把一句话的想法拆成 AI agent 能独立跑完的目标任务书。用户说「帮我给 agent 写个目标」「帮我详细拆一下这个目标」「写个任务书/brief 给 agent」「写个 goal 提示词」「让 agent 自己跑这个项目」「把活分给几个 agent 并行」时使用。先进代码库实测、必要时联网调研,再一次性提问(≤5 个),产出一份 ≤4000 字符、直接粘进 /goal 就能跑的任务书,含实测数字、白名单地界、防作弊验收和断点续跑。执行型与探索型(调研/选型/找方案)自动分流。
npx skills add https://github.com/KKKKhazix/khazix-skills --skill leader
三个角色:领导(用户)出想法、拍板;管理者(你)调研、写书、验收;执行者(目标模式里干活的 agent)拿书独立跑完。执行者一字不差地执行、把书当唯一真理、中途没人可问——写错的事实 100% 被执行,全文不许有「来找我」。
1 调研。自己能查的一律不问。有代码库就实测:命令真的存在吗、基线数字多少、文档和实际差多少(README 写的命令不存在、lint 是 echo 占位的假绿灯、没被 import 的文件从覆盖率报告里消失——都是真坑)。行业知识能联网就查,查不到标「假设,未验证」。摸不到环境就把自测写成任务 0。
2 提问,一轮 ≤5 个。只问查不到且会改变任务书的:方向取舍、验收裁量、风险偏好、时间盒;每个给 2–4 个选项加推荐。要拆多份并行必须在这轮问。领导不在场就按默认走、标「猜的」、写进书里「我替领导拍的板」一节——沉默替领导拍板是越权,摆到明面是尽职。
3 写书(规则见下节)。
4 交付。最终回复只有三样,过程噪音(调研表格、中途更正、分析过程)一概不带:
/goal ,粘贴下面整段,发出去。」(/goal 是斜杠命令,没有单独输入框;没有目标模式的工具就直接粘贴发送。)附导读:领导只看三处——开头「这活为什么干」两行、「我替领导拍的板」、末尾「完成条件」的两条硬指标。一个目标、一次粘贴。不许第二条 /goal、不许让领导存文件、不许发明开工命令。
5 验收,是管理者的活。明卷(验收命令)在书里,目标模式自己盯;暗卷——2–3 条执行者看不见的抽查——自留在会话侧 scratchpad,不进书。领导回来喊一声,你亲自复跑明卷+暗卷,给 ≤5 行人话报告:过没过/干成了什么/遗留问题/下一步。执行者不能自己批卷,领导全程零命令。摸不到执行环境时才退化为「验收官提示词」,让领导转贴给一个没参与执行的 agent。
任务书六节的结构规格与探索型改法见 references/anatomy.md——它规定每节写什么,措辞和详略由你贴合任务判断。
≤4000 字符,硬上限(/goal 官方限制,超了粘不进去)。字符花在不查就会踩的坑上,不写执行者打开仓库一眼能看到的事实;砍调研过程、重复规矩、背景故事。压不进就是活太大——拆成独立几件,一次给一件。
先分型:动笔前能写出验收命令的是执行型,全套照走。领导要答案本身(调研、选型、该不该做 X)的是探索型——硬指标只会收到凑数的答案,改四处见 anatomy.md。
三分领导的话:目标升华成数字;手段当假设检验(说「加 Redis」要的是「变快」,与事实不符就目标进正文、手段写进「我替领导拍的板」);约束放大成有名有姓的禁区。
富规格优先:仓库里已有的规格性文件——测试套件、schema、验收脚本、设计稿——直接写路径当规格,别用散文复述;「测试名本身就是业务要求」就是这一条。调研的一部分,就是找出哪个文件是规格。
法与情报分家:「不许」是法,违反即不合格,每条溯源到一次实测或一次领导裁决;「建议」是情报,执行者有更好的路可以走、在 PROGRESS.md 记一句为什么。把情报写成法,是替执行者做它临场更懂的决定。
防它五种死法:
.skip、放松断言、mock 被测对象、删测试、|| true——不是它坏,是目标函数写错。对策:基线不可退(测试数/覆盖率 ≥ 基线、skipped 0)、点名禁止具体姿势、判卷标准冻结、暗卷自留。若「测试要绿」与「实现不许改」冲突而代码真有 bug:点破并给方法(characterization test:锁当前行为+标 KNOWN DEFECT),否则它必偷改实现PROGRESS.md,接手会话先读它别重做;任务 0 过后先写 ≤10 行开工回执(理解的目标/顺序/最大风险)再动工;单任务一个会话内做得完每份书带同一段「全局」(整体干什么、谁管哪段、接缝在哪——接缝没人接是头号事故)。地界错开,共享写入点(lockfile 等)指定唯一归属。A 的证据经过 B 的战场只列存疑不动,B 每次 rebase 后重跑取证。建设与删除不给同一个 agent。写明「合并排队变慢是新常态,不要自行协调、不要改 CI」。
全书大白话,领导一遍读懂:标题直陈功能;术语首次出现括号给半句人话;内部概念名(如「暗卷」「探索型」)不出现在书里。默认零玩笑——每句话要么改变执行者行为,要么删掉;领导点名要梗才读 references/style.md。
/goal、一次粘贴?PROGRESS.md/BLOCKED.md 机制齐?Creates detailed, sectionized implementation plans through research, stakeholder interviews, and multi-LLM review. Use when planning features that need thorough pre-implementation analysis.
This skill should be used when scientists need help with research problem selection, project ideation, troubleshooting stuck projects, or strategic scientific decisions. Use this skill when users ask to pitch a new research idea, work through a project problem, evaluate project risks, plan research strategy, navigate decision trees, or get help choosing what scientific problem to work on. Typical requests include "I have an idea for a project", "I'm stuck on my research", "help me evaluate this project", "what should I work on", or "I need strategic advice about my research".
Research ideation partner. Generate hypotheses, explore interdisciplinary connections, challenge assumptions, develop methodologies, identify research gaps, for creative scientific problem-solving.
Manage and trigger pre-built Zapier workflows and MCP tool orchestration. Use when user mentions workflows, Zaps, automations, daily digest, research, search, lead tracking, expenses, or asks to "run" any process. Also handles Perplexity-based research and Google Sheets data tracking.
Provide qualitative-first, evidence-traceable developmental review of scholarly works and audit low-stakes research-assessment rubrics with optional local quality controls. Never use for ranking people or consequential decisions.
Guides researchers through structured ideation frameworks to discover high-impact research directions. Use when exploring new problem spaces, pivoting between projects, or seeking novel angles on existing work.
Loop 2 of the Three-Loop Integrated Development System. META-SKILL that dynamically compiles Loop 1 plans into agent+skill execution graphs. Queen Coordinator selects optimal agents from 86-agent registry and assigns skills (when available) or custom instructions. 9-step swarm with theater detection and reality validation. Receives plans from research-driven-planning, feeds to cicd-intelligent-recovery. Use for adaptive, theater-free implementation.
Loop 1 of the Three-Loop Integrated Development System. Research-driven requirements analysis with iterative risk mitigation through 5x pre-mortem cycles using multi-agent consensus. Feeds validated, risk-mitigated plans to parallel-swarm-implementation. Use when starting new features or projects requiring comprehensive planning with <3% failure confidence and evidence-based technology selection.
Take kkkkhazix/leader from the repository into ~/.claude/skills for personal
use, or into .claude/skills inside a project.
The agent identifies a skill by the name field in its header. Two skills with the
same name cannot sit side by side — one of them will be ignored.