mathias7799/MaaawKit

orchestration

Decompose work and delegate to subagents, bridge workers, and fleets effectively. Use for large multi-part tasks, parallel research, codebase-wide analysis, swarm audits, independent workstreams, or when the user mentions agents, subagents, parallelization,…

原文语言：英语

agent-bridge

Delegate bounded tasks to other agent CLIs (Codex, Gemini, Copilot, Cursor, opencode) through the MaaawKit bridge engine — prepared-by-default jobs, isolated worktrees for write modes, structured results, guard policy enforced on every command.

原文语言：英语

agent-handoff

Hand work off to a different agent CLI (Codex, Gemini, Cursor, opencode) — or receive work from one — with rules, verification commands, project memory, and selected prompt contract intact. Use when user says "hand off to codex/gemini", "continue in another…

原文语言：英语

cross-model-prompting

Write prompts that work when delegating to a different model family or agent runtime via bridge/handoff. Use whenever composing bridge tasks, worker prompts, AGENTS.md-style guidance, or instructions another model will consume.

原文语言：英语

coding-standards

Production coding standards for .NET/C#, PowerShell, TypeScript (React/Next.js), and Python. Use this skill whenever writing, refactoring, or reviewing code in any of these languages — even for "quick scripts" or small fixes. Also use when scaffolding new…

原文语言：英语

session-handoff

Preserve working state across sessions by writing a structured handoff (.agent/handoff/HANDOFF.md + handoff.json) before context runs out or the session ends. Use when a task will span multiple sessions, when context is getting long, when the user says…

原文语言：英语

cross-review

Get a second model's independent review of work this session produced (or review another agent's work here) — blind, structured, and adjudicated rather than averaged. Use when the user wants "a second opinion", "have codex review this", cross-model review…

原文语言：英语

rules-sync

Keep every agent's instruction file (AGENTS.md, CLAUDE.md, .cursor/rules, copilot-instructions, GEMINI.md, .windsurfrules) compiled from ONE canonical source — .agent/rules.md + detected repo facts + promoted memory. Use when the user asks to sync/update…

原文语言：英语

codebase-audit

Structured audit of a codebase at a configurable depth (quick smoke pass, standard, or deep) — security, correctness, architecture, dependencies, tests, operational readiness — producing a prioritized findings report. Use when the user asks for an audit,…

原文语言：英语

deep-thinking

Structured planning before non-trivial implementation. Use whenever a task involves 3+ files, a new feature, an architectural decision, a migration/refactor, ambiguous requirements, or anything the user calls "complex", "important", or "production". Also use…

原文语言：英语

verification-loop

Iterative verify-fix loops for driving work to actually-done. Use when running long autonomous tasks, fixing test suites, achieving "all tests green", satisfying a spec checklist, or whenever the user says "keep going until it works", "loop", "don't stop…

原文语言：英语

vibe-to-prd

Turn a vague idea, vibe, voice-note-style ramble, or "I want an app that kinda..." into a sharp PRD optimized for agentic coding (Claude Code can build directly from it). Use when the user describes a product/feature loosely, asks for a PRD/spec/requirements…

原文语言：英语

memory-and-learning

First-class project memory — capture lessons, decisions, and repo facts as schema-valid records in .agent/memory/ via the maaaw engine, so every future session (of any agent) starts smarter. Use PROACTIVELY whenever the user corrects you, a debugging session…

原文语言：英语

codebase-documenter

Generate or refresh documentation for an existing codebase — README, architecture overview, CLAUDE.md, onboarding guide, module docs, ADRs. Use when the user asks to document a repo, write/update a README, explain the architecture, create onboarding docs,…

原文语言：英语

debugging

Systematic debugging methodology. Use whenever something is broken, failing, crashing, flaky, "not working", producing wrong output, or when tests fail — including build errors, runtime exceptions, CI failures, performance problems, and "it works on my…

原文语言：英语

grill-me

Adversarial interrogation of the user's plan, design, architecture, PRD, estimate, or code — stress-testing assumptions before reality does. Use when the user says "grill me", "challenge this", "poke holes", "red-team this", "play devil's advocate", "am I…

原文语言：英语