Pull Request Explorer

Summary

Fix CI: support perf commits in desktop canary trigger

Ensures performance commits trigger desktop canary builds, improving release pipeline reliability.

Health Assessment

• • Rapid review and merge indicate low complexity and high confidence in the change.

Summary

Add thread notifications and deep link for comments

Enables comprehensive comment notifications and deep linking, improving collaboration and user experience.

Health Assessment

• • The PR was reviewed and merged within 5 hours, with AI assistance speeding up code authoring and review. The large code change was handled efficiently, indicating a well-managed iteration process.

Summary

Improve review panel performance by unmounting hidden panel

Optimizes memory and latency by unmounting the review panel when hidden, improving user experience during tab switches.

Health Assessment

• • Quick review and merge within a day, minimal changes, low risk.

Summary

Show reclaimable renderer memory in memory widget

Enhances memory monitoring for developers, enabling quicker detection of leaks and better performance tuning.

Health Assessment

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

Summary

Gate first‑screen import graph, reduce size 41%

Adds a CI gate that limits the first‑screen bundle size, cutting the desktop entry from 3.84 MB to 2.26 MB gzip and improving initial load performance.

Health Assessment

• • Fast review and merge within 6.6 hours, minimal rework, indicating a smooth process.

Summary

Add shared-viewer notice on acceptance page

Non-owners opening a shared acceptance link now see a neutral notice indicating read‑only access, improving clarity and preventing confusion.

Health Assessment

• • PR merged quickly with minimal review, indicating straightforward implementation and clear documentation.

Summary

Add task assignment notifications and member mentions

Enables workspace task assignment notifications, comment pings, and member mentions, improving collaboration and visibility.

Health Assessment

• • The PR had a rapid review cycle (under 3 hours) but required multiple iterations and a large code change, indicating high complexity and potential risk despite the short merge time.

Summary

Add optional Elasticsearch to Docker Compose

Provides an opt‑in single‑node Elasticsearch profile for Docker Compose, enabling self‑hosted deployments to use Elasticsearch for full‑text search without altering default behavior.

Health Assessment

• • Rapid iteration with AI‑assisted review, but large scope and many comments indicate moderate complexity; quick cycle time mitigates risk.

Summary

Fix device gateway silent failure handling

Ensures device read failures are surfaced instead of silently treated as offline, improving reliability and traceability for agent runs.

Health Assessment

• • Quick, single-commit fix with no review needed; minimal code changes improve error visibility and prevent silent fallback to cloud sandbox.

Summary

Fix device gateway handshake timeout

Prevents device clients from hanging during TLS handshake, ensuring timely reconnection and reducing downtime for device-bound tasks.

Health Assessment

• • Merged within 6 hours with no rework, indicating a straightforward bug fix.

Summary

Fix double-writing masked credentials in sandbox

Prevents duplicate credential writes that caused authentication failures, improving reliability for sandboxed environments.

Health Assessment

• • Quick fix with minimal review, merged within 0.8 hours, indicating efficient process.

Summary

Add model cards for GLM-5.3-Flash, V4 Flash, Fable 5.1

Enables users to access newly launched multimodal models with accurate pricing and capabilities, improving model selection and usage.

Health Assessment

• • Rapid AI-assisted development and review led to a 3.5‑hour cycle time with minimal friction and a small code footprint.

Summary

Fix LocalFile chip preview and rerun handoff

Ensures file previews open correctly, user messages render LocalFile chips, and acceptance rerun functionality is restored, improving user experience and reliability.

Health Assessment

• • Fast cycle time and single review round indicate smooth integration; AI assistance likely accelerated development, but the large code change warrants moderate risk monitoring.

Summary

Add task intent reading before creation

Enables intelligent task creation by analyzing draft intent, reducing errors and improving user experience.

Health Assessment

• • PR required extensive rework after initial review, with multiple commits to address feedback, indicating complex changes. Cycle time over 3 days suggests slower integration.

Summary

Fix OIDC consent redirect to preserve origin

Ensures OIDC consent and callback redirects use the original request origin, preventing unintended domain jumps and preserving user session integrity. This change improves security and user experience for cross-domain authentication flows.

Health Assessment

• • Quick review and merge within 14 hours, minimal rework, indicating a straightforward bug fix with clear intent.

Summary

Add workspace document likes

Enables users to like documents within a workspace, triggering notifications and real‑time updates, improving engagement.

Health Assessment

• • Rapid iteration with multiple review rounds indicates complexity, but the short cycle time shows efficient resolution.

Summary

Add project working directory schema

Adds a schema for project working directories, enabling device‑backed execution directories and improving topic association management.

Health Assessment

• • Large schema change with minimal review iterations, but long cycle time indicates possible blockers or slow merge process.

Summary

Add KB sidebar search, add button, comment image thumbnails

Enhances knowledge base navigation and content creation, improving user productivity and content quality.

Health Assessment

• • Merged within 10.7 hours with only one review round and three automated comments, indicating a smooth review process despite the large 1,964 line code churn.

Summary

Fix CLI goal run resilience and graph edge logic

Improves reliability of long-running goal executions by handling transient failures, idle polling, and correcting graph edge representation, reducing manual intervention. Ensures accurate goal graph relationships for better planning.

Health Assessment

• • Fast turnaround with minimal review rounds; quick fix to production issue.

Summary

Fix model-runtime: ensure tool call with Fable 5.1

Adds logic to force tool calls for Fable 5.1 models by appending explicit instructions, preventing silent failures and improving reliability for users.

Health Assessment

• • Quick fix with minimal changes, no review needed, merged within 0.8 hours.

Summary

Fix workspaceId passing to cloud-sandbox MarketService

Aligns MarketService construction with workspaceId to ensure credential injection works for workspace-scoped requests, preventing silent failures.

Health Assessment

• • Resolved in under an hour with minimal changes, indicating a straightforward bug fix with low risk.

Summary

Extend local terminal observation timeout

Improves stability of local terminal interactions by extending the observation timeout, reducing premature termination of processes.

Health Assessment

• • Quick fix with minimal changes, AI-assisted review helped catch issues early.

Summary

Classify 9 error patterns from triage

Adds new error pattern matching for model runtime, improving error classification and user experience.

Health Assessment

• • Quick review and merge within 10 hours, minimal rework, indicating high confidence in changes.

Summary

Fix reasoning passback for GLM and Kimi models

Ensures reasoning content is echoed back for GLM and Kimi thinking models, preventing tool call failures and improving reliability for users.

Health Assessment

• • Quick fix with minimal changes, fast review and merge, low risk.

Summary

Fix agent-runtime stream error classification

Ensures content policy violations are correctly classified instead of being collapsed to generic 500 errors, improving error reporting and user experience.

Health Assessment

• • Quick AI-assisted fix with minimal review and rework, resolving a critical error classification issue.

Summary

Fix model-runtime: classify SubscriptionPlanLimit in error spec table

Adds missing error spec for SubscriptionPlanLimit, enabling proper classification and visibility of quota-related failures.

Health Assessment

• • Quick AI-assisted fix with minimal code changes and fast review turnaround, indicating low complexity and low risk.

Summary

Reset paginated lists before search

Ensures memory search results are accurate by resetting pagination and handling failures, improving user experience and reliability.

Health Assessment

• • The PR required rapid iteration with multiple automated review comments, indicating a complex bug fix that needed several adjustments before approval.

Summary

Add Claude Fable 5.1 tool_choice support

Enables Claude Fable 5.1 to use auto + strict for generateObject, preventing 400 errors and improving runtime compatibility.

Health Assessment

• • Fast review and merge with minimal back‑and‑forth; small iteration and low complexity suggest a low‑risk change.

Summary

Fix CLI daemon startup readiness

Ensures the CLI only reports daemon start success after the daemon process is fully ready, improving reliability and error reporting for users.

Health Assessment

• • Fast cycle time and minimal review rounds indicate efficient resolution of a critical CLI issue.

Summary

Fix topic and message search navigation

Improves user experience by ensuring search results in the topic sidebar and conversation list are correctly highlighted and scrolled, reducing navigation friction.

Health Assessment

• • The PR was merged within 3 hours with a single AI-assisted review, but required three additional commits after the review, indicating iterative refinement. The large code churn suggests a high-impact bugfix that was resolved quickly.

Summary

Add cost attribution metrics to search

Adds cost attribution metrics for full‑text search, enabling better telemetry and resource usage insights for the platform.

Health Assessment

• • Rapid review and merge within 30 minutes indicates high confidence and minimal friction.

Summary

Bump CLI version to 0.0.52

Update the CLI package and its man page to version 0.0.52, ensuring consistency across documentation and enabling new features.

Health Assessment

• • Quick merge with minimal changes, indicating a straightforward version bump.

Summary

Fix Telegram webhook verification and SDK pinning

Ensures bots reliably receive updates by enforcing webhook verification and stabilizes SDK behavior, preventing future runtime changes that could disrupt bot functionality.

Health Assessment

• • The PR was reviewed and merged quickly with minimal back‑and‑forth, but the change touches over 2,700 lines across 20 files, indicating a substantial codebase impact.

Summary

Fix hide agent-management callAgent inside sub-agent runs

Prevents sub-agents from attempting unsupported callAgent calls, improving reliability and reducing unnecessary errors.

Health Assessment

• • The PR was merged in just over six hours with minimal review, indicating a straightforward bug fix that was quickly validated.

Summary

Refactor aiAgent: extract approvalResume pipeline and finalize execAgent skeleton

Reorganizes the aiAgent service into modular pipelines, shrinking code size and improving maintainability while keeping the public API unchanged, enabling easier future enhancements.

Health Assessment

• • The PR was reviewed and merged within 4 hours, with a single commit and no API changes, indicating a smooth, low-risk refactor.

Summary

Fix slow login shell CLI detection

Improves reliability of CLI binary detection by caching login shell PATH and handling timeouts, preventing task failures for users.

Health Assessment

• • The PR was reviewed and merged within 7.2 hours, with a single commit and minimal back‑and‑forth, indicating a straightforward, low‑friction fix.

Summary

Refactor search routing to exclusive provider

Simplifies full‑text search configuration by making provider selection exclusive, reducing routing complexity and ensuring consistent backend usage.

Health Assessment

• • Single commit with a 19.5‑hour cycle time indicates a smooth review process and minimal friction.

Summary

Fix auth callback origin preservation

Ensures users are redirected to the correct origin after authentication, preventing misdirected callbacks.

Health Assessment

• • Resolved in under a day with a single commit, indicating a straightforward bug fix with minimal risk.

Summary

Preserve default action lists in chat input store

Fixes crashes in the Electron renderer by ensuring default action lists are preserved when omitted, improving stability during agent creation.

Health Assessment

• • Merged within 3 hours with minimal changes, indicating a straightforward fix and efficient review process.

Summary

Pin polling regression test with fixed count

Adds a hardcoded test case to detect polling regression when limit changes, ensuring test fails if limit reverts.

Health Assessment

• • Quick test addition with AI assistance, minimal risk.

Summary

Fix credential injection into sandbox

Ensures injected credentials are written to the sandbox environment, improving tool reliability and security.

Health Assessment

• • Quick resolution with minimal review, indicating a straightforward bug fix that required only a single commit and a single review.

Summary

Fix credential routing to personal endpoint

Ensures member-owned credentials in the workspace list use the correct personal API, preventing 404 errors when editing or deleting them.

Health Assessment

• • Quick fix with minimal rework; AI review suggestions helped identify the routing issue and guided the implementation.

Summary

Fix agent tool completion propagation

Ensures successful tool executions correctly trigger side-effect hooks, preventing missed updates to worktree metadata.

Health Assessment

• • Single commit, minimal changes, regression test added, merged within a day.

Summary

Fix IPC listeners and hidden editor roots

Removes lingering IPC listeners and cleans up hidden editor roots to prevent memory leaks, improving desktop app stability and performance.

Health Assessment

• • The PR was reviewed quickly but required multiple automated review iterations, indicating complex changes that were thoroughly vetted before merging.

Summary

Fix upload: support memory-safe large file uploads

Improves stability and performance for large file uploads by streaming data and using multipart uploads, preventing memory crashes and ensuring reliable uploads.

Health Assessment

• • AI-assisted review enabled rapid feedback; extensive tests validate new streaming logic; large code changes but no major risk due to thorough testing.

Summary

Non-blocking goal creation and detail page polish

Improves user experience by making goal creation instant, adding live requirement editing, and enhancing the goal detail UI with better graphics and controls.

Health Assessment

• • PR underwent multiple review rounds and many commits in a short period, indicating complex changes and high iteration intensity.

Summary

Refactor aiAgent pipeline stages extraction

Modularizes large execAgent stages, reducing code size and complexity for easier maintenance and potential performance gains.

Health Assessment

• • Rapid refactor with minimal review, high code churn but no new functionality, likely low risk to business.

Summary

Add role/date filters and topic info to message list

Enables advanced filtering of messages by role, date, and topic, improving data retrieval and user experience. CLI now displays agent and topic details, enhancing visibility into conversations.

Health Assessment

• • Fast turnaround with minimal review iterations indicates a well‑scoped, straightforward enhancement.

Summary

Fix tool-call repeat limit to 20

Adjusts the threshold for repeated tool calls to reduce false positives in polling scenarios, improving reliability for agents.

Health Assessment

• • Quick AI‑assisted fix with minimal changes and no back‑and‑forth, indicating a low‑risk, fast‑track PR.

Summary

Fix lint errors in acceptance feature

Resolves lint errors that were blocking code quality checks, ensuring smooth CI and stable code quality for the acceptance feature.

Health Assessment

• • Quick AI‑assisted fix with minimal changes, resolved lint errors promptly, keeping CI pipeline healthy.