claude-corps

Iterative review-fix loop for accumulated milestone/branch changes. Runs parallel reviewers, fixes findings autonomously, repeats until clean. Use after multiple tasks merge to a milestone branch, or before merging to main. Invoke with /milestone-review --base-branch main. Supports --dry-run and --max-iterations.

This skill should be used when the user wants a comprehensive code review using multiple specialized reviewers in parallel. Invoked with /multi-review or when user asks for 'thorough review', 'full code review', or 'review from multiple perspectives'. Use --plan <path> to review an implementation plan pre-coding.

Use when facing a bug, test failure, or unexpected behavior that isn't immediately obvious

Use when starting a new feature or idea that needs thorough pre-execution planning. Use instead of calling /product-review, /spec, and review skills individually.

Use when you have an idea, goal, or feature description and need to turn it into an actionable plan with tasks

Use when you need measured performance evidence by running a repeatable command on the current branch and a baseline ref

Use when you need 3-5 intentionally different UI directions before committing to a single design approach

Use when you need acceptance-first release confidence: validate acceptance criteria, regressions, and browser workflows before calling work ready

Autonomous dispatch-reconcile loop for batch task processing. Use with /auto-run --through <id> to execute tasks unattended. Requires Linear MCP or tasks to exist. Supports --resume for checkpoint recovery and --skip-milestone-review.

Use when implementation and tests are complete and you're ready to close out a task

Use at the start of an orchestrator session to understand project state and find available work

Review a worker session summary and reconcile tasks with implementation reality. Use after a worker completes /finish-task to sync the task board with what was actually implemented. Invoked by orchestrator sessions, or with /reconcile-summary <task-id>.

Start working on a task (claim, gather context, define acceptance criteria)

Use when multiple tasks are ready and you want to assign them to workers. Supports parallel (worktree-isolated) and sequential (direct on branch) modes.

Use for a read-only progress snapshot without committing, pushing, or closing anything

Interactive engineering plan review: architecture, code quality, tests, performance. One issue per question. Run after /spec before implementation, or standalone on any plan.

Product-taste review that challenges 'are we building the right thing?' Interrogation mode with recommended answers, assumption mapping (V/U/Vi/F), devil's advocate challenges, and alternatives analysis. Four modes: EXPAND / HOLD / REDUCE / DESIGN. Run before /spec or standalone.

Scored design plan review: 7 UI/UX dimensions rated 0-10, explains what would make each a 10, then fixes the plan. Run after /spec for UI-heavy features, or standalone on any plan with UI scope.

Use when the user wants to audit CLAUDE.md files for bloat, staleness, misplacement, or information architecture issues. Invoked with /claudemd-audit. Supports --project, --global, --all scopes and --fix for auto-remediation. Use periodically or before shipping to ensure CLAUDE.md stays lean.

Engineering retrospective analyzing git commit history for shipping metrics, work patterns, and trends. Persistent history with trend tracking. Use weekly to review shipping velocity, or after milestones. Invoke with /retro [7d|14d|30d|24h] or /retro compare to see trends.

Use when writing a new SKILL.md, revising an existing skill's instructions, or converting a repeated workflow into a reusable skill.

Identify and remove AI writing patterns to make text sound natural and human. Use when the user wants to 'humanize' text, remove AI-isms, or make writing sound less like ChatGPT.

Use when making any completion claim, reviewing findings, or processing agent reports