claude-code

Self-looping adversarial-review-to-merge pipeline. Codex runs an adversarial review (a given area/topic, else the whole repo), findings are neutrally verified, each survivor becomes its own PR, and a per-PR review gauntlet (two fresh, context-isolated SATISFIED verdicts on the same PR content, reviewed one at a time over the whole diff) plus event-driven CI monitoring gate an auto-merge. Multiple isolated runs (each keyed by a run-id, with a lease so only one agent drives each) can run concurrently in one repo. Drives its own loop via ScheduleWakeup — invoke once, no /loop wrapper. Args: [--run id] [area or topic]

Evaluate and address GitHub Copilot PR review items for a GitHub pull request. Use when user provides a GitHub PR link and wants Copilot review comments checked, verified, fixed, committed, and summarized. Fetch review items with `gh`, verify each claim against source/tests before changing code, ask user before making subjective or constraint-driven changes, then work items one by one.

Continuously find and fix failing Go tests in a worktree, syncing with a parent branch between iterations. Use when user wants to fix all (or many) test failures in a branch via an automated loop. Args: <parent-branch> [test-args...]

Execute instructions in an isolated worktree, commit, fast-forward merge to parent, cleanup. Args: <parent-branch> <instructions...>

Delegate a task to Codex CLI via `codex exec`. Use for lightweight tasks (exploration, simple searches, file reads) that don't require heavy reasoning. Only available from Claude Code sessions.

Multi-agent design collaboration — Claude Code (reviewer) and Codex (designer) iterate on a design doc via file-based chat. Gathers requirements from user, then runs a review loop until design is satisfactory. Args: <topic>

Reorganize Go package file layout by actual responsibility. Inspects file contents, fixes naming mismatches, merges tiny files, splits oversized ones, consolidates tests, and regenerates generated files — all without changing behavior or public API. Args: <package-dir>

Two-pass code review focused on security, API consistency/symmetry, and user experience. Pass 1 is hostile (assume the worst, surface everything). Pass 2 is a neutral audit that confirms, adjusts, or refutes each finding. Use when the user asks for a hostile, skeptical, or hard review — not a friendly pass.

Run golangci-lint on Go code and fix any issues found. Use after writing or modifying Go code, or when the user explicitly requests linting.

Clean up local branches and worktrees that have been merged into a target branch (default: main). Args: [target-branch]

Discover Go test failures, partition by package/count, spawn N parallel test-fix-loop agents, verify results. Args: <parent-branch> [max-agents]

Split a large Go file into smaller files with logical groupings. Analyzes code structure and extracts cohesive units (types, functions, interfaces) into separate files within the same package.

Create a resumable checkpoint file for another agent to continue work later. Use when user asks to save progress, create handoff notes, checkpoint current state, or leave a resume file for a future agent/session.

Triage a single code-review comment. First verify the claim against source/tests; if invalid or not worth fixing, report the analysis and stop. If valid, enter plan mode, design the fix, and exit plan mode to get user confirmation before any edit.

Detect whether local branches/worktrees have been merged into a target branch (default: main), including squash merges. Args: [target-branch]

Reorganize Go test files — consolidate related tests into subtests, convert to table-driven style, shorten excessive test names, and reposition tests to match implementation ownership. Args: <package-dir>

Initialize .claude/docs/ and CLAUDE.md for a repository. Analyzes the codebase and creates agent-optimized documentation so Claude Code can navigate, modify, and test the project effectively.