vini-workflow

Detect the project stack and verify the hooks system is properly configured for this project.

Read-only code review agent. Use after sprint execution to verify quality, catch issues the executor might have missed, and check adherence to project patterns. Cannot modify files.

Task orchestration and sprint lifecycle management. Use when the user has a PRD with multiple sprints, when sprint coordination is needed, or when the user says "orchestrate", "run the sprints", "execute the PRD". Manages sprint delegation, coherence checks, and completion verification.

Executes a single sprint from a sprint spec file. Use when a sprint needs to be implemented in isolation. Receives sprint spec content from orchestrator or direct user invocation.

Adopt a PRD owned by another (typically dead) session — list adoptable plans in this repo, pick one, force-release any stuck sprint claims, and write a fresh active-plan pointer for THIS session. Auto-invoke when the user says "adopt plan", "take over plan", "resume someone else's plan", "my pointer is gone", or runs /adopt-plan. Do NOT auto-invoke on every fresh terminal — /plan-build-test Phase 0 already self-adopts when this session is the legitimate owner.

End-to-end hands-off run from PRD to staging. Chains /plan-build-test then /ship-test-ensure with CLAUDE_PIPELINE_MODE=staging-only,aggressive-fix-loop. Halts before production with exit 99 — the canonical success state for staging-only delivery. User verifies staging, then runs /ship-test-ensure manually for prod. Use for "ship the PRD to staging unattended", "fire and forget", or "autonomous staging".

Post-task learning capture and knowledge promotion. Auto-invoke when a task or sprint is completed, when the user says "done", "finished", "wrap up", or when all acceptance criteria are checked off. Do NOT invoke when user says "ship it" — that triggers /ship-test-ensure instead.

Creates a new project from scratch with a production-grade PRD. Use this skill whenever the user wants to start a new project, build a new product, create a new application, or says things like "new project", "start a project", "build me an app", "create a new API", "I have an idea for", "let's build something new". This is for greenfield projects — not for adding features to existing ones (use /plan for that). Runs a structured discovery interview, applies battle-tested architecture defaults, generates a full PRD with adversarial analysis, and outputs sprint-ready specs compatible with /plan-build-test.s

Local-only feature builder that plans, implements, and tests code using agent teams — all locally, no deploy. Use this skill when the user describes a problem, feature request, bug, or improvement to build. Triggers on phrases like "build this", "implement", "create feature", "fix this problem", "add functionality", or when user provides a feature description to execute. Also triggers when user wants to work through pending task files. After local testing completes, user manually verifies, then uses /ship-test-ensure to deploy.

Task planning and PRD generation. Auto-invoke when the user describes a new task, feature request, bug to fix (multi-file), refactoring need, or says "plan", "let's build", "I need", "implement", "create a PRD". Do NOT invoke for simple questions, conversations, or single-line fixes.

Stealth web browsing and content extraction for accessing the user's own content on sites with bot detection. Uses Patchright (Playwright fork) + Xvfb virtual display + headed Chromium. Use this skill when the user wants to access their own content on sites that block automated browsers, read their own blog posts, verify their own published pages, or interact with sites that misidentify automation as bot traffic. Also use when the user says "stealth browse", "get page content", "open this URL", "read this page", "check this site", or mentions bot detection.

Deep research via Stochastic Consensus & Debate. Spawns N researcher agents (sonnet) in parallel, each from a distinct angle, then synthesizes with a single opus agent. Use when the user asks "how should I...", "what's the best way to...", "compare approaches for...", or invokes /research explicitly. Do NOT trigger for: simple factual questions, code generation, task execution, or single-file fixes.

End-to-end shipping pipeline that commits, pushes, follows CI/CD deploy to staging, runs all E2E tests on staging, deploys to production, follows that deploy, then runs Google PageSpeed Insights and Lighthouse audits on production until all categories score 100 with zero errors. Iterates and fixes at every step. Use when the user says "ship it", "deploy everything", "ship and verify", "push to prod", "full deploy pipeline", or wants to go from committed code to verified production with perfect scores.

Mines the last 30 days of session transcripts for recurring friction patterns and emits a review directory of proposed edits to skills/rules/CLAUDE.md. Never modifies any skill, rule, agent, hook, or CLAUDE.md directly — outputs to `~/.claude/docs/skill-evolution-proposals/<run-id>/` only; the user reviews and applies manually.

Analyze the codebase and update project documentation (README.md, docs/) to reflect the current state of the code. Use this skill when the user says "update docs", "update readme", "sync docs", "docs are stale", "refresh documentation", or when workflow files changed and docs need updating. Also auto-invoke when check-docs-updated.sh blocks a push, or when the user wants to document a project for the first time. Works on any project — not just the workflow repo.

Verify a feature is live and healthy on staging without deploying. Hits each staging health endpoint, runs the project's Playwright smoke suite against the staging URL, checks the active PRD's acceptance criteria one-by-one with cited evidence, and flags any unexpected production auto-deploys. Use when the user says "verify staging", "check staging", "is it live yet", or after a PR has merged to staging and they want a quick sign-off before manual QA. Does NOT deploy anything — that's `/ship-test-ensure`.

Periodic self-audit of the workflow system. Reviews rule effectiveness, model performance, error patterns, and stale configuration. Use monthly or when the user says "audit workflow", "review the system", "check workflow health".

24 hooks across 8 events