roderik.github.io

Brainstorm and define a project as a PRD stored in Linear. Use when starting a new project idea, refining an existing project spec, or turning a vague concept into a structured PRD. Triggers: "brainstorm", "new project", "define a project", "write a PRD", "refine the spec", "project brainstorm", "I have an idea", "let's scope this out", "spec this", "plan a feature", "break this down into tickets".

Record browser and CLI flows as video and screenshots, upload evidence to Linear tickets and embed in GitHub PRs. Owns the self-learning composable flow system — step-by-step procedures for navigating the application that adapt when they break and get created when new flows are needed. Use when the user says "evidence", "record evidence", "capture proof", "test and record", or needs visual verification for PR reviews and CC-xx items. Also triggers when the testing skill's manual tier needs browser-based verification, when uploading screenshots/recordings to Linear or GitHub, or when you need to create, update, or execute application flows. If you're about to do manual browser testing, load this skill first.

Execute implementation work end-to-end — from a Linear ticket or an ad-hoc task. Handles the full lifecycle: status routing, workpad persistence, research-first planning, tracer-bullet implementation, TDD, self-verification, and state transitions. Triggers: "execute", "implement", "work on", "pick up ticket", "start PRD-xxx", or any non-trivial implementation task.

Research-first implementation planning with parallel expert review. Takes a task description (ticket, ad-hoc request, or PRD), runs codebase exploration, library/web research, codex second opinion, constitution check, and parallel expert review (feasibility, scope, security, architecture, devil's advocate, design, interaction). Produces a structured implementation plan with module mapping, anti-requirements, completion checklist, and verification protocol. Pure content producer — does NOT interact with Linear or any ticket system. Orchestrated by execute, commands, or other skills. Not user-invocable. Do NOT use for PRD creation (use prd-builder) or ticket decomposition (use task-builder) or verification (use verifier).

Build a structured PRD from a project idea through iterative brainstorming, research, and refinement. Outputs a complete PRD markdown document following the strict 6-section schema (including Agent Execution Context). Pure content producer — does NOT interact with Linear or any ticket system. Orchestrated by brainstorm, commands, or other skills. Not user-invocable. Use when a PRD needs to be drafted or refined from an idea, problem statement, or existing spec. Do NOT use for ticket decomposition (use task-builder) or Linear operations (use brainstorm).

Shepherd a PR to merge — resolve review comments and keep CI green until the PR is merge-ready. Use this skill whenever the user mentions PR issues, failing CI, red checks, review comments to address, "make CI green", "fix the build", "get this merged", or any variation of babysitting a pull request through to completion.

Review changed code for reuse, quality, efficiency, and security using six parallel review agents (reuse, quality, efficiency, security, cubic, adversarial), then fix any issues found. Use this skill whenever the user says "review my code", "check my changes", "sweep", "quality check", "code review", "clean up", or after completing a feature — even if they don't explicitly ask for a review. Also trigger when the user finishes a batch of edits and wants a sanity check before committing.

Decompose an approved PRD into milestones and detailed, self-contained task descriptions. Each task gets its own sub-PRD with implementation checklist, module mapping, anti-requirements, completion checklist, and verification protocol — detailed enough for a fresh Claude session to implement from the description alone. Pure content producer — does NOT interact with Linear or any ticket system. Orchestrated by brainstorm, commands, or other skills. Not user-invocable. Use when a PRD needs to be broken into implementation tasks. Do NOT use for PRD creation (use prd-builder) or Linear operations (use brainstorm).

Test-driven development with red-green-refactor loop. Use when user wants to build features or fix bugs using TDD, mentions "red-green-refactor", wants integration tests, or asks for test-first development.

Multi-tier testing skill for development quality gates. Four tiers: fast (lint+format+typecheck on affected packages), checkpoint (fast+unit tests), full (checkpoint+e2e on affected), and manual (delegates to evidence skill for browser-based verification with self-learning flows). Use during development for fast feedback, after implementation phases for checkpoint verification, before PRs for full verification, and as a command for manual browser-based testing. Also invoked by the verifier skill for functional CC-xx item verification. Do NOT use for code review (use sweep).

Execute completion checklist items (CC-xx), run sweep for code quality review, and run CI to verify an implementation is correct and complete. Uses agent-browser for UI verification, sweep skill for code review, and the testing skill's Full CI tier for build health. Reports pass/fail for every item. Pure verification — does NOT interact with Linear or any ticket system. Orchestrated by execute, commands, or other skills. Not user-invocable. Do NOT use for planning (use planner) or implementation. Use when code is written and needs verification before declaring done.