Claude Skills by howar31

Commit the current change set with optional drift backfill (sprint mode) or as a generic Conventional Commits commit (standalone mode). Auto-detects which mode based on sprint context. Sprint mode verifies declared spec deltas (PLAN.md / SPEC.md `## Spec deltas`) against the actual diff before staging; missing or undeclared root-doc changes are surfaced interactively (lenient default; `--strict-deltas` makes them blocking). Falls back to Level 1/2 heuristics when no deltas are declared. Never...

Execute the next undone task(s) from a sprint's TASKS.md by dispatching the magi-developer subagent (Sonnet by default). Updates WORKS.md as a development log. Pauses for user confirmation before any commit. Default behavior auto-parallelizes tasks with disjoint file scopes; --parallel forces parallel, --sequential forces serial. Supports --model, --task, --milestone for fine-grained control.

Quick reference for magi-workflow. Bare invocation prints the full command roster, the standard workflow diagram, subagent roles, and common override flags, plus a state-aware "next step" hint when run inside a magi project. Pass a command name (e.g., /magi:help plan) to print details for one command extracted from its SKILL.md. Always allowed regardless of project state — this is the entry point users reach for when they don't remember which command to run.

One-time, idempotent project bootstrap. Detects which magi-workflow project files are missing (root CLAUDE/README/SPEC, magi/PRD/TECHSTACK/BACKLOG) and offers to scaffold them. Never overwrites existing files. Run once when adopting magi-workflow on a new or existing project. Safe to re-run.

Smart entry point for any commit-worthy work. Classifies the request by type (feat/fix/hotfix/refactor/chore/docs/perf/test/style/ci) and scale (trivial/minor/major), then routes to the right artifact (PLAN.md / SPEC.md / TICKET.md / HOTFIX.md / no-artifact). Multi-language semantic classification (zh-TW / en / mixed). User can override via flags or interactive confirm. Bare invocation reads magi/BACKLOG.md Pending entries.

Review the current git diff and produce DRIFT.md when a sprint context exists. **Mandatory before /magi:commit in sprint mode** — DRIFT.md is required input for the commit step's drift backfill flow. Default behaviour is multi-CLI MAGI cross-review (orchestrator + magi-consensus). Use --single to fall back to a single Opus reviewer (the magi-reviewer subagent) — saves tokens but loses cross-validation. Supports --magi <mode> override and --reviewers override. Never auto-fixes; always presents...

Run a multi-CLI MAGI review of a sprint's PLAN.md / SPEC.md. **Optional step** — can be skipped to save tokens if you trust the plan; human review of the doc is a valid alternative. Spawns reviewers in parallel via the orchestrator, then applies semantic dedup + weighted voting per references/MAGI_VOTING.md. Default reviewers and voting mode come from ~/.config/magi-workflow/config.json. Override with --reviewers and --magi.

First-run onboarding wizard for magi-workflow. Runs healthchecks on installed CLIs (claude/gemini/codex), records the user's preferred reviewer roster and weights into ~/.config/magi-workflow/config.json, and validates end-to-end with a small dry-run. Use --recheck to re-validate without resetting config; --reset to wipe and start over.

Terse "where am I, what's next" printer for magi-workflow projects. Outputs only the current project state, sprint dir (if any), suggested next command, and any active warnings — typically 3–6 lines. Use this when you're mid-flow and just need a quick orientation; reach for /magi:help when you want the full command roster, workflow diagram, and flag reference.

Decompose a confirmed PLAN.md or SPEC.md into a TASKS.md milestone+checklist file in the same magi/<num>-<name>/ folder. Coordinator-only — does not write production code. Pauses for user confirmation before allowing /magi:go to start.

Augment a sprint's SPEC.md with a Backend section (API contract, data model changes, authn/authz, validation, observability, contract test plan) tailored to the detected stack (Express/Fastify/Next/FastAPI/Django/Go/Rails/etc). Coordinator-only — does not write production code. Pauses for user confirmation. Run before /magi:tasks.

Produce a sprint CI.md covering pipeline stages, secrets handling, deployment strategy, and rollback for GitHub Actions / Cloud Build / GitLab CI / etc. Coordinator-only — produces drafts, never modifies live workflows. Pauses for user confirmation.

Augment a sprint's SPEC.md with a Frontend section (component tree, state, a11y, routing, performance budget, Playwright e2e plan) tailored to the project's stack (React/Vue/Svelte/Solid/Astro/RN). Coordinator-only — does not write production code. Pauses for user confirmation. Read before /magi:tasks so the test plan is captured in TASKS.md.

Produce a sprint INFRA.md covering Terraform / gcloud changes — dry-run plan, IAM diff, cost estimate, rollback. Coordinator-only — does not apply infra changes. Pauses for user confirmation. Bias toward GCP+Terraform but works with Pulumi / CloudFormation / Serverless / CDK.

Headless / walk-away mode. Two modes — fresh (`/magi:yolo "<description>"`) creates a new sprint and runs the full pipeline; resume (`/magi:yolo --resume`) continues an existing sprint from its current state. Both run plan/tasks/work/review-code/commit phases without user prompts, with conservative auto-decisions and an audit log at YOLO_LOG.md. Aborts and preserves intermediate state on first failure. --push optionally pushes after commit (refused on default branch unless explicitly allowed)...