dev

Pick one code-review wave, run it in an isolated git worktree, apply every member fix, run QA, merge it back, and close only fully completed waves

Runs every open code-review wave concurrently in isolated git worktrees, lands the run on main as a single PR through a merge lock, and reports which waves closed and which parked

Reviews Rust code for idioms and ownership, error handling, concurrency and async soundness, performance and unsafe code, OWASP security, test quality, and NATS/JetStream patterns. Use while writing or editing Rust as an implementation guardrail, or run a formal review that files one backlog task per finding.

Triage backlog tasks in Triage status, group them semantically, record each wave's file scope, create a code-review-plan-waveN parent task, and flip grouped tasks to To Do

Reviews React and TypeScript code in Vite SPAs for hooks and type safety, async error handling, rendering performance, accessibility, XSS and Web Crypto security, and test quality including socket.io real-time patterns. Use while writing or editing frontend code as an implementation guardrail, or run a formal review that files one backlog task per finding.

Review unstaged/staged changes and git log, then generate a shell script that groups related files into conventional commits, optionally on a branch that ends in a pull request

Surveys a Rust workspace's build cost without touching the tree, reading its profiles, cargo config, gates, nextest config and dependency graph, taking sccache stats and a warm cargo build --timings, and files one backlog task per finding, each with its measured cost, the date and machine load, and a safe or trade-off classification. A cold build is opt-in via --measure-cold, into a target directory on real disk, never tmpfs. Passing --apply writes the safe fixes into Cargo.toml profiles, .ops.toml gates and .config/nextest.toml; trade-offs such as dependency opt-level are never applied. Use when a Rust workspace's builds, tests or gates feel slow, or before tuning its profiles by hand.

Runs Clippy at pedantic strength over a clean checkout without touching the source tree, then files one backlog task per warning, each labelled pedantic, and reports a high-level effort estimate for clearing them. Test-only style findings, generated files, and out-of-tree warnings are dropped, and a lint firing more than twenty times in one crate becomes a single aggregate task. Passing --apply additionally writes the lint policy into Cargo.toml and clippy.toml; without it the run only shows what those files would contain. Use when a Rust project should be held to stricter lint levels than its current configuration enforces.

Processes external Rust content (URLs, articles, pastes), evaluates it against existing Rust skills and project agents, and integrates approved knowledge into the right files. Use to learn from external sources and keep Rust guidance current.