{"id":"5dbea74c-de40-4b8e-8870-90f5c6776d0c","slug":"craftpowers","kind":"skill","name":"craftpowers","summary":"","owner":"anhdt19942020","repoUrl":"https://github.com/anhdt19942020/craftpowers","license":"MIT","licenseTier":"permissive","trust":"community","stars":1,"match":null,"fewStars":true,"format":"skill","createdAt":"2026-04-28T03:57:40.000Z","addedAt":"2026-10-22T02:46:57.285Z","logoUrl":"https://avatars.githubusercontent.com/u/80743620?u=20b539a67af1fc54fed83406e71d4a689bca9324&v=4","linkOnly":false,"collection":null,"skillPath":null,"categories":["writing"],"tags":[],"runtime":"binary","local":true,"cloudCapable":false,"cloudCapability":"unknown","cloudReadyOnPlatform":false,"needsApiKey":false,"reviewed":false,"syncedAt":"2026-10-23T09:50:06.444Z","sample":false,"url":"https://conduits.dev/tools/craftpowers","language":"Python","transports":["stdio"],"install":{},"requirements":{"env":[],"upstreamMayCharge":false},"tools":null,"readme":"# craftpowers\n\n**A complete software development methodology for AI coding agents** — combining the battle-tested workflows of [Superpowers](https://github.com/obra/superpowers) with the craft-focused skills from [Matt Pocock](https://github.com/mattpocock/skills), plus automated security hooks and specialized agents.\n\n> *Your coding agent doesn't just write code — it thinks, plans, stress-tests, builds with TDD, reviews its own work, and ships safely.*\n\n---\n\n## What craftpowers does for you\n\nMost AI coding agents jump straight to writing code. craftpowers changes that. It gives your agent a **complete development process** — the same discipline a senior engineer follows, enforced automatically:\n\n- **Reads your docs first** — scans `docs/**/*.md` for existing specs, architecture notes, and feature documentation before brainstorming or planning, so it doesn't re-ask questions already answered or contradict decisions already made\n- **Thinks before coding** — asks what you actually need, stress-tests the design, explores multiple approaches\n- **Plans in small steps** — breaks work into 2-5 minute tasks with file paths, verification steps, and clear acceptance criteria\n- **Builds with TDD** — writes the test first, watches it fail, then makes it pass\n- **Reviews its own work** — dispatches specialized subagents for code review, security audit, and test quality\n- **Guards against mistakes** — blocks dangerous commands, detects hardcoded credentials, warns when context is running low\n- **Keeps docs in sync** — after implementation, updates existing documentation in-place to match the new code (no changelog bloat, just accurate docs)\n- **Ships cleanly** — verifies tests pass, chooses the right merge strategy, cleans up branches\n\n---\n\n## The Workflow\n\n```\n Idea → Brainstorm → Plan → Build → Review → Ship\n  │         │          │       │        │       │\n  │    stress-test   tasks   TDD    agents   verify\n  │    explore       PRD    code    secure   merge\n  │    ADR           issues  test   debug    cleanup\n```\n\n| Phase | What happens | Key skills |\n|-------|-------------|------------|\n| **Think** | Refine ideas, challenge assumptions, explore alternatives | brainstorming, adversarial-design, design-exploration |\n| **Plan** | Break into tasks, create PRDs, file GitHub issues | writing-plans, to-prd, to-issues |\n| **Build** | TDD cycle, subagent dispatch, parallel execution | test-driven-development, subagent-driven-development, executing-plans |\n| **Review** | Code review, security audit, debug, verify completion | code-review-workflow, systematic-debugging, verification-before-completion |\n| **Ship** | Merge strategy, cleanup, branch management | finishing-a-development-branch |\n| **Maintain** | Refactoring, pre-commit hooks, architecture records | structured-refactoring, architecture-decision-records |\n\n---\n\n## What's Inside\n\n### 57 Skills\n\nSkills are behavioral instructions that shape how your agent works. They activate automatically based on context or when you invoke them directly.\n\n<details>\n<summary><b>Think</b> — before any code</summary>\n\n| Skill | What it does |\n|-------|-------------|\n| **brainstorming** | Socratic design refinement — asks questions one at a time, proposes 2-3 approaches, validates design section by section |\n| **adversarial-design** | Stress-tests your design by relentlessly questioning every decision before you commit |\n| **design-exploration** | Spawns 3+ subagents to create radically different designs, then compares trade-offs |\n| **ubiquitous-language** | Extracts a DDD-style glossary from the conversation — standardizes domain language before coding |\n| **architecture-decision-records** | Captures architectural decisions as ADRs — finds improvement opportunities and documents the \"why\" |\n\n</details>\n\n<details>\n<summary><b>Plan</b> — after design</summary>\n\n| Skill | What it does |\n|-------|-------------|\n| **writing-plans** | Creates implementation plans with 2-5 minute tasks, exact file paths, code snippets, and verification steps |\n| **to-prd** | Converts conversation context into a structured PRD and submits it as a GitHub issue |\n| **to-issues** | Decomposes a plan into independent vertical-slice GitHub issues |\n\n</details>\n\n<details>\n<summary><b>Build</b> — during coding</summary>\n\n| Skill | What it does |\n|-------|-------------|\n| **test-driven-development** | Enforces RED → GREEN → REFACTOR. Deletes code written before tests exist |\n| **subagent-driven-development** | Dispatches a subagent per task with two-stage review (spec compliance + code quality) |\n| **executing-plans** | Loads the plan document and executes all tasks with checkpoints |\n| **dispatching-parallel-agents** | Runs independent tasks concurrently via parallel subagents |\n| **using-git-worktrees** | Creates isolated git worktrees for parallel feature development |\n| **nodejs-patterns** | Node.js best practices from Matteo Collina (Fastify creator): async, streams, error handling, testing, TypeScript type stripping, caching, profiling, graceful shutdown — 15 rule files |\n| **websocket-patterns** | WebSocket/Socket.IO engineering: rooms, namespaces, presence, scaling with Redis, sticky sessions, security, rate limiting, Fastify integration — 6 rule files |\n\n</details>\n\n<details>\n<summary><b>Review</b> — after coding</summary>\n\n| Skill | What it does |\n|-------|-------------|\n| **systematic-debugging** | 4-phase root-cause analysis: understand → hypothesize → trace → confirm |\n| **debug-flight-recorder** | Instrument with temp console.logs → repro → collect → auto-cleanup. Finds where bugs fail |\n| **verification-before-completion** | Requires fresh evidence (test output, screenshots) before claiming a task is done |\n| **code-review-workflow** | Pre-review checklist, dispatches code-reviewer subagent, structured response process, and automation-specific rules |\n\n</details>\n\n<details>\n<summary><b>Ship & Maintain</b></summary>\n\n| Skill | What it does |\n|-------|-------------|\n| **finishing-a-development-branch** | Verify tests → choose merge/PR/keep/discard → execute → clean up worktree |\n| **structured-refactoring** | Martin Fowler approach: interview → tiny commits → GitHub issues |\n| **setup-pre-commit** | Installs Husky pre-commit hooks with lint-staged, type checking, and tests |\n| **git-guardrails-claude-code** | Adds hooks to block dangerous git commands before execution |\n| **writing-skills** | TDD-based process for creating and testing new skills |\n| **using-man** | Meta skill — teaches the agent how to discover and use all other skills |\n\n</details>\n\n---\n\n### 13 Agents\n\nSpecialized subagents that are dispatched automatically when relevant, or invoked directly.\n\n| Agent | Role | Purpose |\n|-------|------|---------|\n| **debugger** | `debugger` | 5-phase systematic root-cause analysis — complex bugs, intermittent failures, regressions |\n| **codebase-explorer** | `codebase-explorer` | Read-only repo scout — maps files, patterns, conventions, and duplicate risks before planning |\n| **test-engineer** | `test-engineer` | Test coverage and quality review — gaps, brittle patterns, weak assertions |\n| **release-prep** | `release-prep` | Pre-deploy gate — audits env vars, migrations, breaking changes, drafts changelog |\n| **doc-writer** | `doc-writer` | Generate or update documentation — README, API refs, inline comments (runs on Haiku) |\n| **code-reviewer** | `code-reviewer` | Code review against the implementation plan and coding standards |\n| **journal-writer** | `journal-writer` | Brutally honest failure log — records bugs, setbacks, and lessons to `docs/mankit/journal/` |\n| **implementer** | `implementer` | Implements ONE task from a plan — writes code, runs tests, commits, reports |\n| **quick-fix** | `quick-fix` | Surgical 1-2 file edit — typo fix, single-function rewrite, mechanical rename |\n| **secure-reviewer** | `secure-reviewer` | OWASP Top 10 security audit — injection, auth, crypto, access control |\n\n---\n\n### Hooks (automatic protection)\n\nHooks run automatically in the background — no action needed from you.\n\n| Hook | Trigger | What it does |\n|------|---------|-------------|\n| **session-start** | Every new session | Injects craftpowers context so the agent knows its skills are available |\n| **security-gate** | Before any Bash command | Blocks dangerous commands: `rm -rf`, force-push, `DROP TABLE`, `chmod 777`, `curl \\| bash`, fork bombs, and 11 more patterns |\n| **credential-scanner** | After every Write/Edit | Scans written code for hardcoded credentials (AWS keys, API tokens, private keys, passwords) with false-positive filtering |\n| **context-tracker** | Every user message | Estimates context window usage and warns at 70% (yellow) and 87% (critical) so you can `/compact` before losing context |\n\n---\n\n### Commands\n\nQuick-access shortcuts for common workflows:\n\n| Command | What it does |\n|---------|-------------|\n| `/man-cook` | End-to-end implementation: scout → plan → code → review → test → finalize |\n| `/man-assess` | Interactive quiz to discover the right workflow for your situation |\n| `/man-debug` | Start systematic debugging for a bug |\n| `/man-trace` | Instrument code with temp logs to trace bug execution flow |\n| `/man-loop` | Manage a tracked agentic loop (`start` / `status` subcommands) |\n| `/man-release` | Pre-deploy audit: env vars, migrations, breaking changes, changelog |\n| `/man-update` | Update man: pull latest → reinstall → verify |\n| `/man-check` | Health check: verify all hooks, agents, permissions are working |\n\n---\n\n## Installation\n\n### Option 1: CLI (recommended)\n\n```bash\nnpm install -g mankit\nmankit init\n```\n\nThat's it. One command to install, one to setup.\n\n**Update anytime:**\n```bash\nmankit update\n```\n\n**Health check:**\n```bash\nmankit doctor\n```\n\n### Option 2: Plugin install\n\n```bash\n/install-plugin https://github.com/anhdt19942020/craftpowers\n```\n\nThen run the full setup:\n\n```bash\npython scripts/install.py\n```\n\n### Option 3: Clone directly\n\n```bash\ngit clone https://github.com/anhdt19942020/craftpowers ~/.claude/plugins/man\npython ~/.claude/plugins/man/scripts/install.py\n```\n\n### What install.py does\n\nThe install script is **idempotent** — safe to run multiple times. It configures:\n\n1. **Hooks** — writes security-gate, credential-scanner, and context-tracker to `~/.claude/settings.json`\n2. **Permissions** — adds ~70 safe permission rules (git read ops, test runners, build tools) to reduce interruption prompts\n3. **Agents** — links `~/.claude/agents/` → `craftpowers/agents/` (junction on Windows, symlink on Unix)\n4. **Commands** — links `~/.claude/commands/` → `craftpowers/commands/`\n5. **`.claudeignore`** — creates a template to exclude build outputs, dependencies, logs, and lock files from context (saves tokens every session)\n\n### Verify your setup\n\nAfter installing, restart Claude Code and run:\n\n```\n/man-check\n```\n\nThis runs 22 checks across 6 categories: installation, settings, hook scripts, hook paths, agents, and hook smoke tests. Every check should show `[PASS]`.\n\n### Updating\n\n```\n/man-update\n```\n\nPulls the latest code, re-runs install, and verifies everything works.\n\n---\n\n## Platform Support\n\n| Platform | Status | Notes |\n|----------|--------|-------|\n| **Claude Code** | Full support | Skills, hooks, agents, commands |\n| **Copilot CLI** | Partial (skills + tool mapping) | See `references/copilot-tools.md` |\n| **Cursor** | Planned | Not yet implemented |\n| **Gemini CLI** | Planned | Not yet implemented |\n| **Codex** | Planned | Not yet implemented |\n| **OpenCode** | Planned | Not yet implemented |\n\n---\n\n## Token Optimization\n\ncraftpowers includes several built-in token-saving features:\n\n- **`.claudeignore`** — auto-generated by `install.py`, excludes `node_modules/`, `dist/`, `*.log`, lock files, and other low-signal content from context\n- **Permission allowlists** — ~70 pre-approved rules eliminate permission prompt round-trips\n- **Context checkpoints** — executing-plans and subagent-driven-development auto-compact at natural breakpoints (every 3 tasks or after long review loops)\n- **Context tracker hook** — warns at 70% and 87% context usage so you can `/compact` before auto-compaction\n- **Session summary** — reports token usage and RTK savings at the end of each workflow\n\n### Recommended companion tools\n\nThese external tools stack with craftpowers for additional savings:\n\n| Tool | Savings | What it does |\n|------|---------|-------------|\n| [RTK](https://github.com/rtk-ai/rtk) | ~60% CLI output | Compresses command output before it enters context |\n| [Context Mode](https://github.com/mksglu/context-mode) | ~98% tool output | Routes large outputs to sandbox, only summaries enter context |\n| [MCPlex](https://github.com/ModernOps888/mcplex) | ~97% tool definitions | Hides MCP tools behind 3 meta-tools, loads on-demand |\n| [Caveman](https://github.com/JuliusBrussee/caveman) | ~65% output | Enforces terse, filler-free responses |\n| [Headroom](https://github.com/chopratejas/headroom) | ~87% compression | Lossless compression of everything the agent reads |\n\nCombined stack: **~58% total savings** — makes the 200K context window function like ~350K.\n\n---\n\n## Philosophy\n\n- **Design before code** — stress-test ideas before committing to an implementation\n- **Test-Driven Development** — write the test first, always\n- **Systematic over ad-hoc** — process beats guessing\n- **Evidence over claims** — verify with real output before asserting completion\n- **Craft over vibe** — deliberate technique, not vibe coding\n\n---\n\n## Quick Reference\n\nSee [`QUICK_REFERENCE.md`](./QUICK_REFERENCE.md) for a one-page lookup of all skills, agents, commands, and hooks.\n\n---\n\n## Credits\n\ncraftpowers is built on the foundation of:\n\n- **[Jesse Vincent](https://blog.fsck.com)** & **[Prime Radiant](https://primeradiant.com)** — creator of [Superpowers](https://github.com/obra/superpowers), the methodology and workflow foundation\n- **[Matt Pocock](https://github.com/mattpocock)** — creator of [skills](https://github.com/mattpocock/skills), the craft-focused toolset for software quality\n- **[Matteo Collina](https://github.com/mcollina)** — creator of [mcollina/skills](https://github.com/mcollina/skills), Node.js best practices (Fastify creator, Node.js TSC member)\n- **[Jeff Allan](https://github.com/Jeffallan)** — creator of [claude-skills](https://github.com/Jeffallan/claude-skills), WebSocket engineer skill\n\nIf craftpowers is useful to you, please consider:\n- Star [Superpowers](https://github.com/obra/superpowers)\n- Star [mattpocock/skills](https://github.com/mattpocock/skills)\n- Star [mcollina/skills](https://github.com/mcollina/skills)\n- [Sponsor Jesse Vincent](https://github.com/sponsors/obra)\n\n---\n\n## Acknowledgements\n\n- 4 skill (frontend-ui-engineering, browser-testing-with-devtools, api-and-interface-design, source-driven-development) + 4 reference checklist (security, performance, accessibility, testing patterns) imported from [addyosmani/agent-skills](https://github.com/addyosmani/agent-skills) (MIT). See `THIRD-PARTY-LICENSES.md`.\n\n## License\n\nMIT License — see [LICENSE](./LICENSE) for details.\n"}