Claude Skills

Documents architecture WITHOUT implementation. Creates ADR and component docs with status: provisioned. MUST use this skill when user wants to DESIGN or PLAN without implementing: - "provision X", "provision a component", "provision new service" - "design the architecture for X", "design a component" - "plan X", "plan a new service", "plan the architecture" - "document X architecture", "architect X" - "envision X", "sketch X architecture" DO NOT use this skill if prompt contains: - "pattern", "patterns" → route to c3-ref - "ref", "refs" → route to c3-ref - "convention", "standard" → route to c3-ref Key indicator: User wants documentation/design output, NOT code changes. Requires .c3/ to exist. Creates ADR with status: provisioned. For implementation requests, route to c3-alter.

Distribution and packaging patterns including shell completions, man pages, cross-compilation, and release automation. Use when preparing CLI tools for distribution.

Refactors existing skills by searching for better external implementations and adapting them. Use when improving skills with external best practices.

Process GitHub PR review comments by fetching them to local JSON, implementing fixes, and tracking progress. Use when user invokes /gh-process-review command. Fetches reviews to file to avoid context pollution, uses jq for parsing, commits after each fix. Supports optional "continue" argument to skip fetching and resume with existing reviews file.

Expert prompt engineering for creating effective prompts for Claude, GPT, and other LLMs. Use when writing system prompts, user prompts, few-shot examples, or optimizing existing prompts for better performance.

Creates Claude Code rule files for .claude/rules/. Writes conventions and shared guidelines that auto-inject into context. Use for path-specific coding conventions or shared project guidelines. NOT for constitution/laws (use <law> in CLAUDE.md instead).

Create distinctive, production-grade frontend interfaces with high design quality. Use this skill when the user asks to build web components, pages, or applications. Generates creative, polished code that avoids generic AI aesthetics.

Extract learnings from completed work. Trigger phrases: "run a retrospective", "extract learnings", "what did we learn", "capture lessons", "create a retro".

This skill should be used when the user asks to "generate documentation", "validate docs", "check doc coverage", "find missing docs", "create code-map", "sync documentation", "update docs", or needs guidance on documentation generation and validation for any repository type. Triggers: doc, documentation, code-map, doc coverage, validate docs.

Extract decisions and learnings from Claude session transcripts. Triggers: "extract learnings", "process pending", SessionStart hook.

Spawn Agent Farm for parallel issue execution. Mayor orchestrates crew via tmux + MCP Agent Mail. Replaces /crank for multi-agent work. Triggers: "farm", "spawn agents", "parallel work", "multi-agent".

Knowledge flywheel health monitoring. Checks velocity, pool depths, staleness. Triggers: "flywheel status", "knowledge health", "is knowledge compounding".

Expert in CLI/TUI design, command structure, visual design (colors, typography, icons), accessibility, and UX patterns. Automatically activates when designing new CLI tools, improving command interfaces, or reviewing CLI usability.

Elite iOS and macOS development expertise for Swift, SwiftUI, UIKit, Xcode, and the entire Apple development ecosystem. Automatically activates when working with .swift files, Xcode projects (.xcodeproj, .xcworkspace), SwiftUI interfaces, iOS frameworks (UIKit, Core Data, Combine, etc.), app architecture, or Apple platform development.

Résout automatiquement les erreurs PHPStan en analysant et corrigeant les problèmes de types. Boucle jusqu'à zéro erreur ou stagnation.

Use when bootstrapping a new project from scratch. Use when user says "create project", "new project", "initialize project", "start new app".

Use when creating rule files in .claude/rules/, adding project conventions, or scoping guidelines to specific paths. Use when user says "add rule", "create convention", "scope guideline". NOT for laws (use <law> in CLAUDE.md).

Use when creating specialized Claude Code subagents in .claude/agents/. Use when user says "create agent", "add reviewer", "specialized agent", "isolated context task".