skill-master

Hands off completed work to QA by merging the PR into the integration branch, posting a testing guide on the Fizzy card, moving it to the QA column, assigning Elvis, and syncing qa-mirror. Auto-detects per-repo conventions (integration branch, qa-mirror presence, Supabase migrations). Use after a PR is approved (or pushed direct) and ready for QA testing.

Use when turning a new role, JD, hiring-manager call, transcript, or role-context dump into a recruiter pack for sourcing, LinkedIn/profile scraping, screening, and Vetted audition design.

Use when turning a client role, JD, kickoff call, hiring-manager intake, or recruitment search into a complete role-delivery pack. Applies to VFA/VettedAI/partner hiring work, including role calibration, internal role briefs, candidate-facing JDs, outreach emails, work-sample auditions, scoring rubrics, shortlist workflows, and live working docs.

Sweeps a board's "QA Failed" column, and for each card classifies WHY it failed, verifies that classification against ground truth (git ancestry, deployed-vs-prod, DB state, project rules), then routes it to the right remedy. Most QA-Failed cards are NOT "the code is wrong, re-review it" — they're stale checkouts, environment drift, deploy gaps, or rule-misreads. Use when a batch of cards is stuck in QA Failed and you want to clear them efficiently as the senior engineer.

Optimized for Human+AI Agent workflows. Converts high-level product intent into technical tickets that include file anchors, logic constraints, and verification protocols for coding agents.

Cross-repo engineering productivity analysis with bounty estimation. Use when the user wants contributor stats, PR velocity, workload distribution, team performance snapshots, or bounty payout projections.

Triage the "Not Now" (postponed) lane of a Fizzy board — list the auto-postponed cards, classify each (shipped / stale-noise / still-open / human-call), prove "shipped" via target-repo main-ancestry, then close the resolved ones, resurface the still-open ones to a triage column, and leave the ambiguous ones for a human. Use when a board's Not Now lane has accumulated and needs cleaning, or the user says "triage not now", "clean up the postponed cards", "triage the not now section".

Opens or updates a staging→main PR covering all Fizzy cards in the "Merge to Prod" column, audits git vs the column (flags shipped cards for closure and premature cards for move-back), and drafts a terse batched PR title/body. Auto-detects integration/target branches and Fizzy board per repo. Use when a batch of tickets has cleared QA + manual UX testing and is ready to ship to production.

Interact with Fizzy (37signals) project management boards via REST API. Use when the user wants to create, read, update, or close Fizzy cards, add comments, check board status, create new boards, or file bugs/features.

Single entry point for all skill distribution — sync local edits back to skill-master and propagate to all repos, pull latest from skill-master, or copy skills to new repos. Use when the user edits a skill locally, says "sync skills", "update skill-master", "pull skills", or "distribute skill".

Use after bug fixes or corrections to update .claude/rules/ with learned anti-patterns and add a one-liner to the CLAUDE.md index table. Invoke after fixing mistakes to build project memory. Every run also prunes stale rules so the set stabilizes instead of growing unbounded.

Map a Fizzy board into a clear state-of-play, surface the real chaos (floating cards, stale artifacts, duplicate/overloaded work, dependency chains, scope-stale epics), then triage and split with the user's sign-off. Use when the user wants to "get back up to speed", "map out the board", "clean up Fizzy", reduce board chaos, or plan who-owns-what for a sprint/weekend. Repo-agnostic — works on any repo with a corresponding Fizzy board.

Reviews GitHub PRs against Fizzy tickets to verify an engineer delivered what was asked. Fetches PR diff + Fizzy card, runs ticket compliance + regression check, prints verdict, and optionally posts to Fizzy. Single PR or bulk (review a whole "PR Open" column / all open PRs in one parallel pass). Use when reviewing PRs from engineers before merging.

Use when you need silent product-demo videos of a real web app without paid tools, manual screen-recording, or a human — scripts a browser through a feature and renders an mp4. Triggers on "demo video", "feature demo", "record the app", "show this feature", "make a walkthrough", or producing many feature demos at once. For Vite/React/Supabase apps (the VettedAI + Congrats stack); the engine is framework-agnostic.

Generates high-quality, predictable, and efficient skills based on user requirements. Use when the user wants to create a new skill.

Reference for writing, editing, and pruning skills and rules well — the vocabulary and failure modes that keep them lean and predictable.

Pressure-tests an implementation plan or technical approach through a senior full-stack engineer + senior tech lead panel — critique, revised plan, then impact analysis — and prints a clean final plan. Use when reviewing a plan after grooming-architect or ticket-review, mid-hotfix, or before building something new, and the user says "tech review", "pressure-test this plan", "review this approach", or "stress-test this plan".

Merges a staging branch into main while excluding specific tickets that haven't passed QA. Handles QA-fix ticket detection (which implies parent feature exclusion), already-in-main deduplication, dry-run conflict checking with proper cleanup, and safe execution via temp branch. Works across all three repos (vettedai-audition, vetted-congrats, backend-restructing).

Thin wrapper around Basecamp's `fizzy` CLI (https://github.com/basecamp/fizzy-cli) for surface area the existing `/fizzy` skill does NOT cover: cross-board search, file uploads, and multi-account auth juggling (Vetted ↔ Congrats Fizzy contexts). For ALL other Fizzy actions (cards, comments, columns, board/column IDs, user IDs, engineering board template, column-move via POST /triage.json, comment HTML formatting, etc.) defer to `/fizzy`.

Inspects, diagnoses, and resolves application issues using a systematic, hypothesis-driven approach. Use when the user asks to fix a bug, debug an error, or troubleshoot a system.

Reviews Fizzy tickets with a senior eng + PM + designer panel, then posts critique and revised ticket as an HTML comment. Use when reviewing grooming tickets for technical accuracy before engineering picks them up.

Codifies the P0 hotfix flow — base off main (skipping QA), open a PR to main, then cherry-pick the merge commit back to the integration branch so staging doesn't drift. Confirms P0 status before going main-route. Use when there's a live production incident that cannot wait for the standard staging→QA→main pipeline.

Use when a user shares a Figma file URL and wants every (or several) frame translated to static HTML/CSS design references under docs/design-references/. Produces browser-openable specs with per-folder READMEs and a design-system translation guide so engineers don't copy reference CSS into production. Triggers on keywords like "figma", "design references", "translate the designs", or any figma.com/design URL with multiple frames.

Use when a batch of engineering work is complete. Generates team-readable release notes OR polished community-facing newsletter content from git history. Supports internal (default), newsletter, or both output modes.

Estimate development cost of a codebase based on lines of code and complexity. Use when the user wants to know what the codebase would cost to build from scratch.

Verifies pull requests are safe to merge by detecting regressions, destructive patterns, and breaking changes. Use when reviewing PRs from junior engineers, before merging feature branches, or when you need to verify a PR doesn't break existing functionality.

Mandatory grounding source for database operations. Use this before writing SQL, creating Supabase migrations, or querying projects, opportunities, or audition_answers to prevent schema hallucinations.

Compares API endpoints between any two repositories by auto-discovering their frameworks (Express, Supabase, etc.) and identifying missing or changed routes.

Performs deep-dive, cross-repository audits to produce high-density technical manifests ("The Ground Truth"). Analyzes codebase to map API bridges, state management, worker patterns, and security usage. Use when auditing distributed systems or generating architectural documentation.

You MUST use this before any creative work - creating features, building components, adding functionality, or modifying behavior. Explores user intent, requirements and design before implementation.

Analyzes downstream impact of code changes BEFORE implementation to prevent regressions. Use when modifying shared functions, hooks, APIs, database schemas, types, or any code with multiple consumers. Prevents the "hammer everything" myopia of AI agents.

Use when you have a spec or requirements for a multi-step task, before touching code

Safely recovers write-only environment variables from Supabase by deploying a temporary edge function. Use when the user needs to recover lost secrets from a Supabase project.

Analyzes user problems to recommend existing skills from the repository or design new ones based on available resources. Use when the user wants to solve a problem but doesn't know which skill to use, or when proposing a new skill.

Use when executing implementation plans with independent tasks in the current session

Expert AI consultant for Supabase architecture, Realtime, Auth, and Webhooks.

Verifies API behavior by comparing a target API against a source API or a defined contract. Use when migrating backends, switching providers, or regression testing endpoints.