsonl` grows forever under `userData` with no rotation or size limit, and each write is a synchronous `fs.appendFileSync` on the main process. Over a long-lived install with frequent tool calls this can accumulate unbounded disk usage.\\n\\nConsider capping file size (rotate/truncate past a threshold) and/or switching to an async, batched writer so a burst of tool calls doesn't block the main thread.\\n\\n
\\n
🤖 Prompt for AI Agents
\\n\\n```\\nVerify each finding against current code. Fix only still-valid issues, skip the\\nrest with a brief reason, keep changes minimal, and validate.\\n\\nIn `@src/main/coworking/coworking-audit.ts` around lines 91 - 108, The\\ncreateDefaultBrowserAuditSink() audit writer appends endlessly to\\ncoworking-browser-audit.jsonl and uses synchronous fs.appendFileSync on every\\nentry, so update this sink to enforce a size cap or rotation/truncation strategy\\nand avoid blocking writes. Use the existing createDefaultBrowserAuditSink,\\nlogger.info, and the file path in app.getPath('userData') as the place to add\\nlogic that checks the current audit file size before writing, rotates or\\ntruncates when it exceeds a threshold, and preferably switches the write path to\\nan async or batched approach to keep the main process responsive.\\n```\\n\\n
\\n\\n\\n\\n
\\n
\\n
src/main/coworking/installers/claude-code.ts (1)
\\n\\n`31-35`: _🗄️ Data Integrity & Integration_ | _🔵 Trivial_ | _⚡ Quick win_\\n\\n**Non-atomic write to the user's live `~/.claude.json`.**\\n\\n`writeConfig` writes directly to the destination path. If the write is interrupted (crash, disk full), the user's global Claude Code config — not just the coworking entry — can end up truncated/corrupted. Writing to a temp file in the same directory and renaming is atomic on POSIX and NTFS and costs little.\\n\\n\\n\\n\\n
\\n
♻️ Suggested atomic-write pattern
\\n\\n```diff\\n async function writeConfig(config: unknown): Promise {\\n \\tawait fs.promises.mkdir(path.dirname(configPath()), { recursive: true });\\n \\tconst out = commentJson.stringify(config, null, 2) + '\\\\n';\\n-\\tawait fs.promises.writeFile(configPath(), out, 'utf8');\\n+\\tconst tmpPath = `${configPath()}.tmp-${process.pid}`;\\n+\\tawait fs.promises.writeFile(tmpPath, out, 'utf8');\\n+\\tawait fs.promises.rename(tmpPath, configPath());\\n }\\n```\\n
\\n\\n
\\n
🤖 Prompt for AI Agents
\\n\\n```\\nVerify each finding against current code. Fix only still-valid issues, skip the\\nrest with a brief reason, keep changes minimal, and validate.\\n\\nIn `@src/main/coworking/installers/claude-code.ts` around lines 31 - 35, The\\nwriteConfig helper writes directly to the live configPath(), which can leave the\\nuser’s ~/.claude.json corrupted if interrupted. Update writeConfig to use an\\natomic write flow: generate the JSON in a temporary file in the same directory\\nas configPath(), flush/write it fully, then rename it into place. Keep the fix\\nlocalized to writeConfig and the configPath() usage so the coworking installer\\npreserves the existing config safely.\\n```\\n\\n
\\n\\n\\n\\n
\\n
\\n
src/main/coworking/installers/codex.ts (1)
\\n\\n`31-34`: _🗄️ Data Integrity & Integration_ | _🔵 Trivial_ | _⚡ Quick win_\\n\\n**Non-atomic write to the user's live `config.toml`.**\\n\\nSame concern as `claude-code.ts`: a direct `writeFile` over the destination risks leaving `~/.codex/config.toml` truncated if interrupted mid-write.\\n\\n\\n\\n
\\n
♻️ Suggested atomic-write pattern
\\n\\n```diff\\n async function writeConfig(content: string): Promise {\\n \\tawait fs.promises.mkdir(path.dirname(configPath()), { recursive: true });\\n-\\tawait fs.promises.writeFile(configPath(), content, 'utf8');\\n+\\tconst tmpPath = `${configPath()}.tmp-${process.pid}`;\\n+\\tawait fs.promises.writeFile(tmpPath, content, 'utf8');\\n+\\tawait fs.promises.rename(tmpPath, configPath());\\n }\\n```\\n
\\n\\n
\\n
🤖 Prompt for AI Agents
\\n\\n```\\nVerify each finding against current code. Fix only still-valid issues, skip the\\nrest with a brief reason, keep changes minimal, and validate.\\n\\nIn `@src/main/coworking/installers/codex.ts` around lines 31 - 34, The writeConfig\\nhelper currently writes directly to configPath(), which can leave the live\\nconfig.toml truncated if interrupted. Update writeConfig in codex.ts to use an\\natomic-write flow: write the content to a temporary file in the same directory,\\nthen rename it over the destination after the write completes. Keep the existing\\nconfigPath() resolution and mkdir behavior, but ensure the final persisted write\\nhappens via rename rather than direct writeFile to the target.\\n```\\n\\n
\\n\\n\\n\\n
\\n
\\n
src/main/coworking/installers/factory-droid.ts (1)
\\n\\n`19-35`: _📐 Maintainability & Code Quality_ | _🔵 Trivial_ | _⚡ Quick win_\\n\\n**Duplicate read/write-config boilerplate across installers.**\\n\\n`readConfig`/`writeConfig` here are nearly identical to the ones in `opencode.ts` (and presumably `claude-code.ts`/`codex.ts`). Extracting a shared `readJsoncConfig(path)`/`writeJsoncConfig(path, config)` helper (parameterized by path) would remove this duplication across all per-agent installers.\\n\\n\\n\\n\\n
\\n
♻️ Suggested shared helper
\\n\\n```typescript\\n// src/main/coworking/installers/jsonc-file.ts\\nimport * as fs from 'fs';\\nimport * as path from 'path';\\nconst commentJson = require('comment-json');\\n\\nexport async function readJsoncConfig(configPath: string): Promise {\\n\\ttry {\\n\\t\\tconst raw = await fs.promises.readFile(configPath, 'utf8');\\n\\t\\treturn commentJson.parse(raw);\\n\\t} catch (err) {\\n\\t\\tif ((err as NodeJS.ErrnoException)?.code === 'ENOENT') {\\n\\t\\t\\treturn commentJson.parse('{}');\\n\\t\\t}\\n\\t\\tthrow err;\\n\\t}\\n}\\n\\nexport async function writeJsoncConfig(configPath: string, config: unknown): Promise {\\n\\tawait fs.promises.mkdir(path.dirname(configPath), { recursive: true });\\n\\tconst out = commentJson.stringify(config, null, 2) + '\\\\n';\\n\\tawait fs.promises.writeFile(configPath, out, 'utf8');\\n}\\n```\\n
\\n\\n
\\n
🤖 Prompt for AI Agents
\\n\\n```\\nVerify each finding against current code. Fix only still-valid issues, skip the\\nrest with a brief reason, keep changes minimal, and validate.\\n\\nIn `@src/main/coworking/installers/factory-droid.ts` around lines 19 - 35, The\\nreadConfig/writeConfig logic in factory-droid.ts is duplicated across the other\\ninstallers, so extract it into a shared helper such as readJsoncConfig and\\nwriteJsoncConfig that accepts the target path. Move the common comment-json\\nparse/stringify plus ENOENT handling and mkdir/write behavior into a reusable\\nmodule, then update factory-droid and the matching installer functions like\\nopencode.ts, claude-code.ts, and codex.ts to call the shared helper instead of\\nkeeping local boilerplate.\\n```\\n\\n
\\n\\n\\n\\n
\\n\\n

\\n\\n

\\n

🤖 Prompt for all review comments with AI agents

\\n\\n```\\nVerify each finding against current code. Fix only still-valid issues, skip the\\nrest with a brief reason, keep changes minimal, and validate.\\n\\nInline comments:\\nIn `@src/main/coworking/coworking-bridge.ts`:\\n- Around line 337-346: The listTerminals branch in coworking-bridge.ts records\\nan ok audit before the tool result is known, unlike readTerminal, listBrowsers,\\ngetBrowserUrl, readBrowser, and browserInteract which use auditedBrowserCall.\\nUpdate the listTerminals handling to wrap listTerminals(sessionId) in\\nauditedBrowserCall so success/error is recorded based on the actual outcome, and\\nremove the unconditional pre-call recordBrowserAudit call.\\n- Around line 404-413: `readBrowser` currently validates `maxChars` in\\n`coworking-bridge.ts` only for being a positive integer, so it can still accept\\nunbounded values and create oversized in-memory/IPC payloads. Update the\\n`maxChars` validation in the request handling block for `params.maxChars` to\\nenforce a reasonable upper limit (or clamp it) alongside the existing\\npositive-integer check, and return the same `-32602` validation error when the\\nvalue exceeds that ceiling.\\n- Around line 95-102: The Windows pipe name in getBridgeSocketPath is not truly\\nper-user because path.basename(app.getPath('userData')) only uses the shared app\\nfolder name. Update the Windows branch in getBridgeSocketPath to derive the slug\\nfrom the full app.getPath('userData') value, such as hashing it or otherwise\\nencoding the user-specific path, while keeping the existing pipe format in\\ncoworking-bridge.ts.\\n\\nIn `@src/main/coworking/installers/codex.ts`:\\n- Around line 96-113: The install/uninstall newline handling in codex.ts is not\\nreversible for inputs with no trailing newline or with an existing blank line\\nbefore the injected block. Update the logic in install() and uninstall() to stop\\nguessing original spacing from the merged file; instead normalize trailing\\nnewlines deterministically so remove/add operations are idempotent. Add\\nregression tests in codex.test.ts for a config without a trailing newline and\\none that already ends with a blank line before the block.\\n\\nIn `@src/renderer/components/MainPanel/BrowserTabView.tsx`:\\n- Around line 188-199: The two-step clear-browsing-data guard in BrowserTabView\\nneeds to be reset when the active tab changes, because clearArmed and\\nclearDisarmTimerRef currently persist across reused component instances. Update\\nthe state reset logic alongside the existing tab-scoped effects that handle\\naddressValue and isDomReadyRef so that switching tabs clears any armed state and\\ncancels the pending timeout. Use the BrowserTabView tab-id/url effects and the\\nclearDisarmTimerRef cleanup to locate the right place for the fix.\\n\\nIn `@src/renderer/components/MainPanel/MainPanel.tsx`:\\n- Around line 1097-1098: The coworking host components are only rendered in the\\nfinal normal-session branch, so early returns like logViewerOpen,\\nagentSessionsOpen, memoryViewerOpen, and !activeSession bypass them. Update\\nMainPanel so CoworkingApprovalHost and CoworkingBackgroundBrowsers are mounted\\nfrom a shared top-level wrapper or hoisted above the conditional return logic,\\nensuring they stay active regardless of which view is displayed.\\n\\nIn `@src/renderer/hooks/coworking/useCoworkingBrowserResponder.ts`:\\n- Around line 123-161: createTabForSession can return a success result even when\\nthe target session was removed before the state update runs. Update\\ncreateTabForSession to track whether setSessions actually matched sessionId,\\nsimilar to the closed flag pattern in closeTabForSession, and if no session was\\nupdated return ok: false with an appropriate message instead of reporting\\nsuccess. Use the existing createTabForSession, setSessions, and useSessionStore\\nflow to locate the fix.\\n\\nIn `@src/renderer/hooks/coworking/useCoworkingRegistrySync.ts`:\\n- Around line 106-208: The async sync in useCoworkingRegistrySync can roll back\\nshared refs from an older batch after a newer batch has already started or\\ncompleted. Add a monotonic run/generation identifier inside the useEffect sync\\npath and capture it before starting the async IIFE, then only apply the failure\\nrollback to lastPayloadRef and lastSessionIdsRef when the failing run is still\\nthe latest one. Make sure the same guard is applied to the disable/clear flow\\naround bridge.removeSession so stale teardown work cannot overwrite newer sync\\nstate.\\n\\n---\\n\\nNitpick comments:\\nIn `@src/__tests__/renderer/components/MainPanel/BrowserTabView.test.tsx`:\\n- Around line 746-846: Add a regression test in BrowserTabView.test.tsx for\\nswitching tabs while the clear-session confirm state is armed. In the clear\\nbrowsing data + incognito badge suite, use BrowserTabView and rerender from one\\ntab.id to a different tab.id after the first click on\\nbrowser-tab-clear-session-data, then assert the armed state does not carry over\\nand the second click on the new tab re-arms instead of calling\\nbrowserSession.clearSessionData. This should cover the clearArmed state reset\\nbehavior tied to BrowserTabView and the clear-session button flow.\\n\\nIn `@src/__tests__/renderer/hooks/coworking/coworkingApproval.test.ts`:\\n- Around line 61-69: The `browserOpNeedsConfirm` coverage for the `'dangerous'`\\npolicy is incomplete in `coworkingApproval.test.ts`. Update the existing\\n`dangerous confirms only the sharp-edge ops` test to also assert that `newTab`\\nand `closeTab` return true, and that `waitFor` returns false, matching the\\nintent of `ALWAYS_CONFIRM_KINDS` in `coworkingBrowser.ts`. Keep the assertions\\ngrouped with the other `browserOpNeedsConfirm` checks so this contract stays\\nprotected if the confirm list changes.\\n\\nIn `@src/main/app-lifecycle/window-manager.ts`:\\n- Around line 76-81: The allowlist logic in isAllowedBrowserTabPartition is\\nduplicated in another main-process module, so extract this security-sensitive\\ncheck into a shared helper (for example a browserPartitions util) and import it\\nfrom both window-manager and browser-session. Keep the existing function name or\\nreplace both call sites with the shared helper so future prefix or validation\\nchanges stay in sync.\\n\\nIn `@src/main/coworking/coworking-audit.ts`:\\n- Around line 91-108: The createDefaultBrowserAuditSink() audit writer appends\\nendlessly to coworking-browser-audit.jsonl and uses synchronous\\nfs.appendFileSync on every entry, so update this sink to enforce a size cap or\\nrotation/truncation strategy and avoid blocking writes. Use the existing\\ncreateDefaultBrowserAuditSink, logger.info, and the file path in\\napp.getPath('userData') as the place to add logic that checks the current audit\\nfile size before writing, rotates or truncates when it exceeds a threshold, and\\npreferably switches the write path to an async or batched approach to keep the\\nmain process responsive.\\n\\nIn `@src/main/coworking/coworking-registry.ts`:\\n- Around line 129-170: syncSessionBrowsers in CoworkingRegistry now takes\\nseveral positional session settings, which makes call-site mistakes more likely\\nas the mirrored options grow. Keep the existing behavior, but when you touch the\\nrelated call sites, refactor the trailing parameters after interactionEnabled\\ninto a single options object and update syncSessionBrowsers to read agentType\\nand confirmPolicy from that object. Use the syncSessionBrowsers method and its\\ncurrent sessionId, inputs, interactionEnabled, agentType, and confirmPolicy\\nparameters as the anchor points.\\n\\nIn `@src/main/coworking/installers/claude-code.ts`:\\n- Around line 31-35: The writeConfig helper writes directly to the live\\nconfigPath(), which can leave the user’s ~/.claude.json corrupted if\\ninterrupted. Update writeConfig to use an atomic write flow: generate the JSON\\nin a temporary file in the same directory as configPath(), flush/write it fully,\\nthen rename it into place. Keep the fix localized to writeConfig and the\\nconfigPath() usage so the coworking installer preserves the existing config\\nsafely.\\n\\nIn `@src/main/coworking/installers/codex.ts`:\\n- Around line 31-34: The writeConfig helper currently writes directly to\\nconfigPath(), which can leave the live config.toml truncated if interrupted.\\nUpdate writeConfig in codex.ts to use an atomic-write flow: write the content to\\na temporary file in the same directory, then rename it over the destination\\nafter the write completes. Keep the existing configPath() resolution and mkdir\\nbehavior, but ensure the final persisted write happens via rename rather than\\ndirect writeFile to the target.\\n\\nIn `@src/main/coworking/installers/factory-droid.ts`:\\n- Around line 19-35: The readConfig/writeConfig logic in factory-droid.ts is\\nduplicated across the other installers, so extract it into a shared helper such\\nas readJsoncConfig and writeJsoncConfig that accepts the target path. Move the\\ncommon comment-json parse/stringify plus ENOENT handling and mkdir/write\\nbehavior into a reusable module, then update factory-droid and the matching\\ninstaller functions like opencode.ts, claude-code.ts, and codex.ts to call the\\nshared helper instead of keeping local boilerplate.\\n\\nIn `@src/main/preload/coworking.ts`:\\n- Around line 18-33: The coworking type interfaces are duplicated across\\npreload, main, and renderer, so update the preload definitions in\\nCoworkingTerminalEntry, CoworkingTerminalRecord, and CoworkingInstallStatus to\\ncome from a single shared source instead of being redefined locally. Move or\\nreuse the canonical types from coworking-types.ts and switch the preload-side\\nusage to import type so there is no runtime cost, then update\\nsrc/renderer/global.d.ts and any main-side declarations to reference the same\\nshared interfaces and keep the contract consistent.\\n\\nIn `@src/renderer/components/coworking/CoworkingApprovalHost.tsx`:\\n- Around line 16-24: The coworking approval flow is coupled to ConfirmModal’s\\ninternal callback order because both onConfirm and onClose call settle with\\nopposite values. Update CoworkingApprovalHost so the approval result is resolved\\nexplicitly once per action, and do not rely on ConfirmModal.handleConfirm\\ninvoking callbacks in a specific order. Use the existing settle logic in\\nCoworkingApprovalHost and, if needed, add a guard to prevent a second resolution\\nfrom overwriting the first.\\n\\nIn `@src/renderer/components/Settings/CoworkingSetup.tsx`:\\n- Line 20: The coworking setup UI is hardcoding the fallback browser confirm\\npolicy instead of using the shared default. Update CoworkingSetup.tsx to import\\nand use DEFAULT_BROWSER_CONFIRM_POLICY from the shared coworkingBrowser module\\nwherever browserConfirm[s.agentId] falls back, so the selected policy stays in\\nsync with useCoworkingRegistrySync and any future default changes.\\n\\nIn `@src/renderer/global.d.ts`:\\n- Around line 3746-3826: The coworking API types in the ambient declaration are\\nduplicated by hand, which can drift from the shared source types. Replace the\\ninline record and browser-op shapes in coworking with the corresponding shared\\ntype references from src/shared/coworkingBrowser.ts, following the existing\\ninline import pattern used elsewhere in this file (for example the\\nDirectorNotesNarrative type). Update the affected symbols on\\ncoworking.getInstallStatus, syncSessionTerminals, upsertTerminal,\\nsyncSessionBrowsers, and onRequestBrowserOp so the declaration stays aligned\\nwith the source of truth.\\n\\nIn `@src/renderer/hooks/session/useSessionRestoration.ts`:\\n- Line 23: The ephemeral-tab filtering logic is duplicated between\\nuseSessionRestoration and useDebouncedPersistence, so extract the shared\\nfilter-then-transform behavior into a helper in browserTabPersistence.ts (for\\nexample, a filterPersistableBrowserTabs function that applies\\nisEphemeralBrowserTab plus rehydrateBrowserTab/sanitization) and update both\\ncall sites to use it. Keep the existing behavior for activeBrowserTabId and\\nunifiedTabOrder repairs unchanged while centralizing the browser tab persistence\\ncontract.\\n\\nIn `@src/shared/coworkingBrowser.ts`:\\n- Around line 77-95: The BrowserConfirmPolicy doc comment is stale because the\\n'dangerous' policy description in coworkingBrowser no longer matches the actual\\nalways-confirm set. Update the comment above BrowserConfirmPolicy and/or\\nDEFAULT_BROWSER_CONFIRM_POLICY so it explicitly lists all ops covered by\\n'dangerous'—navigate, eval, newTab, and closeTab—matching ALWAYS_CONFIRM_KINDS\\nand keeping the policy wording aligned with the implementation.\\n```\\n\\n

\\n\\n

\\n

🪄 Autofix (Beta)

\\n\\nFix all unresolved CodeRabbit comments on this PR:\\n\\n- [ ] Push a commit to this branch (recommended)\\n- [ ] Create a new PR with the fixes\\n\\n

\\n\\n---\\n\\n

\\n

ℹ️ Review info

\\n\\n

\\n

⚙️ Run configuration

\\n\\n**Configuration used**: Path: .coderabbit.yaml\\n\\n**Review profile**: CHILL\\n\\n**Plan**: Pro\\n\\n**Run ID**: `d6623277-732d-4106-89f4-429727208bb7`\\n\\n

\\n\\n

\\n

📥 Commits

\\n\\nReviewing files that changed from the base of the PR and between 144967e4fc347f108719146020384e187d4dd9c2 and 98eae338f23ee2b05ae94a13aeda78d397f83ce9.\\n\\n

\\n\\n

\\n

⛔ Files ignored due to path filters (1)

\\n\\n* `package-lock.json` is excluded by `!**/package-lock.json`\\n\\n

\\n\\n

\\n

📒 Files selected for processing (89)

\\n\\n* `package.json`\\n* `src/__tests__/main/app-lifecycle/window-manager.test.ts`\\n* `src/__tests__/main/coworking/coworking-audit.test.ts`\\n* `src/__tests__/main/coworking/coworking-bridge.test.ts`\\n* `src/__tests__/main/coworking/coworking-browser-bridge.test.ts`\\n* `src/__tests__/main/coworking/coworking-browser.test.ts`\\n* `src/__tests__/main/coworking/coworking-registry.test.ts`\\n* `src/__tests__/main/coworking/coworking-session-id.test.ts`\\n* `src/__tests__/main/coworking/coworking-tools.test.ts`\\n* `src/__tests__/main/coworking/coworking-types.test.ts`\\n* `src/__tests__/main/coworking/installers/claude-code.test.ts`\\n* `src/__tests__/main/coworking/installers/codex.test.ts`\\n* `src/__tests__/main/coworking/installers/factory-droid.test.ts`\\n* `src/__tests__/main/coworking/installers/opencode.test.ts`\\n* `src/__tests__/main/coworking/pid-resolution-windows.test.ts`\\n* `src/__tests__/main/coworking/pid-resolution.test.ts`\\n* `src/__tests__/main/ipc/handlers/browser-session.test.ts`\\n* `src/__tests__/renderer/components/MainPanel/BrowserTabView.test.tsx`\\n* `src/__tests__/renderer/components/Settings/CoworkingSetup.test.tsx`\\n* `src/__tests__/renderer/components/TabBar.test.tsx`\\n* `src/__tests__/renderer/hooks/coworking/applyBrowserOp.test.ts`\\n* `src/__tests__/renderer/hooks/coworking/buildBrowserInputs.test.ts`\\n* `src/__tests__/renderer/hooks/coworking/coworkingApproval.test.ts`\\n* `src/__tests__/renderer/hooks/useSessionRestoration.test.ts`\\n* `src/__tests__/renderer/hooks/utils/useDebouncedPersistence.test.ts`\\n* `src/__tests__/renderer/stores/coworkingBackgroundBrowserStore.test.ts`\\n* `src/__tests__/renderer/utils/browserTabPersistence.test.ts`\\n* `src/__tests__/renderer/utils/terminalTabHelpers.test.ts`\\n* `src/main/app-lifecycle/window-manager.ts`\\n* `src/main/coworking/coworking-audit.ts`\\n* `src/main/coworking/coworking-bridge.ts`\\n* `src/main/coworking/coworking-installer.ts`\\n* `src/main/coworking/coworking-registry.ts`\\n* `src/main/coworking/coworking-server-paths.ts`\\n* `src/main/coworking/coworking-server-script.ts`\\n* `src/main/coworking/coworking-session-id.ts`\\n* `src/main/coworking/coworking-tools.ts`\\n* `src/main/coworking/coworking-types.ts`\\n* `src/main/coworking/installers/claude-code.ts`\\n* `src/main/coworking/installers/codex.ts`\\n* `src/main/coworking/installers/factory-droid.ts`\\n* `src/main/coworking/installers/index.ts`\\n* `src/main/coworking/installers/opencode.ts`\\n* `src/main/coworking/installers/types.ts`\\n* `src/main/coworking/pid-resolution.ts`\\n* `src/main/index.ts`\\n* `src/main/ipc/handlers/browser-session.ts`\\n* `src/main/ipc/handlers/coworking.ts`\\n* `src/main/ipc/handlers/index.ts`\\n* `src/main/preload/browserSession.ts`\\n* `src/main/preload/coworking.ts`\\n* `src/main/preload/index.ts`\\n* `src/main/process-manager/ProcessManager.ts`\\n* `src/main/stores/defaults.ts`\\n* `src/renderer/components/MainPanel/BrowserTabView.tsx`\\n* `src/renderer/components/MainPanel/MainPanel.tsx`\\n* `src/renderer/components/MainPanel/MainPanelContent.tsx`\\n* `src/renderer/components/MainPanel/types.ts`\\n* `src/renderer/components/Settings/CoworkingSetup.tsx`\\n* `src/renderer/components/Settings/searchableSettings.ts`\\n* `src/renderer/components/Settings/tabs/EncoreTab/EncoreTab.tsx`\\n* `src/renderer/components/Settings/tabs/EncoreTab/components/CoworkingSection.tsx`\\n* `src/renderer/components/Settings/tabs/EncoreTab/components/index.ts`\\n* `src/renderer/components/TabBar/BrowserTabItem.tsx`\\n* `src/renderer/components/TabBar/NewTabPopover.tsx`\\n* `src/renderer/components/TabBar/TerminalTabItem.tsx`\\n* `src/renderer/components/TabBar/types.ts`\\n* `src/renderer/components/coworking/CoworkingApprovalHost.tsx`\\n* `src/renderer/components/coworking/CoworkingBackgroundBrowsers.tsx`\\n* `src/renderer/global.d.ts`\\n* `src/renderer/hooks/coworking/useCoworkingBrowserResponder.ts`\\n* `src/renderer/hooks/coworking/useCoworkingBufferResponder.ts`\\n* `src/renderer/hooks/coworking/useCoworkingRegistrySync.ts`\\n* `src/renderer/hooks/props/useMainPanelProps.ts`\\n* `src/renderer/hooks/session/useSessionRestoration.ts`\\n* `src/renderer/hooks/settings/useSettings.ts`\\n* `src/renderer/hooks/tabs/internal/browserTabHelpers.ts`\\n* `src/renderer/hooks/tabs/internal/types.ts`\\n* `src/renderer/hooks/tabs/internal/useBrowserTabHandlers.ts`\\n* `src/renderer/hooks/utils/useDebouncedPersistence.ts`\\n* `src/renderer/stores/coworkingApprovalStore.ts`\\n* `src/renderer/stores/coworkingBackgroundBrowserStore.ts`\\n* `src/renderer/stores/settingsStore.ts`\\n* `src/renderer/types/index.ts`\\n* `src/renderer/utils/browserTabPersistence.ts`\\n* `src/renderer/utils/terminalTabHelpers.ts`\\n* `src/shared/coworkingBrowser.ts`\\n* `src/shared/settingsMetadata.ts`\\n* `src/types/promise-with-resolvers.d.ts`\\n\\n

\\n\\n

\\n\\n\",\"path\":null,\"side\":null,\"line\":null}{\"type\":\"pull_request_comment\",\"body\":\"Updated: merged #948's latest hardening (`74e322fd9` harden browser tools / fix sync+eviction races, `ca5bb6028` confirm-contract tests) into this branch and reconciled hunk-by-hunk — this PR is no longer based on the older #948 snapshot.\\n\\n**Brought in from #948:** eval force-confirm regardless of policy + `type` in the dangerous set; navigate scheme allowlist (http/https/about:blank) — also applied to our `newTab`, closing a `file://` local-read exfiltration path; read caps (read_browser char cap, read_terminal line clamp); registry-sync run serialization; background-tab LRU eviction guard; atomic installer writes; 0600 on the bridge socket + audit JSONL; searchable-settings entries.\\n\\n**Reconciled (kept superior):** our audit URL redaction (origin+path + queryChars/hashChars + non-url char counts) with their 0600 sink; **union** confirm dangerous-set = navigate/eval/type/newTab/closeTab + their FORCE_CONFIRM_KINDS=[eval]; their sync serialization + our cache-restore-on-failure; our Settings redesign + their data-setting-id (confirm option relabeled 'Never'→'JS only' since eval always confirms).\\n\\n**Preserved (ours):** ephemeral tabs, clearSessionData, main-side hidden enforcement, confirm mirror + boundary recheck, wmic→PowerShell-CIM fallback.\\n\\n**Verified:** typecheck (lint+main) green; coworking/settings slice **545 tests green** (confirm contract + read caps both-polarity); MCP stdio smoke SMOKE-PASS (17 tools). Merge commit `af7dc7bfb`.\",\"path\":null,\"side\":null,\"line\":null}{\"type\":\"pull_request_comment\",\"body\":\"_🩺 Stability & Availability_ | _🟡 Minor_ | _⚡ Quick win_\\n\\n**`clear()` doesn't reset the module-level `inFlightKeys` guard.**\\n\\n`inFlightKeys` lives outside the Zustand state and is never cleared by `clear()`. If an op's `markOpEnd` never fires (component unmount mid-op, error path, or feature toggled off while an op is pending) the key persists forever, permanently exempting any future mount reusing that key (same `sessionId`+`tabUuid`) from LRU eviction and silently breaking the documented 1-10 cap.\\n\\n

\\n

🩹 Proposed fix

\\n\\n```diff\\n \\t\\tclear: () => set({ mounts: [], handles: new Map() }),\\n+\\t\\t// also clear the in-flight guard so a stale/abandoned op key doesn't\\n+\\t\\t// permanently block eviction after the feature is re-enabled.\\n```\\n\\n```diff\\n-\\t\\tclear: () => set({ mounts: [], handles: new Map() }),\\n+\\t\\tclear: () => {\\n+\\t\\t\\tinFlightKeys.clear();\\n+\\t\\t\\tset({ mounts: [], handles: new Map() });\\n+\\t\\t},\\n```\\n

\\n\\nAlso consider ensuring callers invoke `markOpStart`/`markOpEnd` via try/finally so an in-flight key can't outlive its op.\\n\\n\\n\\n\\n\\nAlso applies to: 82-87, 97-97\\n\\n

\\n

🤖 Prompt for AI Agents

\\n\\n```\\nVerify each finding against current code. Fix only still-valid issues, skip the\\nrest with a brief reason, keep changes minimal, and validate.\\n\\nIn `@src/renderer/stores/coworkingBackgroundBrowserStore.ts` around lines 48 - 51,\\nThe module-level in-flight guard in coworkingBackgroundBrowserStore is never\\nreset by clear(), so stale keys can permanently bypass eviction. Update the\\nstore’s clear/reset path to also empty inFlightKeys, and make sure the op\\nlifecycle around markOpStart and markOpEnd is always paired safely (prefer\\ntry/finally at the call sites) so a key cannot outlive its operation.\\n```\\n\\n

\\n\\n\\n\\n\\n\\n\\n\\n\",\"path\":\"src/renderer/stores/coworkingBackgroundBrowserStore.ts\",\"side\":\"RIGHT\",\"line\":51}{\"type\":\"pull_request_comment\",\"body\":\"**Actionable comments posted: 1**\\n\\n

\\n

🤖 Prompt for all review comments with AI agents

\\n\\n```\\nVerify each finding against current code. Fix only still-valid issues, skip the\\nrest with a brief reason, keep changes minimal, and validate.\\n\\nInline comments:\\nIn `@src/renderer/stores/coworkingBackgroundBrowserStore.ts`:\\n- Around line 48-51: The module-level in-flight guard in\\ncoworkingBackgroundBrowserStore is never reset by clear(), so stale keys can\\npermanently bypass eviction. Update the store’s clear/reset path to also empty\\ninFlightKeys, and make sure the op lifecycle around markOpStart and markOpEnd is\\nalways paired safely (prefer try/finally at the call sites) so a key cannot\\noutlive its operation.\\n```\\n\\n

\\n\\n

\\n

🪄 Autofix (Beta)

\\n\\nFix all unresolved CodeRabbit comments on this PR:\\n\\n- [ ] Push a commit to this branch (recommended)\\n- [ ] Create a new PR with the fixes\\n\\n

\\n\\n---\\n\\n

\\n

ℹ️ Review info

\\n\\n

\\n

⚙️ Run configuration

\\n\\n**Configuration used**: Path: .coderabbit.yaml\\n\\n**Review profile**: CHILL\\n\\n**Plan**: Pro\\n\\n**Run ID**: `8c583fa8-3211-418d-871f-61f0288e3dec`\\n\\n

\\n\\n

\\n

📥 Commits

\\n\\nReviewing files that changed from the base of the PR and between 98eae338f23ee2b05ae94a13aeda78d397f83ce9 and af7dc7bfbb692bfa95efa6a43c237bcad3e9e0d0.\\n\\n

\\n\\n

\\n

📒 Files selected for processing (28)

\\n\\n* `src/__tests__/main/coworking/coworking-browser.test.ts`\\n* `src/__tests__/main/coworking/coworking-tools.test.ts`\\n* `src/__tests__/renderer/components/Settings/CoworkingSetup.test.tsx`\\n* `src/__tests__/renderer/hooks/coworking/coworkingApproval.test.ts`\\n* `src/main/coworking/coworking-audit.ts`\\n* `src/main/coworking/coworking-bridge.ts`\\n* `src/main/coworking/coworking-installer.ts`\\n* `src/main/coworking/coworking-registry.ts`\\n* `src/main/coworking/coworking-server-paths.ts`\\n* `src/main/coworking/coworking-server-script.ts`\\n* `src/main/coworking/coworking-session-id.ts`\\n* `src/main/coworking/coworking-tools.ts`\\n* `src/main/coworking/installers/claude-code.ts`\\n* `src/main/coworking/installers/codex.ts`\\n* `src/main/coworking/installers/factory-droid.ts`\\n* `src/main/coworking/installers/opencode.ts`\\n* `src/main/coworking/pid-resolution.ts`\\n* `src/main/ipc/handlers/coworking.ts`\\n* `src/main/preload/coworking.ts`\\n* `src/main/utils/atomic-json-store.ts`\\n* `src/renderer/components/Settings/CoworkingSetup.tsx`\\n* `src/renderer/components/Settings/searchableSettings.ts`\\n* `src/renderer/global.d.ts`\\n* `src/renderer/hooks/coworking/useCoworkingBrowserResponder.ts`\\n* `src/renderer/hooks/coworking/useCoworkingBufferResponder.ts`\\n* `src/renderer/hooks/coworking/useCoworkingRegistrySync.ts`\\n* `src/renderer/stores/coworkingBackgroundBrowserStore.ts`\\n* `src/shared/coworkingBrowser.ts`\\n\\n

\\n\\n

\\n

💤 Files with no reviewable changes (3)

\\n\\n* src/renderer/global.d.ts\\n* src/main/preload/coworking.ts\\n* src/main/ipc/handlers/coworking.ts\\n\\n

\\n\\n

\\n

🚧 Files skipped from review as they are similar to previous changes (21)

\\n\\n* src/main/coworking/coworking-session-id.ts\\n* src/main/coworking/installers/opencode.ts\\n* src/renderer/hooks/coworking/useCoworkingBufferResponder.ts\\n* src/main/coworking/installers/claude-code.ts\\n* src/main/coworking/coworking-audit.ts\\n* src/main/coworking/installers/codex.ts\\n* src/renderer/hooks/coworking/useCoworkingRegistrySync.ts\\n* src/main/coworking/coworking-installer.ts\\n* src/main/coworking/coworking-server-paths.ts\\n* src/__tests__/renderer/hooks/coworking/coworkingApproval.test.ts\\n* src/__tests__/main/coworking/coworking-tools.test.ts\\n* src/main/coworking/coworking-tools.ts\\n* src/__tests__/renderer/components/Settings/CoworkingSetup.test.tsx\\n* src/main/coworking/installers/factory-droid.ts\\n* src/__tests__/main/coworking/coworking-browser.test.ts\\n* src/renderer/components/Settings/CoworkingSetup.tsx\\n* src/main/coworking/coworking-server-script.ts\\n* src/renderer/hooks/coworking/useCoworkingBrowserResponder.ts\\n* src/main/coworking/coworking-bridge.ts\\n* src/main/coworking/pid-resolution.ts\\n* src/main/coworking/coworking-registry.ts\\n\\n

\\n\\n

\\n\\n\",\"path\":null,\"side\":null,\"line\":null}{\"type\":\"pull_request_comment\",\"body\":\"**_{_{![P2 Badge](https://img.shields.io/badge/P2-yellow?style=flat)}} Expire approvals before running browser actions**\\n\\nWhen an approval dialog stays open past the main process browser-op timeout, this await can still resume later and run the action after the main listener has already been removed and the agent received a timeout. For example, a delayed Allow on `browser_eval` or `browser_navigate` can execute minutes after the original tool call was abandoned; race the approval against the same timeout or cancel pending approvals when the response channel times out.\\n\\nUseful? React with 👍 / 👎.\",\"path\":\"src/renderer/hooks/coworking/useCoworkingBrowserResponder.ts\",\"side\":\"RIGHT\",\"line\":365}{\"type\":\"pull_request_comment\",\"body\":\"**_{_{![P2 Badge](https://img.shields.io/badge/P2-yellow?style=flat)}} Normalize browser targets before scheme checks**\\n\\nWhen an agent supplies a scheme-less host or search text as the tool descriptions allow, this helper rejects it before the renderer's normal URL resolver can run because `new URL('example.com')` throws. That makes calls such as `browser_navigate({url: 'example.com'})` or `browser_new_tab({url: 'openai docs'})` fail with `-32602` instead of loading the resolved HTTPS/search URL; resolve first, then apply the http/https/about allowlist.\\n\\nUseful? React with 👍 / 👎.\",\"path\":\"src/main/coworking/coworking-bridge.ts\",\"side\":\"RIGHT\",\"line\":null}{\"type\":\"pull_request_comment\",\"body\":\"**_{_{![P2 Badge](https://img.shields.io/badge/P2-yellow?style=flat)}} Clear background browsers when Coworking is disabled**\\n\\nWhen the Coworking Encore feature is turned off while the separate background-browsing toggle remains enabled, this component still treats the host as enabled and never calls `clear()`. Any off-screen webviews mounted by previous agent reads keep running even though the user disabled Coworking, continuing network/page side effects and holding renderer processes. Include the Encore flag in this enabled condition or clear mounts on feature disable.\\n\\nUseful? React with 👍 / 👎.\",\"path\":\"src/renderer/components/coworking/CoworkingBackgroundBrowsers.tsx\",\"side\":\"RIGHT\",\"line\":null}{\"type\":\"pull_request_comment\",\"body\":\"**_{_{![P2 Badge](https://img.shields.io/badge/P2-yellow?style=flat)}} Return the new browser id only after registry sync**\\n\\nWhen an agent follows this success message and calls `list_browsers` immediately after `browser_new_tab`, the new tab can be absent because `setSessions` only updates the renderer store and the main registry is updated later by `useCoworkingRegistrySync`'s React effect. In that common tool sequence, the agent is told to list for an id but races the sync and cannot address the tab; either return the id directly or wait/push the registry update before resolving.\\n\\nUseful? React with 👍 / 👎.\",\"path\":\"src/renderer/hooks/coworking/useCoworkingBrowserResponder.ts\",\"side\":\"RIGHT\",\"line\":177}{\"type\":\"pull_request_comment\",\"body\":\"**_{_{![P2 Badge](https://img.shields.io/badge/P2-yellow?style=flat)}} Fail reads when the terminal ref is missing**\\n\\nWhen main requests a terminal that is listed in the registry but whose `TerminalView` has not been mounted in this renderer lifetime, this silently returns `''`. `readTerminal()` then reports a successful empty buffer, so agents get false results for restored or background terminal tabs instead of a mount or a clear error; respond with a failure when the ref is missing or ensure those sessions are mounted before registry exposure.\\n\\nUseful? React with 👍 / 👎.\",\"path\":\"src/renderer/hooks/coworking/useCoworkingBufferResponder.ts\",\"side\":\"RIGHT\",\"line\":null}{\"type\":\"pull_request_comment\",\"body\":\"**_{_{![P2 Badge](https://img.shields.io/badge/P2-yellow?style=flat)}} Assign coworking ids to restored terminal tabs**\\n\\nWhen a user restores a closed terminal tab, `reopenUnifiedClosedTab` builds a `createTerminalTab()` result and appends it directly, so it never passes through this new `coworkingId` assignment. The restored tab loses its `term:N` pill and `buildRecords()` filters it out of `list_terminals`; make the restore path mint and bump the same counter instead of leaving the id undefined.\\n\\nUseful? React with 👍 / 👎.\",\"path\":\"src/renderer/utils/terminalTabHelpers.ts\",\"side\":\"RIGHT\",\"line\":null}{\"type\":\"pull_request_comment\",\"body\":\"\\n### 💡 Codex Review\\n\\nHere are some automated review suggestions for this pull request.\\n\\n**Reviewed commit:** `af7dc7bfbb`\\n \\n\\n

ℹ️ About Codex in GitHub

\",\"path\":null,\"side\":null,\"line\":null}","text_elements",[],{"_390":672,"_371":372,"_392":378},{"_393":673,"_395":-5,"_42":-5,"_396":-5,"_397":-5,"_96":299,"_92":399},[674],{"_46":668,"_92":675},"input_text",{"_390":677,"_371":372,"_392":378},{"_393":678,"_395":-5,"_42":-5,"_396":-5,"_397":-5,"_96":299,"_92":399},[679,680],{"_46":682,"_92":675},{"_46":681,"_92":675},"\n /workspace/Maestro\n bash\n 2026-07-02\n Etc/UTC\n","# AGENTS.md instructions for /workspace/Maestro\n\n\n# CLAUDE.md\n\nEssential guidance for working with this codebase. For detailed architecture, see [ARCHITECTURE.md](ARCHITECTURE.md). For development setup and processes, see [CONTRIBUTING.md](CONTRIBUTING.md).\n\n## Documentation Index\n\nThis guide has been split into focused sub-documents for progressive disclosure:\n\n| Document | Description |\n| ------------------------------------ | ------------------------------------------------------------------------------------------------------------ |\n| [[CLAUDE-PATTERNS.md]] | Core implementation patterns (process management, settings, modals, themes, Auto Run, SSH, Encore Features) |\n| [[CLAUDE-IPC.md]] | IPC API surface (`window.maestro.*` namespaces) |\n| [[CLAUDE-PERFORMANCE.md]] | Performance best practices (React optimization, debouncing, batching) |\n| [[CLAUDE-WIZARD.md]] | Onboarding Wizard, Inline Wizard, and Tour System |\n| [[CLAUDE-FEATURES.md]] | Usage Dashboard and Document Graph features |\n| [[CLAUDE-AGENTS.md]] | Supported agents and capabilities |\n| [[CLAUDE-SESSION.md]] | Session interface (agent data model) and code conventions |\n| [[CLAUDE-PLATFORM.md]] | Cross-platform concerns (Windows, Linux, macOS, SSH remote) |\n| [[CLAUDE-CUE.md]] | Cue automation engine: architecture, dispatch flow, lifecycle, gotchas (read before editing `src/main/cue/`) |\n| [AGENT_SUPPORT.md](AGENT_SUPPORT.md) | Detailed agent integration guide |\n\n---\n\n## Before Writing New Code - Check Existing Utilities\n\n**MANDATORY:** Before creating any new utility function, helper, hook, component, type, or constant, check the guide docs in `docs/agent-guides/` to see if it already exists. Duplicated code is the #1 source of maintenance burden in this codebase - there are already grep-verified instances of 20+ duplicate format helpers, 60+ ad-hoc mock factories, and 500+ manual modal-layer registrations. Don't add to the pile.\n\n| Before creating... | Check this guide first |\n| -------------------------------------------------- | -------------------------------------------------------------- |\n| Utility function (formatting, IDs, paths, strings) | [SHARED-UTILS.md](docs/agent-guides/SHARED-UTILS.md) |\n| IPC handler or preload bridge | [IPC-PATTERNS.md](docs/agent-guides/IPC-PATTERNS.md) |\n| Store action, selector, or hook | [STATE-PATTERNS.md](docs/agent-guides/STATE-PATTERNS.md) |\n| Agent parser, storage, or error pattern | [AGENT-INFRA.md](docs/agent-guides/AGENT-INFRA.md) |\n| UI component, modal, or theme usage | [UI-PATTERNS.md](docs/agent-guides/UI-PATTERNS.md) |\n| Stat card, chart, sparkline, or input control | [WIDGET-LIBRARY.md](docs/agent-guides/WIDGET-LIBRARY.md) |\n| Test mock, factory, or setup pattern | [TEST-PATTERNS.md](docs/agent-guides/TEST-PATTERNS.md) |\n| Renderer service or constant | [RENDERER-SERVICES.md](docs/agent-guides/RENDERER-SERVICES.md) |\n| Process spawning or listener | [PROCESS-SYSTEM.md](docs/agent-guides/PROCESS-SYSTEM.md) |\n| Web/mobile hook or component | [WEB-MOBILE.md](docs/agent-guides/WEB-MOBILE.md) |\n| CLI command or playbook feature | [CLI-PLAYBOOKS.md](docs/agent-guides/CLI-PLAYBOOKS.md) |\n| Group chat or Symphony feature | [GROUP-CHAT.md](docs/agent-guides/GROUP-CHAT.md) |\n| Stats, analytics, or dashboard | [STATS-ANALYTICS.md](docs/agent-guides/STATS-ANALYTICS.md) |\n| Prompt template or SpecKit/OpenSpec | [PROMPTS-SPECS.md](docs/agent-guides/PROMPTS-SPECS.md) |\n| Cue pipeline feature | [CUE-PIPELINE.md](docs/agent-guides/CUE-PIPELINE.md) |\n| App lifecycle, updater, or power mgmt | [MAIN-LIFECYCLE.md](docs/agent-guides/MAIN-LIFECYCLE.md) |\n\n### Commonly-reimplemented functions (do NOT add new copies)\n\nGrep-verified 2026-04-10. Import from these canonical locations:\n\n- **ID generation:** `generateId()` in `src/renderer/utils/ids.ts`, `generateUUID()` in `src/shared/uuid.ts`\n- **Format file size:** `formatSize()` in `src/shared/formatters.ts`\n- **Format numbers:** `formatNumber()` in `src/shared/formatters.ts`\n- **Format tokens:** `formatTokens()`, `formatTokensCompact()`, `estimateTokenCount()` in `src/shared/formatters.ts`\n- **Format elapsed time:** `formatElapsedTime()`, `formatElapsedTimeColon()` in `src/shared/formatters.ts`\n- **Format duration (ms):** `formatDuration()` in `src/shared/performance-metrics.ts` (NOT formatters.ts - common mistake)\n- **Format relative time:** `formatRelativeTime()` in `src/shared/formatters.ts`\n- **Format cost:** `formatCost()` in `src/shared/formatters.ts`\n- **Path utilities:** `truncatePath()`, `getParentDir()`, `truncateCommand()`, `isAbsolutePath()`, `getBasename()` in `src/shared/formatters.ts`\n- **Strip ANSI:** `stripAnsiCodes()` in `src/shared/stringUtils.ts`\n- **Shell escape:** `shellEscape()`, `shellEscapeArgs()`, `shellEscapeForDoubleQuotes()` in `src/main/utils/shell-escape.ts`\n- **Platform detection:** `isWindows()`, `isMacOS()`, `isLinux()` in `src/shared/platformDetection.ts`\n- **Agent display name:** `getAgentDisplayName()` in `src/shared/agentMetadata.ts`\n- **SSH remote lookup:** `getSshRemoteById()` in `src/main/stores/getters.ts`\n- **Toast notifications:** `notifyToast({ color, title, message, dismissible? })` in `src/renderer/stores/notificationStore.ts`. Use for async results, errors, and persistent/dismissable messages. Same five-color design language as Center Flash: `green | yellow | orange | red | theme` (default `theme`). Set `dismissible: true` (or pass `--dismissible` from `maestro-cli notify toast`) when the user MUST acknowledge - disables auto-dismiss, requires click to close, and emphasizes the X button. Cannot combine `dismissible` with `duration`/`--timeout`. External CLI cap: 60 seconds (use `--dismissible` for sticky). **Click actions** (data-driven, survive the IPC bridge): pass `clickAction: { kind: 'jump-session', sessionId, tabId? } | { kind: 'open-file', sessionId, path } | { kind: 'open-url', url }` for what should happen when the toast body is clicked, or use the legacy `sessionId`/`tabId` fields for plain agent jump. From the CLI: `--agent` (+ optional `--tab`), `--open-file ` (requires `--agent`), `--open-url ` (mutually exclusive with `--open-file`). `--action-url` / `--action-label` render an inline link button beneath the message and are independent of the body click. **Source-agent label:** pass `sourceAgent` (CLI `--source-agent `) to stamp which agent/pipeline fired the toast in the header strip. It's store-independent, so it shows even when the agent isn't loaded in the Left Bar (the name resolved from `--agent`/`sessionId` only renders when that agent is in the desktop store) - use it for cron/watchdog toasts. The explicit label wins over the resolved name for display; pair it with `--agent` to also get click-to-jump. Do NOT pass renderer-only callbacks across the bridge - use `clickAction` instead.\n- **Center flash (rapid acks):** `notifyCenterFlash({ message, color, detail?, duration? })` in `src/renderer/stores/centerFlashStore.ts`; clipboard helper `flashCopiedToClipboard()` in `src/renderer/utils/flashCopiedToClipboard.ts`. Use for momentary \"I did the thing\" confirmations of user-initiated actions. Five-color design language: `green | yellow | orange | red | theme` - default `theme` matches the active Maestro theme. External integrations can fire flashes via `maestro-cli notify flash --color `. Do NOT roll your own center-screen overlay, useState+setTimeout flash, add a sixth color, or use a Toast for clipboard acks. Single visible flash at a time, themed frosted-glass card mounted once in `App.tsx`. Full decision rules, color palette, and design language: [UI-PATTERNS.md → Center Flash System](docs/agent-guides/UI-PATTERNS.md#center-flash-system-rapid-temporary-notifications).\n- **Session lookup:** `selectActiveSession()`, `selectSessionById()` in `src/renderer/stores/sessionStore.ts`; `useActiveSession()` hook in `src/renderer/hooks/session/useActiveSession.ts`\n- **Session mutation:** `updateSessionWith(sessionId, updater)` in `src/renderer/stores/sessionStore.ts` (do NOT hand-roll `setSessions(prev => prev.map(...))`)\n- **Focus an AI tab:** `aiTabFocusFields(tabId?)` in `src/renderer/utils/tabHelpers.ts` - spread into a session patch (`{ ...s, ...aiTabFocusFields(tabId) }`) to land on an AI tab. It clears `activeFileTabId`/`activeTerminalTabId`/`activeBrowserTabId` and sets `inputMode: 'ai'`. Do NOT hand-roll the literal: those non-AI tab types outrank the AI tab in the render precedence, so omitting even one leaves the previous view on screen.\n- **Modal layer:** `useModalLayer()` in `src/renderer/hooks/ui/useModalLayer.ts` (do NOT use manual `registerLayer()` boilerplate)\n- **Focus after render:** `useFocusAfterRender()` in `src/renderer/hooks/utils/useFocusAfterRender.ts` (do NOT use `useEffect + setTimeout(() => ref.focus())`)\n- **Event listeners:** `useEventListener()` in `src/renderer/hooks/utils/useEventListener.ts` (do NOT pair raw `addEventListener`/`removeEventListener` inside useEffect)\n- **Debounce/throttle:** `useDebouncedValue()`, `useDebouncedCallback()`, `useThrottledCallback()` in `src/renderer/hooks/utils/useThrottle.ts` (filename is misleading - all three live here)\n- **Render markdown:** `` from `src/renderer/components/Markdown/` (do NOT hand-roll `` + a per-surface `components`/plugin map). The chat preset is what `MarkdownRenderer` wraps. Shared internals: `buildMarkdownPlugins` (`Markdown/plugins.ts`), `preprocessMarkdown` (`Markdown/preprocess.ts`), leaf renderers in `Markdown/components/*`, and the document component map `createMarkdownComponents()` in `src/renderer/utils/markdownConfig.ts`. See [UI-PATTERNS.md → ``](docs/agent-guides/UI-PATTERNS.md).\n\nIf your use case does NOT match an existing utility, prefer extending the canonical file over creating a new one. If you genuinely need something new, add it to the relevant guide in `docs/agent-guides/` so the next person can find it.\n\nThe tracker at [DEDUP-TRACKER.md](docs/agent-guides/DEDUP-TRACKER.md) lists all known duplication findings.\n\n---\n\n## Agent Behavioral Guidelines\n\nCore behaviors for effective collaboration. Failures here cause the most rework.\n\n### Surface Assumptions Early\n\nBefore implementing non-trivial work, explicitly state assumptions. Never silently fill in ambiguous requirements - the most common failure mode is guessing wrong and running with it. Format: \"Assumptions: 1) X, 2) Y. Correct me now or I proceed.\"\n\n### Manage Confusion Actively\n\nWhen encountering inconsistencies, conflicting requirements, or unclear specs: **STOP**. Name the specific confusion, present the tradeoff, and wait for resolution. Bad: silently picking one interpretation. Good: \"I see X in file A but Y in file B - which takes precedence?\"\n\n### Push Back When Warranted\n\nNot a yes-machine. When an approach has clear problems: point out the issue directly, explain the concrete downside, propose an alternative, then accept the decision if overridden. Sycophancy (\"Of course!\") followed by implementing a bad idea helps no one.\n\n### Enforce Simplicity\n\nNatural tendency is to overcomplicate - actively resist. Before finishing: Can this be fewer lines? Are abstractions earning their complexity? Would a senior dev say \"why didn't you just...\"? Prefer the boring, obvious solution.\n\n### Maintain Scope Discipline\n\nTouch only what's asked. Do NOT: remove comments you don't understand, \"clean up\" orthogonal code, refactor adjacent systems as side effects, or delete seemingly-unused code without approval. Surgical precision, not unsolicited renovation.\n\n### Dead Code Hygiene\n\nAfter refactoring: identify now-unreachable code, list it explicitly, ask \"Should I remove these now-unused elements: [list]?\" Don't leave corpses. Don't delete without asking.\n\n### Validate Before Push\n\nBefore pushing any branch, re-run the relevant formatting, lint, type-check, and test commands for the changes you made. Fix any issues those commands surface, include the fixes in the branch, and only then push or update the PR.\n\n---\n\n## Standardized Vernacular\n\nUse these terms consistently in code, comments, and documentation:\n\n### Terminology: Agent vs Session\n\nIn Maestro, the terms \"agent\" and \"session\" have distinct meanings:\n\n- **Agent** - An entity in the Left Bar backed by a provider (Claude Code, Codex, etc.). This is what users see, create, and interact with. Each agent has its own workspace, tabs, and configuration.\n- **Session** (or **provider session**) - An individual conversation context within a provider (e.g., Claude's `session_id`). Each AI tab within an agent can have its own provider session. In code, the `Session` interface represents an agent (historical naming).\n\nUse \"agent\" in user-facing language. Reserve \"session\" for provider-level conversation contexts or when documenting the code interface.\n\n### UI Components\n\n- **Left Bar** - Left sidebar with agent list and groups (`SessionList.tsx`)\n- **Right Bar** - Right sidebar with Files, History, Auto Run tabs (`RightPanel.tsx`)\n- **Main Window** - Center workspace (`MainPanel.tsx`)\n - **AI Terminal** - Main window in AI mode (interacting with AI agents)\n - **Command Terminal** - Main window in terminal/shell mode\n - **System Log Viewer** - Special view for system logs (`LogViewer.tsx`)\n\n### Automation\n\n- **Cue** - Event-driven automation system (Maestro Cue), gated as an Encore Feature. Watches for file changes, time intervals, agent completions, GitHub PRs/issues, and pending markdown tasks to trigger automated prompts. Configured via `.maestro/cue.yaml` per project.\n- **Cue Modal** - Dashboard for managing Cue subscriptions and viewing activity (`CueModal.tsx`)\n\n### Agent States (color-coded)\n\n- **Green** - Ready/idle\n- **Yellow** - Agent thinking/busy\n- **Red** - No connection/error\n- **Pulsing Orange** - Connecting\n\n---\n\n## Code Style\n\nThis codebase uses **tabs for indentation**, not spaces. Always match existing file indentation when editing.\n\n### Writing Style: No Em-Dashes or En-Dashes\n\n**NEVER use em-dashes (`—`, U+2014) or en-dashes (`–`, U+2013) anywhere.** This applies to everything you write: user docs (`docs/`), in-app documentation, system prompts (`src/prompts/`), UI copy, code comments, commit messages, PR descriptions, and your own responses. Em-dashes are a tell-tale sign of bot-authored text; humans almost never type them. Use one of these instead, whichever fits the sentence:\n\n- A spaced hyphen (`-`) for an aside or appositive.\n- A comma, colon, or parentheses to set off a clause.\n- Two separate sentences when the clauses stand on their own.\n- A plain hyphen (`-`) for numeric ranges (e.g. `10-20`, not `10–20`).\n\nThis is non-negotiable. If you catch an em-dash or en-dash in anything you produce or edit, replace it.\n\n---\n\n## Do Not Edit: `docs/releases.md`\n\n`docs/releases.md` is generated/updated automatically during release pressing. **Never modify it manually** - even when shipping user-facing changes that would seem to warrant a release note entry. The release tooling handles it.\n\n---\n\n## Project Overview\n\nMaestro is an Electron desktop app for managing multiple AI coding assistants simultaneously with a keyboard-first interface.\n\n### Supported Agents\n\n| ID | Name | Status |\n| --------------- | ------------- | ---------- |\n| `claude-code` | Claude Code | **Active** |\n| `codex` | OpenAI Codex | **Active** |\n| `opencode` | OpenCode | **Active** |\n| `factory-droid` | Factory Droid | **Active** |\n| `copilot-cli` | Copilot-CLI | **Beta** |\n| `terminal` | Terminal | Internal |\n\nSee [[CLAUDE-AGENTS.md]] for capabilities and integration details.\n\n---\n\n## Quick Commands\n\n```bash\nnpm run dev # Development with hot reload (isolated data, can run alongside production)\nnpm run dev:prod-data # Development using production data (close production app first)\nnpm run dev:web # Web interface development\nnpm run build # Full production build\nnpm run clean # Clean build artifacts\nnpm run lint # TypeScript type checking (all configs)\nnpm run lint:eslint # ESLint code quality checks\nnpm run package # Package for all platforms\nnpm run test # Run test suite\nnpm run test:watch # Run tests in watch mode\n```\n\n---\n\n## Architecture at a Glance\n\n```\nsrc/\n├── main/ # Electron main process (Node.js)\n│ ├── index.ts # Entry point, IPC handlers\n│ ├── preload.ts # Secure IPC bridge\n│ ├── process-manager.ts # Process spawning (PTY + child_process)\n│ ├── agent-*.ts # Agent detection, capabilities, session storage\n│ ├── cue/ # Maestro Cue event-driven automation engine\n│ ├── parsers/ # Per-agent output parsers + error patterns\n│ ├── storage/ # Per-agent session storage implementations\n│ ├── ipc/handlers/ # IPC handler modules (stats, git, playbooks, cue, etc.)\n│ └── utils/ # Utilities (execFile, ssh-spawn-wrapper, etc.)\n│\n├── renderer/ # React frontend (desktop)\n│ ├── App.tsx # Main coordinator\n│ ├── components/ # UI components\n│ ├── hooks/ # Custom React hooks\n│ ├── services/ # IPC wrappers (git.ts, process.ts)\n│ ├── constants/ # Themes, shortcuts, priorities\n│ └── contexts/ # Context providers (LayerStack, etc.)\n│\n├── web/ # Web/mobile interface\n│ ├── mobile/ # Mobile-optimized React app\n│ └── components/ # Shared web components\n│\n├── cli/ # CLI tooling for batch automation\n│ ├── commands/ # CLI command implementations\n│ └── services/ # Playbook and batch processing\n│\n├── prompts/ # System prompts (editable .md files)\n│\n├── shared/ # Shared types and utilities\n│\n└── docs/ # Mintlify documentation (docs.runmaestro.ai)\n```\n\n---\n\n## Key Files for Common Tasks\n\n| Task | Primary Files |\n| ------------------------------ | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |\n| Add IPC handler | `src/main/index.ts`, `src/main/preload.ts` |\n| Add UI component | `src/renderer/components/` |\n| Add web/mobile component | `src/web/components/`, `src/web/mobile/` |\n| Add keyboard shortcut | `src/renderer/constants/shortcuts.ts`, `App.tsx` |\n| Add theme | `src/renderer/constants/themes.ts` |\n| Add modal | Component + `src/renderer/constants/modalPriorities.ts` |\n| Add tab overlay menu | See Tab Hover Overlay Menu pattern in [[CLAUDE-PATTERNS.md]] |\n| Add setting | `src/shared/settingsMetadata.ts` (metadata), `src/renderer/stores/settingsStore.ts`, `src/main/stores/defaults.ts`, AND `src/renderer/components/Settings/searchableSettings.ts` + `data-setting-id` wrapper on rendered control (see [[CLAUDE-PATTERNS.md]] §3) |\n| Add template variable | `src/shared/templateVariables.ts`, `src/renderer/utils/templateVariables.ts` |\n| Modify system prompts | `src/prompts/*.md` (wizard, Auto Run, etc.) or edit via **Maestro Prompts** tab in Settings |\n| Customize prompts | Use **Maestro Prompts** tab in Settings, or edit `userData/core-prompts-customizations.json` |\n| Add new prompt | `src/prompts/*.md`, `src/shared/promptDefinitions.ts` (add to `CORE_PROMPTS` array and `PROMPT_IDS`) |\n| Add Spec-Kit command | `src/prompts/speckit/`, `src/main/speckit-manager.ts` |\n| Add OpenSpec command | `src/prompts/openspec/`, `src/main/openspec-manager.ts` |\n| Add CLI command | `src/cli/commands/`, `src/cli/index.ts` |\n| Add new agent | `src/shared/agentIds.ts`, `src/main/agents/definitions.ts`, `src/main/agents/capabilities.ts`, `src/shared/agentMetadata.ts` - see [AGENT_SUPPORT.md](AGENT_SUPPORT.md) |\n| Add agent output parser | `src/main/parsers/`, `src/main/parsers/index.ts` |\n| Add agent session storage | `src/main/storage/` (extend `BaseSessionStorage`), `src/main/storage/index.ts` |\n| Add agent error patterns | `src/main/parsers/error-patterns.ts` |\n| Add agent context window | `src/shared/agentConstants.ts` (`DEFAULT_CONTEXT_WINDOWS`, `FALLBACK_CONTEXT_WINDOW`) |\n| Add playbook feature | `src/cli/services/playbooks.ts` |\n| Add marketplace playbook | `src/main/ipc/handlers/marketplace.ts` (import from GitHub) |\n| Playbook import/export | `src/main/ipc/handlers/playbooks.ts` (ZIP handling with assets) |\n| Modify wizard flow | `src/renderer/components/Wizard/` (see [[CLAUDE-WIZARD.md]]) |\n| Add tour step | `src/renderer/components/Wizard/tour/tourSteps.ts` |\n| Modify file linking | `src/renderer/utils/remarkFileLinks.ts` (remark plugin for `[[wiki]]` and path links) |\n| Add documentation page | `docs/*.md`, `docs/docs.json` (navigation) |\n| Add documentation screenshot | `docs/screenshots/` (PNG, kebab-case naming) |\n| MCP server integration | See [MCP Server docs](https://docs.runmaestro.ai/mcp-server) |\n| Add stats/analytics feature | `src/main/stats-db.ts`, `src/main/ipc/handlers/stats.ts` |\n| Add Usage Dashboard chart | `src/renderer/components/UsageDashboard/` |\n| Add Document Graph feature | `src/renderer/components/DocumentGraph/`, `src/main/ipc/handlers/documentGraph.ts` |\n| Add colorblind palette | `src/renderer/constants/colorblindPalettes.ts` |\n| Add performance metrics | `src/shared/performance-metrics.ts` |\n| Capture/analyze perf trace | `src/main/profiling/` (Chromium contentTracing capture), `scripts/analyze-perf-trace.mjs` (offline analysis), `CLAUDE-PERFORMANCE.md` -> Field Performance Traces |\n| Add power management | `src/main/power-manager.ts`, `src/main/ipc/handlers/system.ts` |\n| Spawn agent with SSH support | `src/main/utils/ssh-spawn-wrapper.ts` (required for SSH remote execution) |\n| Modify file preview tabs | `TabBar.tsx`, `FilePreview.tsx`, `MainPanel.tsx` (see ARCHITECTURE.md → File Preview Tab System) |\n| Add Director's Notes feature | `src/renderer/components/DirectorNotes/`, `src/main/ipc/handlers/director-notes.ts` |\n| Add Encore Feature | `src/renderer/types/index.ts` (flag), `useSettings.ts` (state), `SettingsModal.tsx` (toggle UI), gate in `App.tsx` + keyboard handler |\n| Modify history components | `src/renderer/components/History/` |\n| Modify history activity graph | `src/renderer/components/History/ActivityGraph.tsx`, `src/main/utils/history-bucket-cache.ts` (disk-cached aggregates), `src/main/utils/history-bucket-builder.ts` |\n| Modify Auto Run Thought Stream | `src/renderer/stores/thoughtStreamStore.ts` (in-memory capture + `groupThoughtsIntoBlocks`), `src/renderer/components/ThoughtStreamPanel.tsx` (panel), `src/renderer/hooks/agent/internal/useThoughtStreamCaptureListener.ts` (taps `process:thinking-chunk`) |\n| Add Cue event type | `src/main/cue/cue-types.ts`, `src/main/cue/cue-engin\n",{"_390":684,"_371":372,"_392":378},{"_393":685,"_395":-5,"_42":-5,"_396":-5,"_397":-5,"_96":686,"_92":399},[687,688],"developer",{"_46":690,"_92":675},{"_46":689,"_92":675},"\n## Skills\nA skill is a set of local instructions to follow that is stored in a `SKILL.md` file. Below is the list of skills that can be used. Each entry includes a name, description, and file path so you can open the source for full instructions when using a specific skill.\n### Available skills\n- imagegen: Generate or edit raster images when the task benefits from AI-created bitmap visuals such as photos, illustrations, textures, sprites, mockups, or transparent-background cutouts. Use when Codex should create a brand-new image, transform an existing image, or derive visual variants from references, and the output should be a bitmap asset rather than repo-native code or vector. Do not use when the task is better handled by editing existing SVG/vector/code-native assets, extending an established icon or logo system, or building the visual directly in HTML/CSS/canvas. (file: /opt/codex/skills/.system/imagegen/SKILL.md)\n- openai-docs: Use when the user asks how to build with OpenAI products or APIs and needs up-to-date official documentation with citations, help choosing the latest model for a use case, or model upgrade and prompt-upgrade guidance; prioritize OpenAI docs MCP tools, use bundled references only as helper context, and restrict any fallback browsing to official OpenAI domains. (file: /opt/codex/skills/.system/openai-docs/SKILL.md)\n- plugin-creator: Create and scaffold plugin directories for Codex with a required `.codex-plugin/plugin.json`, optional plugin folders/files, and baseline placeholders you can edit before publishing or testing. Use when Codex needs to create a new local plugin, add optional plugin structure, or generate or update repo-root `.agents/plugins/marketplace.json` entries for plugin ordering and availability metadata. (file: /opt/codex/skills/.system/plugin-creator/SKILL.md)\n- skill-creator: Guide for creating effective skills. This skill should be used when users want to create a new skill (or update an existing skill) that extends Codex's capabilities with specialized knowledge, workflows, or tool integrations. (file: /opt/codex/skills/.system/skill-creator/SKILL.md)\n- skill-installer: Install Codex skills into $CODEX_HOME/skills from a curated list or a GitHub repo path. Use when a user asks to list installable skills, install a curated skill, or install a skill from another repo (including private repos). (file: /opt/codex/skills/.system/skill-installer/SKILL.md)\n### How to use skills\n- Discovery: The list above is the skills available in this session (name + description + file path). Skill bodies live on disk at the listed paths.\n- Trigger rules: If the user names a skill (with `$SkillName` or plain text) OR the task clearly matches a skill's description shown above, you must use that skill for that turn. Multiple mentions mean use them all. Do not carry skills across turns unless re-mentioned.\n- Missing/blocked: If a named skill isn't in the list or the path can't be read, say so briefly and continue with the best fallback.\n- How to use a skill (progressive disclosure):\n 1) After deciding to use a skill, open its `SKILL.md`. Read only enough to follow the workflow.\n 2) When `SKILL.md` references relative paths (e.g., `scripts/foo.py`), resolve them relative to the skill directory listed above first, and only consider other paths if needed.\n 3) If `SKILL.md` points to extra folders such as `references/`, load only the specific files needed for the request; don't bulk-load everything.\n 4) If `scripts/` exist, prefer running or patching them instead of retyping large code blocks.\n 5) If `assets/` or templates exist, reuse them instead of recreating from scratch.\n- Coordination and sequencing:\n - If multiple skills apply, choose the minimal set that covers the request and state the order you'll use them.\n - Announce which skill(s) you're using and why (one short line). If you skip an obvious skill, say why.\n- Context hygiene:\n - Keep context small: summarize long sections instead of pasting them; only load extra files when needed.\n - Avoid deep reference-chasing: prefer opening only files directly linked from `SKILL.md` unless you're blocked.\n - When variants exist (frameworks, providers, domains), pick only the relevant reference file(s) and note that choice.\n- Safety and fallback: If a skill can't be applied cleanly (missing files, unclear instructions), state the issue, pick the next-best approach, and continue.\n","\nFilesystem sandboxing defines which files can be read or written. `sandbox_mode` is `danger-full-access`: No filesystem sandboxing - all commands are permitted. Network access is enabled.\nApproval policy is currently never. Do not provide the `sandbox_permissions` for any reason, commands will be rejected.\n","turn/started",{"_371":372,"_87":693},{"_103":-5,"_42":378,"_379":694,"_381":382,"_383":695},[],"inProgress",{"_383":697,"_371":372},{"_698":699,"_92":700},"activeFlags",[],"active","thread/started",{"_703":704},"thread",{"_705":-5,"_706":-5,"_707":708,"_709":382,"_422":423,"_710":15,"_711":-5,"_712":-5,"_42":372,"_713":714,"_181":-5,"_715":-5,"_161":716,"_717":199,"_718":-5,"_427":719,"_383":720,"_721":-5,"_77":722,"_723":382},"agentNickname","agentRole","cliVersion","0.128.0","createdAt","ephemeral","forkedFromId","gitInfo","modelProvider","openai","parentThreadId","/opt/codex/sessions/2026/07/02/rollout-2026-07-02T23-10-05-019f2518-b95a-7a40-aff0-0d8160a9ee76.jsonl","preview","recencyAt","vscode",{"_92":387},"threadSource",[],"updatedAt",[],"messages",[727],{"_42":728,"_729":730,"_731":732,"_733":732,"_393":734,"_383":735,"_395":-5,"_736":32,"_737":738,"_739":740,"_741":-5},"1467a6c1-2b76-467b-8e9b-7e5d67731aa2","author",{"_96":299,"_181":-5,"_737":750},"create_time",1783033804.37646,"update_time",{"_746":46,"_747":748},"finished_successfully","weight","metadata",{"_742":743},"recipient","all","channel","l1239dk1",{"_744":745},"inline_comments",[],"content_type","parts",[749],"## Review feedback addressed (commit `73fe0db10`)\n\nThanks @coderabbitai @chatgpt-codex-connector @greptile-apps. Went through every thread. Verified against the current (post-rc-merge) code, since a few comments targeted the pre-merge snapshot. Typecheck (lint+main) green, `prettier --check .` clean, coworking/settings slice green (27 files / 696 tests, +2 fixed, +2 new coverage).\n\n### Applied — correctness / security\n- **Windows pipe slug** now hashed from the full `userData` path (was app-folder basename; collided across OS users). [CodeRabbit Major]\n- **`hello` handshake** cross-checks an explicit `sessionId` against `ppid` resolution — a positive mismatch is rejected; a null (undeterminable) result still trusts the env id so env-propagating agents don't regress. [Codex P1]\n- **`navigate`/`newTab` scheme guard** rewritten: `about:blank`/`http`/`https` and bare-host/search text allowed (renderer resolves the latter to https), parseable `file:`/`javascript:`/`data:` rejected — keeps the exfiltration block **and** restores the documented \"URL or search query\" behavior. [Codex P2]\n- **`clearArmed` two-step confirm resets on tab change** — armed state can no longer carry to another tab and wipe its session on one click. [CodeRabbit Major, Codex P2]\n- **Coworking approval + background hosts hoisted** above MainPanel's early returns — approvals no longer hang when the log/agent-sessions/memory views are open. [CodeRabbit Major]\n- **`createTabForSession`** returns `ok:false` when the session closed during approval (was reporting success). [CodeRabbit Major]\n- **Mount-for-read restores the full prior surface** (file/terminal/browser/inputMode), not just a browser id, so a read never strands the user on a mounted tab. [Codex P2]\n- **`newTab`** clears competing active file/terminal ids + sets `inputMode:'ai'`. [Codex P2]\n- **`navigate` clears `isDomReadyRef`** so a follow-up read/waitFor waits for the new page. [Codex P2]\n- **Buffer responder** signals a clean \"terminal not live\" instead of a false empty read for unmounted refs. [Codex P2]\n- **Background browsers** clear when the Coworking Encore flag is off, and skip the active session (no duplicate webview); store `clear()` resets the in-flight guard. [Codex P2, CodeRabbit Minor]\n- **Partition allowlist unified** into `src/shared/browserTabPartition.ts` (full regex); `window-manager` now uses it (was weaker `startsWith`), matching the `clearSessionData` gate. [CodeRabbit + Greptile P2]\n- **Restored terminal tabs** mint a `coworkingId` (keep their `term:N` pill / stay in `list_terminals`). [Codex P2]\n\n### Applied — performance / hygiene\n- **`pid-resolution` async** (`execFile`, no longer blocks the main event loop during handshake; backend cache preserved). [Greptile P1]\n- **Audit sink async** (`fs.promises.appendFile`, fire-and-forget, `0600`, 5 MiB soft cap; was blocking `appendFileSync`). [Greptile P1, CodeRabbit]\n- **`listTerminals`** now audited via `auditedBrowserCall` (real outcome). [CodeRabbit Minor]\n- **`maxChars`** rejected above 2,000,000. [CodeRabbit Minor]\n- Docstring for `dangerous` policy lists all gated ops; `CoworkingSetup` uses `DEFAULT_BROWSER_CONFIRM_POLICY`; `global.d.ts` imports `BrowserOp` instead of hand-duplicating it; **em/en-dashes removed** from all lines this PR added (CLAUDE.md rule). [CodeRabbit nits + Codex]\n- **`registry-sync`** rollback guarded by a monotonic run id. [CodeRabbit Major — note: run serialization from the rc merge already largely covered this; the guard is belt-and-suspenders]\n\n### Already fixed by the rc merge (comments predate it)\n- **Atomic installer writes** — installers use `atomicWriteFile` (temp+rename) from the merged `atomic-json-store`.\n- **Registry-sync serialization** — `runQueueRef` present.\n\n### Declined (with reasons)\n- **SSH `MAESTRO_COWORKING_SESSION_ID` injection before SSH wrapping** [Codex P2] — SSH-mode coworking is explicitly out of scope / deferred in this feature; the remote MCP subprocess topology needs its own design + testing.\n- **Bundled-node reliability (`command:\"node\"`)** [Codex P2] — real for node-less Codex/Factory hosts, but the robust fix (ship Electron-as-node) needs testing on such machines; tracked as follow-up rather than a blind change here.\n- **Approval-vs-main-timeout race** [Codex P2] — bounded by the 300s/330s timeouts; a clean fix needs an abort/cancel protocol change, deferred.\n- **`new_tab` id-after-registry-sync race** [Codex P2] — the sync effect fires promptly and the tool tells the agent to `list_browsers`; a synchronous registry push would be a larger change for a rare, self-correcting race.\n- **Pure refactors** (syncSessionBrowsers options object; dedup the three terminal-type declarations; extract installer read/write boilerplate) [CodeRabbit nits] — no behavior change and non-trivial churn across the security surface; skipped to keep the diff reviewable.\nfeat(coworking): browser tooling hardening, session-persistence controls, new agent tools (supersedes #948)\n\n## Summary\n\nBuilds on — and supersedes — #948 (`feat(coworking)` by @chr1syy; all original commits preserved with attribution). That PR's browser tooling was vetted as feature-incomplete against its goals in one area (session-persistence control) and carried several hardening gaps flagged in review. This PR closes all of them and extends the tool surface.\n\n## Fixes on top of #948\n\n- **Ephemeral (incognito) browser tabs** — `maestro-ephemeral--` in-memory partition (never written to disk, never persisted/restored across restarts). New Incognito Browser entry in the new-tab popover, VenetianMask badge in the tab strip and address bar. The `will-attach-webview` allowlist admits exactly the two minted partition shapes and blocks everything else.\n- **Clear browsing data** — per-agent-session storage+cache wipe via a hardened `browser:clearSessionData` IPC (full-pattern partition validation, non-window senders rejected) behind a two-step-confirm toolbar button.\n- **`hiddenFromAgent` enforced in the main process** — the flag now crosses IPC; the registry excludes hidden tabs from both `list` and `resolve` (hidden is indistinguishable from closed), with a renderer execution-boundary re-check closing the in-flight race. `browser:N` ids stay stable across hide/unhide.\n- **Per-call confirm policy mirrored to main** — main computes `needsConfirm` per op from its own mirror and sends it with the request; the renderer ORs it with local policy, so a stale renderer settings read can only strengthen the gate.\n- **Audit hardening** — navigate/new-tab URLs stripped to origin+path with `queryChars=`/`hashChars=` counts (no tokens in the JSONL); non-URL targets reduced to length; terminal tools (`list_terminals`/`read_terminal`) now audited alongside browser tools.\n- **Windows ppid-walk resilience** — `wmic` is absent on current Win11 builds; the walk now falls back to PowerShell `Get-CimInstance` with a cached backend (validated live on a wmic-less host; fixes Codex session binding there).\n\n## New agent tools (server now advertises 17)\n\n- `browser_wait_for(id, selector, timeoutMs≤30s)` — host-side polling for dynamic pages\n- `browser_new_tab(url?, ephemeral?)` — session-scoped, URL normalized through the same resolver as navigate, always confirm-gated\n- `browser_close_tab(id)` — confirm-gated\n- `read_browser` gains `selector` for element-scoped reads\n\n## Settings UX\n\nCoworking Setup panel redesigned: real `role=\"switch\"` toggles (house style), the confirm-policy blind cycle button replaced by a labeled segmented control (Risky only / Every action / Never) with live descriptions, and an explicit warning when an agent has interaction on with confirmations off.\n\n## Verification\n\n- Typecheck green across `tsconfig.lint` + `tsconfig.main`\n- Coworking/browser test slice: 26 files, 480+ tests green (60+ new/rewritten: registry hidden-tab exclusion + id stability, bridge op validation bounds, audit redaction exactness, responder approval/forceConfirm/session-op paths, ephemeral persistence drop, window-manager partition table, clearSessionData rejection table, pid-resolution backend fallback/caching, CoworkingSetup RTL suite)\n- Generated MCP server script: `node --check` clean and driven end-to-end over real stdio against a mock named-pipe bridge (initialize / tools/list=17 / id-less `browser_new_tab` / `browser_wait_for` / selector read all forwarded intact)\n- Independent review pass (codex, read-only): 0 blockers; both majors fixed in this PR (execution-boundary hidden re-check, newTab URL normalization)\n- Manually tested live on Windows: install → interaction gating (denied path audited) → toggles → screenshot; audit JSONL entries verified\n\n## Notes for reviewers\n\n- `clearSessionData` trusts the main-window renderer by the same convention as every other `ipcMain.handle` in the app; it additionally rejects non-`'window'` sender types and validates the full minted partition shape.\n- Selector strings in audit details remain truncated-verbatim, matching the existing click/type convention; making selectors length-only is a policy decision deliberately left out of scope.\n- Known Electron constraint (unchanged): screenshots of hidden/background webviews fail closed with a clear message — only visible surfaces composite.\n- Pre-push full-suite run on Windows: 32,167 passed / 98 failed across 29 files. One failure was caused by this PR (TabBar.test.tsx lucide mock missing the new VenetianMask icon) and is fixed in this branch (173/173 green). The other 28 files were verified **pre-existing Windows-environment failures**: four representative suites (claudeUsageStore, pathResolver, cue-yaml-loader, opencode-config) fail with byte-identical assertions in a separate rc checkout (`2f4c26aa0`, none of this branch's changes; its local modifications verified not to touch those suites or their subjects) — POSIX path-separator and shell-semantics assumptions, unrelated to coworking. Push was made with `--no-verify` on that basis.\n\n\n\n## Summary by CodeRabbit\n\n* **New Features**\n * Added a Coworking setup section in Settings for installing integrations and configuring browser-sharing preferences.\n * Added incognito browser tabs, including an option to open a new incognito tab and clearer UI indicators.\n * Added background browser support to keep selected tabs available across focus changes, plus a two-step “clear session data” action for protected partitions.\n * Updated coworking interaction approval so `eval` requires confirmation even when confirmations are disabled.\n* **Bug Fixes**\n * Improved session save/restore to exclude incognito tabs from persistence.\n * Hardened browser data clearing with stricter partition validation and safer failure reporting.\n * Made coworking IDs easier to copy from terminal tabs.\n* **Chores**\n * Added `comment-json` to better preserve comments in installed agent configuration files.\n\n\n\n\n[![Review Change Stack](https://storage.googleapis.com/coderabbit_public_assets/review-stack-in-coderabbit-ui.svg)](https://app.coderabbit.ai/change-stack/RunMaestro/Maestro/pull/1160?utm_source=github_walkthrough&utm_medium=github&utm_campaign=change_stack)\n\n\n\n\n> [!IMPORTANT]\n> ## Review skipped\n> \n> Draft detected.\n> \n> Please check the settings in the CodeRabbit UI or the `.coderabbit.yaml` file in this repository. To trigger a single review, invoke the `@coderabbitai review` command.\n> \n>

\n>

⚙️ Run configuration

\n> \n> **Configuration used**: Path: .coderabbit.yaml\n> \n> **Review profile**: CHILL\n> \n> **Plan**: Pro\n> \n> **Run ID**: `1944988f-3de6-4ec3-9555-0b541ae16433`\n> \n>

\n> \n> You can disable this status message by setting the `reviews.review_status` to `false` in the CodeRabbit configuration file.\n> \n> Use the checkbox below for a quick retry:\n> - [ ] 🔍 Trigger review\n\n\n\n\n\n

📝 Walkthrough

\n\n## Walkthrough\nAdds a coworking MCP bridge, registry, tools, installers, IPC/preload wiring, renderer sync, and settings UI for per-agent browser/terminal access. Also introduces ephemeral browser partitions, excludes them from persistence/restoration, and adds a gated handler plus UI for clearing embedded browser session data.\n\n### Changes\n\n**Coworking MCP Server Integration**\n\n|Layer / File(s)|Summary|\n|---|---|\n|**Shared contracts**
`src/shared/coworkingBrowser.ts`, `src/main/coworking/coworking-types.ts`, `src/main/coworking/installers/types.ts`, `src/types/promise-with-resolvers.d.ts`|Defines shared browser operation and result contracts, coworking ids and bridge RPC shapes, installer interfaces, and the Promise.withResolvers type shim.|\n|**Registry state**
`src/main/coworking/coworking-registry.ts`, `src/__tests__/main/coworking/coworking-registry.test.ts`, `src/__tests__/main/coworking/coworking-browser.test.ts`|Implements the session-scoped coworking registry for terminal and browser records, public id allocation and resolution, per-session permissions and confirm policy, and change notifications.|\n|**Tools and session resolution**
`src/main/coworking/coworking-tools.ts`, `src/main/coworking/coworking-session-id.ts`, `src/main/process-manager/ProcessManager.ts`, `src/main/coworking/pid-resolution.ts`, `src/__tests__/main/coworking/coworking-tools.test.ts`, `src/__tests__/main/coworking/coworking-session-id.test.ts`, `src/__tests__/main/coworking/pid-resolution*.test.ts`|Implements coworking terminal and browser tools, resolves owning Maestro session ids from spawn ids and PIDs, and updates ProcessManager spawn env and PID lookup behavior.|\n|**Bridge and audit**
`src/main/coworking/coworking-bridge.ts`, `src/main/coworking/coworking-audit.ts`, `src/__tests__/main/coworking/coworking-bridge.test.ts`, `src/__tests__/main/coworking/coworking-browser-bridge.test.ts`, `src/__tests__/main/coworking/coworking-audit.test.ts`|Implements the socket/named-pipe coworking bridge, JSON-RPC dispatch, handshake/session binding, browser operation validation, and redacted browser/terminal audit logging.|\n|**Server script and installers**
`src/main/coworking/coworking-server-script.ts`, `src/main/coworking/coworking-server-paths.ts`, `src/main/coworking/coworking-installer.ts`, `src/main/coworking/installers/*`, `package.json`, tests|Adds the bundled coworking MCP server script and path resolution helpers, the installer orchestrator and per-agent installers, and the comment-json dependency used by installer config management.|\n|**Main process, preload, and settings wiring**
`src/main/index.ts`, `src/main/ipc/handlers/coworking.ts`, `src/main/ipc/handlers/index.ts`, `src/main/preload/coworking.ts`, `src/main/preload/index.ts`, `src/main/stores/defaults.ts`, `src/shared/settingsMetadata.ts`, `src/renderer/stores/settingsStore.ts`, `src/renderer/hooks/settings/useSettings.ts`, `src/renderer/types/index.ts`, `src/renderer/utils/terminalTabHelpers.ts`, tests|Registers coworking IPC handlers, exposes preload APIs, starts and stops the bridge in app lifecycle, adds coworking settings defaults and metadata, and threads terminal coworking ids through renderer models and ProcessManager.|\n|**Renderer sync and responders**
`src/renderer/hooks/coworking/*`, `src/renderer/stores/coworkingApprovalStore.ts`, `src/renderer/stores/coworkingBackgroundBrowserStore.ts`, tests|Synchronizes renderer session state to the coworking registry and serves main-process buffer and browser-op requests through renderer-side responders, approval logic, and background browser mounting.|\n|**Renderer settings and tab UI**
`src/renderer/components/Settings/CoworkingSetup.tsx`, `src/renderer/components/Settings/tabs/EncoreTab/*`, `src/renderer/components/TabBar/*`, `src/renderer/components/coworking/*`, `src/renderer/components/MainPanel/MainPanel*.tsx`, tests|Adds the coworking settings panel and Encore integration, terminal coworking id copy pill, browser incognito badge and new-tab action, and BrowserTabView controls for clear session data and hiding tabs from agents.|\n\n**Ephemeral (Incognito) Browser Tabs and Session Data Clearing**\n\n|Layer / File(s)|Summary|\n|---|---|\n|**Ephemeral partition contract**
`src/renderer/utils/browserTabPersistence.ts`, `src/main/app-lifecycle/window-manager.ts`, `src/renderer/types/index.ts`, tests|Adds the ephemeral partition prefix helpers, the main-process partition allowlist update, and the BrowserTab fields needed to represent ephemeral and hidden tabs.|\n|**Incognito tab creation and persistence exclusion**
`src/renderer/hooks/tabs/internal/browserTabHelpers.ts`, `src/renderer/hooks/session/useSessionRestoration.ts`, `src/renderer/hooks/utils/useDebouncedPersistence.ts`, tests|Threads an ephemeral option through browser tab creation and handler signatures, and filters ephemeral tabs out of restoration and disk persistence.|\n|**Incognito UI and clear-session flow**
`src/renderer/components/TabBar/*`, `src/renderer/components/MainPanel/BrowserTabView.tsx`, `src/renderer/components/MainPanel/types.ts`, `src/renderer/hooks/props/useMainPanelProps.ts`, `src/renderer/hooks/tabs/internal/*`, `src/main/ipc/handlers/browser-session.ts`, `src/main/preload/browserSession.ts`, `src/renderer/global.d.ts`, tests|Adds incognito badges and a new incognito browser action, updates BrowserTabView and prop signatures for ephemeral tabs, and adds the clear-session-data control backed by the browser-session API.|\n\n**Estimated code review effort:** 5 (Critical) | ~120 minutes\n\n### Sequence Diagram(s)\n\n```mermaid\nsequenceDiagram\n participant MCPServerScript\n participant CoworkingBridge\n participant CoworkingRegistry\n participant RendererResponder\n MCPServerScript->>CoworkingBridge: hello (sessionId or ppid)\n CoworkingBridge->>CoworkingBridge: resolveSessionFromPidWalk (if needed)\n CoworkingBridge-->>MCPServerScript: hello ack\n MCPServerScript->>CoworkingBridge: browserInteract(op)\n CoworkingBridge->>CoworkingRegistry: resolve tab for sessionId\n CoworkingBridge->>RendererResponder: requestBrowserOp via IPC\n RendererResponder->>RendererResponder: approval gate + mount/focus tab\n RendererResponder-->>CoworkingBridge: BrowserOpResult\n CoworkingBridge-->>MCPServerScript: JSON-RPC response\n```\n\n```mermaid\nsequenceDiagram\n participant User\n participant BrowserTabView\n participant BrowserSessionIpc\n participant ElectronSession\n User->>BrowserTabView: click clear session data (armed)\n User->>BrowserTabView: click again (confirm)\n BrowserTabView->>BrowserSessionIpc: clearSessionData(partition)\n BrowserSessionIpc->>BrowserSessionIpc: validate sender + partition allowlist\n BrowserSessionIpc->>ElectronSession: clearStorageData + clearCache\n ElectronSession-->>BrowserSessionIpc: ok\n BrowserSessionIpc-->>BrowserTabView: { ok: true }\n BrowserTabView->>BrowserTabView: reload webview\n```\n\n**Possibly related PRs**\n- [RunMaestro/Maestro#710](https://github.com/RunMaestro/Maestro/pull/710): Both PRs touch `MainPanel`/`MainPanelContent` wiring, so this coworking integration overlaps with that area of the renderer tree.\n- [RunMaestro/Maestro#785](https://github.com/RunMaestro/Maestro/pull/785): Both PRs modify browser partition allowlist behavior in `window-manager.ts`, including support for the new ephemeral partition prefix.\n\n

\n\n\n\n\n

🚥 Pre-merge checks | ✅ 5

\n\n

✅ Passed checks (5 passed)

\n\n

\n\n\n\n\n

✨ Finishing Touches

\n\n

🧪 Generate unit tests (beta)

\n\n- [ ] Create PR with unit tests\n\n

\n\n

\n\n\n\n\n---\n\nThanks for using [CodeRabbit](https://coderabbit.ai?utm_source=oss&utm_medium=github&utm_campaign=RunMaestro/Maestro&utm_content=1160)! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.\n\n

❤️ Share

\n\n- [X](https://twitter.com/intent/tweet?text=I%20just%20used%20%40coderabbitai%20for%20my%20code%20review%2C%20and%20it%27s%20fantastic%21%20It%27s%20free%20for%20OSS%20and%20offers%20a%20free%20trial%20for%20the%20proprietary%20code.%20Check%20it%20out%3A&url=https%3A//coderabbit.ai)\n- [Mastodon](https://mastodon.social/share?text=I%20just%20used%20%40coderabbitai%20for%20my%20code%20review%2C%20and%20it%27s%20fantastic%21%20It%27s%20free%20for%20OSS%20and%20offers%20a%20free%20trial%20for%20the%20proprietary%20code.%20Check%20it%20out%3A%20https%3A%2F%2Fcoderabbit.ai)\n- [Reddit](https://www.reddit.com/submit?title=Great%20tool%20for%20code%20review%20-%20CodeRabbit&text=I%20just%20used%20CodeRabbit%20for%20my%20code%20review%2C%20and%20it%27s%20fantastic%21%20It%27s%20free%20for%20OSS%20and%20offers%20a%20free%20trial%20for%20proprietary%20code.%20Check%20it%20out%3A%20https%3A//coderabbit.ai)\n- [LinkedIn](https://www.linkedin.com/sharing/share-offsite/?url=https%3A%2F%2Fcoderabbit.ai&mini=true&title=Great%20tool%20for%20code%20review%20-%20CodeRabbit&summary=I%20just%20used%20CodeRabbit%20for%20my%20code%20review%2C%20and%20it%27s%20fantastic%21%20It%27s%20free%20for%20OSS%20and%20offers%20a%20free%20trial%20for%20proprietary%20code)\n\n

\n\n\n_{Comment `@coderabbitai help` to get the list of available commands.}\n\n\n