Pull Request Explorer

Summary

Add signed commits guidance to docs

Provides contributors with instructions on signing commits, enhancing author verification and security.

Health Assessment

• • Docs-only PR merged within minutes, no code changes, minimal risk and review friction.

Summary

Unify Ctrl-C and Ctrl-D quit behavior

Ensures consistent exit handling in the agent TUI, improving user experience by aligning Ctrl-C with the two-press exit flow.

Health Assessment

• • Long review cycle but minimal code change, resulting in a low-risk bugfix.

Summary

Improve CUDA runtime staging and ccache

Ensures CUDA runtime is bundled correctly and build performance is improved via ccache, reducing CI failures and improving user experience.

Health Assessment

• • The PR required multiple CI and build-related iterations, with several changes requested before approval, but was merged quickly due to a short review cycle.

Summary

Upgrade llama.cpp to 0.3.0, fix Windows build, add grammar paths

This PR updates the llama.cpp engine to version 0.3.0, resolves Windows MAX_PATH build issues, and adds support for absolute grammar and template file paths, improving stability and configurability for users.

Health Assessment

• • The PR required a long review period and involved many changes across backend and frontend, indicating significant complexity and potential risk.

Summary

Fix agent: heal poisoned tool-call history

Ensures tool calls with non-object arguments are sanitized, preventing runtime errors and improving system reliability.

Health Assessment

• • The PR required a rebase and multiple review rounds, indicating some complexity, but ultimately merged after thorough review.

Summary

Add /context to break down context window usage

Introduces a TUI command that displays detailed context window usage, giving users visibility into token consumption and preventing oversized tool schemas.

Health Assessment

• • The PR had a moderate cycle time of 47.6 hours with a single review, but the large code changes required extensive testing to ensure correct context reporting.

Summary

Scope Rust-Check Caches to Main Branch

Improves CI caching efficiency by limiting cache storage to the main branch, preventing GitHub Actions cache limit breaches and ensuring reliable builds.

Health Assessment

• • Rapid review and merge with minimal changes; no blockers or extensive rework required.

Summary

Merge dev branch into main

Integrates latest development changes, including new CLI features, TUI enhancements, and CI pipeline updates, ensuring the main branch reflects current functionality.

Health Assessment

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

Summary

Port Cowork agent to Vercel AI SDK

Enables the desktop Cowork agent to run via the Vercel AI SDK, adding sandboxing, permissions, and UI updates for improved isolation and user experience.

Health Assessment

• • Rapid iteration with multiple commits in a single day and minimal review time suggests high confidence in the changes, though the large scope warrants careful monitoring for downstream impact.

Summary

Make RepoRoot lazy for Windows install

Fixes Windows install one-liner by deferring RepoRoot calculation, preventing errors when piped via irm|iex.

Health Assessment

• • Quick, single-commit fix with minimal code changes and no rework.

Summary

Add OAuth login for Claude and Codex

Enables OAuth account login for Claude and Codex BYOK subscriptions, fixing redirect issues when callback port is occupied.

Health Assessment

• • PR took over 9 days to merge with a single review after a long review period, indicating potential bottlenecks in the review process.

Summary

Improve TUI streaming and budget handling

Adds real‑time tool output to the terminal UI, refines layout, and prevents budget exhaustion errors, enhancing user experience and reliability.

Health Assessment

• • Fast cycle time and minimal rework indicate efficient review, but the large code churn and multiple review rounds suggest a high‑risk change that required careful validation.

Summary

Fix Claude Code 403 and port-in-use errors

Resolve authentication and server start errors for Claude Code integration, improving reliability on Windows.

Health Assessment

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

Summary

Add configurable ask‑tool timeout with auto‑select

Enables agents to auto‑select recommended options after a configurable timeout, preventing indefinite hangs and improving unattended run reliability.

Health Assessment

• • Fast cycle time (3.3h) and single review request indicate a smooth process; CI blocker was quickly resolved and the PR introduced a large code change but with minimal back‑and‑forth.

Summary

Count update checks in agent CLI

Improves efficiency by aggregating update checks, reducing network overhead.

Health Assessment

• • Rapid approval with minimal changes indicates straightforward enhancement.

Summary

Fix TUI resize issue on kitty

Resolves persistent black screen during terminal resize by disabling synchronized output in kitty, improving user experience.

Health Assessment

• • Quick fix with single commit and approval, minimal impact.

Summary

Stub engine worker for make test

Ensures Windows CI builds include required worker binary, preventing build failures.

Health Assessment

• • Fast turnaround, minimal changes, no blockers.

Summary

Refactor llama.cpp integration with worker process

Improves stability and performance by isolating the llama.cpp engine in a worker process and caching prompts across sessions, enhancing user experience.

Health Assessment

• • Extensive refactor with many iterations and thorough reviews indicates high complexity but ultimately delivered a stable feature.

Summary

Cap MCP tool output size to prevent context overflow

Limits tool-generated content before adding to conversation history, preventing model context exhaustion and giving users control via settings.

Health Assessment

• • Rapid turnaround with minimal review rounds indicates confidence in the fix; large code changes suggest significant impact on MCP tool handling.