Taku

Execution reliability expert. ALWAYS invoke this skill after any coding skill completes its work. Do not accept completed work without running this guard first. Use this skill to audit step execution and catch silent omissions.

Execute an approved implementation plan. Triggers after /taku-plan, or on "build this", "implement the plan", "start coding", "run the plan", "execute tasks", "写代码", "开始实现", "执行计划", "开发吧", "开始写". Also handles worktree setup on "create a worktree", "isolated environment", "new branch workspace", "创建工作树".

Use after implementation is complete. Triggers after /taku-build. Analyzes diffs for security issues, bugs, and code quality. Run when asked to "review this", "check my diff", "code review", "审查代码", "看看有什么问题", "检查一下", "准备合并", "代码质量", or before shipping. Proactively invoke when the user is about to merge or land code changes.

Clarify ambiguous development requests, new feature ideas, product/design choices, and idea-stage work before planning. Three adaptive modes: Quick (small clear changes -> fast alignment), Design (feature requests -> architecture + DESIGN.md), Explore (new ideas -> forcing questions before committing). Triggers on "add a feature", "I have an idea", "should we build", "let's design", "设计一下", "需求分析", "我有个想法", or when the user describes a desired end state but implementation choices or success criteria are not yet settled. Also handles design system creation. Bug fixes and refactors use `/taku-debug` or `/taku-review` directly unless they need fresh design decisions.

Create a recoverable active-work brief for context-heavy coding, design, debugging, review, research, or handoff sessions. Multi-mode context-control habit with resume, handoff, debug, review, design, and research policies.

When the user reports a bug, crash, error, or anything that is broken or behaving unexpectedly, invoke this skill FIRST — before writing any code or proposing any fix. This skill is the mandatory entry point for ALL bug-fix tasks. The user expects you to investigate root cause systematically before fixing. Without this skill, you will likely fix the symptom instead of the cause, which is the #1 debugging failure mode. Use this skill whenever the user says anything like: "fix this bug", "crashes", "returns null", "500 error", "not working", "wrong result", "throws exception", "timeout", "silent failure", "was working yesterday", "broke after deploy", "sometimes fails", "only in production", "works locally but not in CI/staging", "CLI crashes", "config missing", "test fails", or describes any error/stack trace. Also: "调试", "找bug", "报错", "排查问题", "不工作了", "挂了", "崩了", "数据不对". Do NOT use for: lint fixes, dependency updates, adding features, refactoring, or any task that is NOT about fixing broken behavior.

Turn an approved design into an executable implementation plan. Three-step pipeline: scope + architecture review → design review (UI only) → plan writing. Auto-detects which step to start from based on project state. Invoke when DESIGN.md exists and is approved. Triggers on "write the plan", "create implementation plan", "plan this", "review this plan", "scope check", "architecture review", "design review", "is this plan buildable", "what could go wrong", "写计划", "制定方案", "规划实施", "架构评审", "这个方案可行吗", "有什么风险", or after /taku-think completes.

User-invoked reflection. Three modes: Learn (script-backed recording, searching, pruning, exporting, and optional bootstrap for user-approved patterns, pitfalls, preferences, discoveries), Retro (weekly engineering retrospective with git commit analysis, metrics, team breakdowns, trends), and Write Skill (codify recurring learnings into reusable skills). Triggers on "what have we learned", "add learning", "show learnings", "weekly retro", "what did we ship", "engineering retrospective", "write a skill", "create a skill", "codify this pattern", "总结一下", "学到了什么", "记录一下", "回顾这周", "做个retro", "写个技能", "把这个模式固化". Also invoke when the user expresses satisfaction or frustration after completing work — these are natural reflection moments.