Pull Request Explorer

Summary

Fix remote-device picker on device-locked runs

Ensures that when a run is locked to a device, the remote-device picker is removed, preventing redundant activation and improving user experience.

Health Assessment

• • Quick turnaround with minimal rework; AI review bots provided feedback but no major changes required.

Summary

Restore device execution branch for execScript

Fixes critical device execution path, ensuring local device runs work correctly and preventing silent fallback to sandbox, improving reliability and user experience.

Health Assessment

• • The PR had a rapid review but required many commits to address extensive changes across backend and frontend, indicating significant rework; the large code churn and multiple iterations suggest a high-risk fix that was ultimately merged within a day.

Summary

Exclude system topics from unread badges

Fixes unread badge counts by excluding system-triggered topics from the home sidebar, aligning badge behavior with the main chat sidebar.

Health Assessment

• • Fast 0.4‑hour cycle, no review needed, straightforward bug fix.

Summary

Unify save-state feedback in settings

Ensures autosave failures are clearly shown and closes UX gaps across settings tabs, improving reliability and user clarity.

Health Assessment

• • Fast cycle time of 3.9 hours and immediate review indicate efficient collaboration and low friction.

Summary

Fix deletion of errored step in heterogeneous run

Ensures that deleting a failed step in a heterogeneous chat run only removes that step, preserving prior successful steps and preventing accidental data loss.

Health Assessment

• • Quick resolution with minimal rework; AI review helped catch edge cases.

Summary

Gate react-scan behind REACT_SCAN flag

Enables optional react-scan diagnostics, reducing dev overhead when not needed.

Health Assessment

• • Fast review and merge with minimal changes, no risk to production.

Summary

Fix crawl failures traceability per implementation

Improves debugging of crawler failures by adding implementation context to logs, enabling faster issue resolution and reducing downtime.

Health Assessment

• • Fast turnaround with minimal changes and quick review indicates low complexity and high confidence in the fix.

Summary

Add verify interaction cost reporting

Adds detailed interaction cost reporting for agents, enabling better cost analysis and optimization. Provides new metrics for user-equivalent interaction cost, improving transparency and performance tuning.

Health Assessment

• • The PR was merged quickly with minimal review, indicating a straightforward implementation of a new reporting feature. Large scope but no rework suggests high confidence in code quality.

Summary

Fix auto-run handling for dynamic approvals

Ensures dynamic approval policies honor auto‑run, preventing unnecessary approval prompts and improving agent runtime reliability.

Health Assessment

• • Quick fix with minimal changes, reviewed and merged within minutes, indicating low complexity and high confidence.

Summary

Fix desktop notification deep link handling

Ensures desktop notifications correctly navigate to specific chat topics, improving reliability across workspaces and tabs.

Health Assessment

• • Merged within 2 hours with minimal back‑and‑forth, indicating a smooth review process.

Summary

Hide agent signal threads from topic lists

Removes agent signal self‑iteration isolation threads from user‑facing topic lists while keeping them for audit and debugging, improving clarity for end users.

Health Assessment

• • Fast cycle time and no review rounds indicate a straightforward, low‑risk change.

Summary

Fix guard for AskUserQuestion payloads

Prevents UI crashes by validating payloads, improving stability and user experience.

Health Assessment

• • Fast cycle time and single review with many small commits indicate an efficient, low‑risk bugfix process.

Summary

Prefetch unread completed topic messages

Preloads messages for completed topics to eliminate loading delays when users switch topics, improving responsiveness.

Health Assessment

• • Merged in 2.5 hours with only two brief AI review comments, indicating a smooth review process.

Summary

Fix agent runtime lock conflict handling

Improves reliability by correctly handling stale and fresh lock conflicts, preventing duplicate execution and ensuring proper retry behavior.

Health Assessment

• • The PR was merged within 1.8 hours with minimal rework; AI-assisted reviews helped identify edge cases early, resulting in a smooth, low‑friction process.

Summary

Update system announcement notification

Updates the system announcement notification UI to reflect new design guidelines, improving clarity for users.

Health Assessment

• • Quick, single-commit PR with minimal changes and no review back-and-forth, indicating a low-risk, straightforward update.

Summary

Fix agent runtime step lock retries

Improves reliability of agent runtime by preventing duplicate step execution and reducing unnecessary retries, ensuring smoother operation and better resource usage.

Health Assessment

• • Quick fix with minimal review, indicating low complexity and high confidence.

Summary

Fix Codex file change diff rendering

Corrects how Codex file change diffs are rendered, preventing entire file diffs when no snapshot is provided, improving UI accuracy.

Health Assessment

• • Fast turnaround with minimal rework; quick review and merge within 2 hours.

Summary

Add inline short verify text evidence

Enables inline display of verification evidence, improving user experience and reducing modal interactions.

Health Assessment

• • Rapid iteration with AI-assisted review, minimal friction, quick merge.

Summary

Stabilize QStash step startup

Ensures agent runtime starts reliably by delaying QStash messages for sub‑second delays, preventing race conditions. Adds diagnostics for missing Redis metadata to aid incident investigation.

Health Assessment

• • The PR was merged within 18 minutes of opening, with a single commit and no back‑and‑forth reviews, indicating a straightforward fix. No major scope or risk concerns were identified.

Summary

Fix working directory path rendering guard

Prevents crashes caused by dirty working directory data, improving agent page stability.

Health Assessment

• • Merged within 1.3 hours with minimal review comments, indicating a straightforward bug fix with low complexity and risk.

Summary

Add inline expand/collapse for long task content

Enables users to view full task run content without navigating away, improving usability and reducing friction.

Health Assessment

• • Fast review and merge with a single commit; UI-only change with minimal scope and no backend impact.

Summary

Fix agent-signal receipt anchor preservation

Ensures receipt and trigger anchors are retained during source normalization, improving UI placement accuracy and preventing data loss.

Health Assessment

• • Quick turnaround with AI review, minimal rework, and a focused bug fix across backend and frontend.

Summary

Fix stale running topics without alive ops

Prevents UI spinners from persisting by cleaning up stale topics, improving user experience.

Health Assessment

• • Fast turnaround with AI review, minimal human intervention, large code changes focused on UI cleanup.

Summary

Fix weak model tool selection and redundant device activation

Improves agent behavior by ensuring weak models use the agent-browser skill for dynamic pages and prevents unnecessary device activation, enhancing reliability and efficiency.

Health Assessment

• • Quick, low‑line change with no tests added; fast review and merge cycle indicates minimal risk and high confidence in correctness.

Summary

Fix CLI binary resolver for headless runs

Ensures sandbox and remote runs use the same binary resolution logic as the desktop app, preventing crashes due to missing global shims.

Health Assessment

• • Merged within 11 hours with a single review comment, indicating a smooth review process despite the large code change.

Summary

Generalize home free credit badge to promo banner

Creates a reusable promo banner slot on the home page, simplifying future promotional content without affecting current user experience.

Health Assessment

• • Fast cycle with a single review and minimal changes indicates low risk and high process efficiency.

Summary

Fix sign‑in UX audit findings

Improves the sign‑in flow with consistent button hierarchy, a persistent inbox state, inline error handling, and full i18n support, enhancing user experience and reducing friction.

Health Assessment

• • Fast turnaround with AI‑assisted review, minimal rework, and a single commit indicates an efficient process.

Summary

Fix LLM outbound guards for empty enum, messages, budget

Improves reliability of LLM requests by preventing empty payloads and ensuring correct token budgeting, reducing failures and improving user experience. This reduces error rates and improves system stability.

Health Assessment

• • Fast turnaround with immediate review and minimal rework indicates a straightforward bug fix with clear boundaries. The PR was reviewed by an AI tool, adding a low friction layer to the process.

Summary

Fix agent signal receipt anchor semantics

Ensures agent signal receipts correctly reference the originating message, improving chat accuracy and preventing misattributed responses.

Health Assessment

• • Rapid review and merge with minimal changes, indicating low complexity and risk.

Summary

Fix chat snapshot topic branch/PR at send time

Prevents overwriting stored branch/PR when opening a topic, ensuring historical snapshots are preserved.

Health Assessment

• • The PR had a fast cycle time of 12.2 hours, indicating efficient review and merge processes. However, the large scope and multiple commits may have introduced some complexity.

Summary

Add Create Worktree Flow to Worktree Switcher

Enables users to create new git worktrees directly from the UI, streamlining branch management and reducing context switching.

Health Assessment

• • Fast review and merge (1.8h cycle, 0h review time) indicates high confidence and minimal friction; the change spans many layers but was delivered quickly with no major blockers.

Summary

Align tool_end payload+result for CLI agents

Ensures consistent event stream so onAfterCall hooks fire for CLI tools, improving reliability of tool execution and worktree detection.

Health Assessment

• • Rapid merge with minimal review indicates high confidence but large scope may increase risk.

Summary

Fix OIDC route to log 500 errors

Ensures critical OIDC errors are logged in production, improving observability and faster incident response.

Health Assessment

• • Quick fix with minimal changes, no rework, fast review and merge.

Summary

Add Weixin business bot integration

Enables Weixin business channel support, allowing users to interact via Weixin and expanding the platform’s communication reach.

Health Assessment

• • Rapid iteration with AI review bots and minimal human review led to a quick 4‑hour merge, indicating high confidence in code quality despite the large scope.

Summary

Remove internal issue markers from source comments

This PR cleans up internal Linear issue references from code comments, preserving readability while keeping the codebase public-friendly.

Health Assessment

• • The PR was merged within 14 hours with no review delays, indicating a straightforward, low-risk change that required minimal rework.

Summary

Fix sidebar topic spinner and ordering glitches

Improves user experience by preventing spurious topic creation, ensuring spinner stops correctly, and maintaining consistent sidebar ordering during new-topic flows.

Health Assessment

• • Rapid iteration with AI-assisted review led to quick fixes and thorough testing, resulting in a smooth merge within 4 hours.

Summary

Refactor CLI to add connect service management

Adds commands to install, start, stop, and manage a systemd service for the connect feature, improving reliability and user experience.

Health Assessment

• • The PR was reviewed solely by AI bots, resulting in a 4‑day cycle time and potential oversight.
• • Large code additions (~490 lines) without human review increase risk of undiscovered issues.

Summary

Add image preview for Claude Read tool

Enables image content from Claude Read to be uploaded and displayed as thumbnails, improving multimodal chat experience.

Health Assessment

• • Rapid iteration with AI-generated code and minimal review, but large code change introduces moderate risk.

Summary

Flatten agent-browser block handling to single decision

Simplifies anti‑bot block handling by immediately switching to real Chrome, reducing retries and improving scraping efficiency.

Health Assessment

• • Merged within 24 minutes with minimal changes and only automated review comments, indicating a smooth, low‑risk update.

Summary

Fix placeholder reuse in tool result resume

Fixes a UI glitch where agent placeholders were orphaned during tool result resumption. Ensures consistent avatar rendering and smoother user interactions.

Health Assessment

• • Fast turnaround with minimal rework; quick review and merge indicates low complexity.

Summary

Define explicit artifact TTLs for OIDC provider

Fixes OIDC device authorization flow by ensuring all artifact TTLs are positive integers, preventing server errors and enabling successful login.

Health Assessment

• • Quick fix with minimal changes, no rework, merged within 1 hour.

Summary

Fix memory tool argument validation

Ensures memory tool arguments are validated before persistence, preventing malformed data and improving reliability.

Health Assessment

• • Quick fix with minimal rework; AI review helped catch issues early.

Summary

Optimize persona upserts for memory performance

Reduces memory usage by skipping unchanged persona upserts and optimizing database queries.

Health Assessment

• • The PR was reviewed and merged quickly, with minimal changes and a clear description of the changes made.

Summary

Fix agent-browser skill anti-bot diagnosis

Improves the agent-browser skill to detect anti-bot blocks instead of retrying, reducing unnecessary retries and improving reliability.

Health Assessment

• • Rapid review and merge within minutes indicates a straightforward bug fix with minimal friction.

Summary

Add July 6 weekly changelog

Adds the July 6 weekly changelog entry in English and Chinese, covering user‑visible chat, workspace, Agent, Task, verification, reliability, and security updates from the past week.

Health Assessment

• • Quick merge with minimal review, indicating straightforward documentation update.

Summary

Fix memory render guard for non-array args

Prevents React component crashes caused by malformed model output, improving stability.

Health Assessment

• • Quick fix with minimal changes, no rework, fast review.

Summary

Paginate verify report list with infinite scroll

Adds cursor pagination and server‑side search to the verify report list, improving scalability and search accuracy for users.

Health Assessment

• • Fast review and merge with minimal friction; the PR was approved within minutes and merged shortly after, indicating a smooth process.

Summary

Show full connector catalog in settings

Users can now see all available connectors, improving discoverability and onboarding.

Health Assessment

• • Quick review and merge with minimal changes, indicating low complexity and risk.

Summary

Add interactive code context selections

This PR introduces interactive code context selection, enabling users to inject selected code into agent prompts and improving the chat experience.

Health Assessment

• • The PR was merged quickly after a single review, indicating confidence in the changes, but the extensive code modifications increase risk.

Summary

Fix cloud-sandbox fallback for offline devices

Ensures commands intended for a local device are not silently executed in the cloud sandbox when the device is offline, improving reliability and user trust.

Health Assessment

• • Fast review and merge (18.8h cycle time) with minimal comments and no major rework, indicating a straightforward bug fix with low risk.