SkillAtlasSkill 详情

executing-plans

Your landlord kept your deposit.

审核状态:已审核Quality 80Security 100

复制安装命令

用 Codex 或 Claude 安装复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它先审查 Skill 页面再帮你安装。

复制前请先查看来源、License 和安全提示。

项目 README

来源文件:README.md

抓取于 2026年9月29日

🧠 PM Skills — 1176 Professional Agent Skills for Claude, ChatGPT, Gemini, Cursor, Codex & Hermes

PM Skills — 1176 professional skills your AI assistant can read. Plain markdown, works with Claude, ChatGPT, Gemini, Cursor, and Codex. MIT licensed.

Quick start · Playground · All skills · Packs · Install guide · Changelog · Roadmap

Stars Skills Version npm In the official Anthropic plugin directory SkillCheck Security Audit License

Your landlord kept your deposit. Your mom got a medical bill that makes no sense. You got laid off on a Tuesday. Someone you love died, and no one handed you the checklist.

Generic AI is a very confident intern. PM Skills is the senior colleague's notes — 1176 of them, one markdown file each, for the moments at work and in life where "it depends" is not an answer. (PM stands for Professional, not just Product Management. Yes, we get asked.)

PM Skills is an open-source library of 1176 Agent Skills — plain-markdown SKILL.md files that teach an AI assistant to do one professional task to a senior professional's standard, from writing a PRD to decoding a lease or running a blameless postmortem. Each skill bundles the framework, an output template, quality checks, and anti-patterns. It is MIT-licensed and works with Claude, ChatGPT, Gemini, Cursor, and Codex.

1176
skills
134
bundles
35
professions
12
platforms
4.8 / 5
eval-scored
0
runtime · telemetry · accounts
MIT
forever

🧭 How it works

StepExample
01Say what you need, in your own words"My landlord is keeping my deposit."
02Your assistant loads the one skill that matchessecurity-deposit-recovery, a single markdown file, read only when it is relevant
03You get the finished document, not advice about itThe challenge to each deduction, the demand letter, and the point at which small claims is worth it

🚪 Pick a door — 30 seconds each

You…Do this
▶just want to see itopen the Playground and run a skill in your browser. No install, no signup, no "enter your email to continue"
🧠use Claude Code/plugin → search pm-skills → install. Ask "decode this lease" and watch
🛠use anything elsenpx pm-claude-skills add and pick your tool (Cursor, Codex, Windsurf, ChatGPT, Gemini…)
🔎don't know what to ask fortype it at find — "my landlord kept my deposit" — and it names the skill
🎒are in the middle of somethingstart from your moment, not the catalogue → Skill Packs · 🍼 new parent · 💼 laid off · 🌍 new country · 👵 caring for a parent · 🕊️ losing someone · 💸 money in crisis

Nothing here can scare your setup. A skill is a markdown file your AI reads. No runtime, no telemetry, no accounts. Installing copies text files; uninstalling is deleting them. Skeptical? Good instinct — read one first, it's written for humans too.

Works with
AssistantsClaude Code · ChatGPT · Gemini · Cursor, Codex, Windsurf · any MCP client
Where you already workTelegram · Slack · Raycast · Obsidian · n8n
For buildersPython · Hugging Face dataset · Docker · GitHub Actions · decision layer

If it can read a markdown file, it can use a skill.

✅ PM Skills is right for you if

  • ✅ You keep retyping the same long prompt and get a different answer each time
  • ✅ You want the finished document (a PRD, a demand letter, a postmortem), not an essay on how to write one
  • ✅ You use more than one assistant and want the same skills in all of them
  • ✅ You want to read exactly what your assistant was told before you trust the result
  • ✅ You are dealing with something outside your field: a lease, a medical bill, a layoff
  • ✅ You want nothing to host, no account, and no telemetry unless you switch it on

🥊 Without a skill vs. with one

You sayGeneric AIWith the skill
"help me with my lease"600 words on the importance of reading leases carefully🔴 clause 14 auto-renews you into a full year · 🟡 deposit terms written to fail · the two sentences to send back — lease-decoder
"write the PRD"a template with [insert goal here] in itproblem, users, requirements, metrics, the open questions — scored 0–40 against its own rubric before you see it — prd-template
"should we ship Friday?""There are several factors to consider…"ship 0.18 · ship_reduced 0.71 · slip 0.11 — and the one fact that would flip it — ship-or-slip
"practice my salary negotiation""Great question! Confidence is key."a hiring manager who pushes back, then an out-of-character debrief on what you gave away — salary-negotiation

✨ See it in action

The Skill Playground: pick the Executive Update skill, fill in a few notes, hit run, and watch a structured executive briefing stream out — all in the browser
▶ Pick a skill → fill a short form → run it → a senior-grade artifact streams out. Your key stays in your browser, or run free with no key.

Galaxy 3D — fly through all 1176 skills as a glowing constellation you orbit and click into
🌌 Galaxy 3D — all 1176 skills as a constellation. The ones you've run burn brighter. Zero productivity value, 100% recommended.
PM Skills Wrapped — your practice turned into a shareable, Spotify-Wrapped-style story
🎁 Wrapped — your practice as a shareable story. 100% local; nobody learns you ran saying-no-kindly eleven times.

🆕 What's new

ReleaseWhat it addsTry sayingRead more
v80.2.0 trust in a releaseThe MCP install command fixed, a check that every release installs and runs, announcements in doctor and the playground, and frozen versions of the flagship skills that a team can pin"Run the doctor."CHANGELOG.md · skill releases
v80.1.0 design tasteA five-stage UI pipeline (direction, system, motion, critique, verification) that stops generic AI-looking interfaces, with pointers to five specialist design skills"Run the design pipeline on this landing page."pm-design-taste · credits
v80.0.0 the promote loopScans your own transcripts for what you keep asking, drafts it as a skill, tests the triggers, publishes it"What do I keep asking you for?"PROMOTE-LOOP.md
v79.0.0 the decision layerTyped questions with a probability per option: which skill, is this input safe, ship or slip"Should we ship Friday?"JEV-DECISION-LAYER.md

The full write-up of the three feature releases, with examples: docs/WHATS-NEW.md. Everything older: CHANGELOG.md.

💬 What can I ask it to do?

Say it in your own words. The description does the routing:

🏠 "decode this lease before I sign" → lease-decoder📋 "write the PRD for our referral feature" → prd-template🚨 "blameless postmortem for Friday's outage" → incident-postmortem
💰 "practice my salary negotiation" → salary-negotiation📉 "why is churn up this quarter?" → churn-analysis⚖️ "rank the backlog with RICE" → rice-prioritisation
🛂 "prep me for the visa interview" → the-visa-interview🔨 "is this contractor quote fair?" → home-contractor-quote-decoder🏡 "should we rent or buy?" → rent-vs-buy
📝 "draft my self-review honestly" → performance-review🚢 "ship Friday or slip a week?" → ship-or-slip♻️ "what do I keep asking you for?" → promoter-scan

…all 1176 live in the catalog.

⚡ Quick start

You want to…Do this
BrowseSKILLS.md · the searchable web catalog
Install in Claude Code/plugin → search pm-skills (official Anthropic directory) — or npx pm-claude-skills add --agent claude
Install in Cursor / Codex / Windsurf / Cline…npx pm-claude-skills add --agent cursor (or codex, windsurf, aider, cline, zed…)
Use one skill in ChatGPT / Geminicopy from exports/chatgpt/ or exports/gemini/ and paste as instructions
Skills over MCP, in any sessionclaude mcp add pm-skills -- npx -y -p pm-claude-skills pm-claude-skills-mcp
Route a prompt to a skillnpm run route -- "board meeting on Thursday"

No npm install needed — npx pm-claude-skills … always runs the latest. Per-tool instructions: docs/installation.md.

📚 The skills

Every skill follows the same discipline: what it produces, the inputs it needs, a real framework (severity scales, decision rules, not vibes), a concrete output template, quality checks, and anti-patterns. All 1176 pass the SkillSpec L3 gate and a security audit in CI.

Decoders bundle crestSimulators bundle crestCalculators bundle crestLive data bundle crestCowork bundle crestTokens bundle crestSeatbelt bundle crestEssentials bundle crest
DecodersSimulatorsCalculatorsLive dataCoworkTokensSeatbeltEssentials

Browse all 1176 → · try one in your browser → · 134 bundles →

Every category, with examples

For everyone — life's paperwork and decisions

FamilyWhat it doesExamples (of many)
🔍 Decoders (25+)Read the document before you sign it — plain language, 🔴🟡🟢 severity, the money mathlease · medical bill · job offer · severance · insurance policy · contractor quote · timeshare
🎭 SimulatorsFace the adversary early — the real meeting, then an out-of-character debriefsalary negotiation · promotion committee · thesis defense · visa interview · due-diligence call
🧮 CalculatorsDeterministic Python scripts + honest models — assumptions labeled, no false precisionrent vs buy · FIRE number · debt payoff · raise vs jump · daycare vs stay-home
📡 Live data (17)Real-time answers with zero API keys — weather, rates, flights, scores, all over plain curlweather · currency · crypto · flights · earthquakes · is-it-down
🏠 Life adminThe unglamorous logistics, done in orderrelocation · new parent · caregiving · doctor visits · records requests
💼 Career momentsThe weeks that decide yearslayoff kit · resignation kit · PIP response · first 90 days as manager · interview gauntlet
🏛 Dead mentors (5) 🆕History's sharpest operators, resurrected — the real methods from public-domain classics, applied to modern workMachiavelli on office politics · Sun Tzu on picking your fights · Franklin's decision algebra · Marcus Aurelius on bad days · Bennett's 1908 time audit
🏛 Life systems (20) 🆕Navigating the bureaucracies and emergencies people face alone — civic, disability, immigration, disastervoting-navigator · disability-benefit-appeal · arrival-setup · credential-recognition · go-bag-builder · after-the-disaster
🧠 Human edges (20) 🆕The parts of life nobody built tools for — neurodivergence, invisible illness, grief, identity, the hard conversationsmasking-budget · spoon-planner · diagnosis-limbo-kit · coming-out-rehearsal · grief-admin · rabbit-hole-rescue
⚡ New-gen (10) 🆕How the next generation lives and earns — creator deals, clips, D&D, ranked, resale, the attention warcreator-deal-decoder · clip-factory · ttrpg-session-forge · the-vibe-check · ranked-climb-coach · attention-reset
🔮 2027 (10) 🆕Problems you don't have yet, but will — the agent era's operational skillsagent-severance · deepfake-drill · agent-hiring-panel · context-bankruptcy · clone-brief · api-for-yourself · the-org-simulator
🎲 Tabletop (5) 🆕Game night, upgraded — teach, judge, plan, design, and practice the tradesteach-the-game · rules-lawyer · game-night-planner · board-game-designer · tabletop-negotiator
🧾 Freelance & renters & parentsSmall bundles for specific livespricing your services · late invoices · deposit recovery · IEP meetings · students
🎲 Hobbies (12) 🆕Life outside work — the genuinely fun stuffwine pairing · houseplant care · board-game night · D&D campaign · stargazing · chess openings
💪 Wellbeing (12) 🆕Body and mind, sustainably — not another app streakhome workout · sleep reset · habit builder · posture reset · screen-time detox
🔐 Digital self-defense (12) 🆕When your digital life is under attackidentity-theft recovery · phishing triage · account recovery · data-broker removal · doxxing response
👪 Family & relationships (12) 🆕The people who matternew-baby logistics · wedding vows · co-parenting messages · condolences · in-law boundaries
💭 Thinking modes (24) 🆕Change how your AI reasons — escape the generic answer, stress-test decisionsthe-third-answer · five-minds · decision-panel · red-team-my-plan · devils-advocate · poke-holes-in-this
🎯 Focus & executive function (26) 🆕Get unstuck and run your own brain — ADHD-friendly, for everyonewhere-do-i-start · task-to-first-step · overwhelm-triage · the-one-thing · build-my-memory-file · weekly-unstuck
📖 Learning & mastery (10) 🆕Learn anything faster and make it sticklearn-anything-roadmap · feynman-explainer · spaced-repetition-setup · skill-plateau-breaker · deliberate-practice-plan
💰 Wealth-building (10) 🆕Build wealth on purpose — educational, not financial adviceinvesting-for-beginners · index-fund-starter · ask-for-a-raise · first-100k-plan · financial-independence-roadmap
📑 Investing literacy (5) 🆕Read a company's accounts before you believe its highlights — every figure page-referenced; educational, not financial adviceannual-report-tutor · audited-boundary-check · profit-to-cash-walk · company-compare-same-definitions · investing-vocabulary-explainer
🤝 Social & relationships (10) 🆕The hard conversations and the human onesmake-friends-as-an-adult · networking-for-introverts · boundary-setting-scripts · give-hard-feedback-kindly · repair-after-a-fight
🩺 Caregiving & aging (10) 🆕Care for aging parents and navigate the system — not medical/legal advicemedical-appointment-advocate · care-team-coordinator · caregiver-burnout-check · long-term-care-options · end-of-life-wishes-conversation
🤖 AI-native life (10) 🆕Use AI itself well — the meta-skills that make every tool betterprompt-library-builder · delegate-to-ai · ai-context-primer · spot-ai-mistakes · get-more-from-ai
🤝 Cowork (100)The office knowledge work an AI coworker actually does — the frameworks — the whole bundleemail triage · spreadsheet audit · meeting cost meter · deck outline first · saying no kindly · delegation brief
⚡ Cowork · Live (12)The same jobs, done — Claude Cowork acts on your real data via connectors + sandbox and returns an artifact — the whole bundleinbox triage (live) · meeting prep (live) · spreadsheet audit (live) · deck from doc · thread → decision · PR description (live)

For professionals — 35 fields

Product Management Engineering Marketing & GTM
Customer Success Data & Analytics Leadership & People
Design & UX Legal Finance
Founders Security Government

…plus HR, sales, operations, research, healthcare, educators, writers, social media, and more — the full profession index, or by bundle in plugins/ (134 bundles). Install any bundle: /plugin install pm-decoders@pm-skills.

Meta

Before installing anyone's skills (including these): skill-vetting — a security read for SKILL.md files. The library's own standard lives in SKILLSPEC.md; every skill's level is enforced in CI.

💭 pm-thinking — think better

For when the model is too correct and gives you the average answer:

🎯 pm-focus — get unstuck

ADHD-friendly executive function, useful for every brain:

🔍 What does a skill look like?

A skill is a single markdown file with a name, a description that tells the assistant when to activate it, and a body containing the working framework: required inputs, decision rules or severity scales, a concrete output template, quality checks, and anti-patterns. The assistant reads it and gains the judgment; humans can read, audit, and edit the same file. No runtime, no lock-in.

---
name: lease-decoder
description: "Decode a residential lease into plain English and rank the
  clauses that can hurt you. Use when someone asks 'what am I signing'…"
---
## Framework: Severity Scale
- 🔴 Can cost you real money — auto-renewal into a full new term, break
  penalties beyond re-rental costs, deposit conditions written to fail…

That's the whole trick. It's markdown. Your agent reads it and gains the judgment; you can audit it, edit it, or write your own.

✅ Quality, not just quantity

Structureevery skill passes the SkillSpec L3 gate — framework, quality checks, anti-patterns — on every commit
Evidenceeval-scored: 208 outputs, avg 4.8/5, judged blind. The benchmark report publishes the negative findings too
Trustrisk tiers on every skill · an expert-review program with a harm-ranked queue · a vendor-neutrality gate · a security audit in CI
Honestydecoders end with a not-advice line, calculators name what they don't model, simulators debrief out of character, and skills that shouldn't ghostwrite (your kid's college essay) coach instead

🎁 Beyond the skills

Explore📄 one-page cheatsheet · 📸 gallery · 🏛 anti-pattern museum (2,900+ rules) · 📖 the handbook (also a printed book) · 🎯 skill router
Runworkflow recipes · journeys (a pack as one guided session) · subagents & slash commands · MCP server + REST API · n8n / Slack / Obsidian · the Boardroom
Savepm-tokens — 30–60% off a session's token flow, stdlib Python, nothing leaves your machine (how-to, 🪙 dashboard) · npx pm-claude-skills mcp-audit — what your MCP servers charge you in rent
Provenpx pm-claude-skills prove --skill ./my-skill --tasks tasks.txt — paired A/B with real token counts and a sha-pinned receipt, because "65% better!" is not a measurement · SkillBench · route-bench
Workpm-cowork (100 skills for office knowledge work) · pm-seatbelt (agent safety pre-flight) · Org Edition · 🇪🇸 🇫🇷 🇨🇳 🇯🇵 translations · offline bundle
Lint your own skills in CI
- uses: mohitagw15856/pm-claude-skills@v79
  with:
    path: .claude/skills   # optional — it finds them otherwise

Checks frontmatter, the Use when … trigger clause a model actually matches on, leftover template text, and structure — and annotates each finding inline on the PR diff. Also npx pm-claude-skills skillcheck. Zero dependencies, no model call.

Companion tools for the bits a skill shouldn't guess — both MIT, zero-dependency, no model call: notugly (contrast and design-system checks the design skills call instead of estimating) and rulebook (37 games, 203 rulings, for the board-game skills).

🚫 What PM Skills is not

Not an agent framework.Skills are instructions your existing assistant reads. There is no runtime to host.
Not a prompt pack.Every skill has required inputs, an output format, quality checks and anti-patterns, and is gated in CI.
Not only for product managers.PM stands for Professional. Most of the library is life admin and 35 professions.
Not a substitute for a professional.The legal, medical and financial skills prepare you for that conversation. They do not replace it.
Not tied to one vendor.The same skill exports to 12 platforms and is served over MCP.
Not a hosted service.The playground and the MCP endpoint are optional. Everything works from files on your machine.

❓ Straight answers

Is it actually free? Yes — MIT, all 1176 skills, forever. The skills are markdown; there is nothing to gate. Sponsors fund the playground's free model runs, not access.
Do I need an API key? Not to browse, read, install, or use skills inside a tool you already have. The playground serves a few sponsor-funded free runs a day. A key only matters for optional extras: running skills from CI, or the typed decision layer (which falls back to keyword routing without one).
The catalog says 1176 but the folder has more. Which is it? Both. There are 1188 directories under skills/; 12 of them are deprecated (marked in their frontmatter, each pointing at the skill that replaced it). They stay on disk so an old install command or a bookmarked name never breaks, and they are hidden from the catalog, the playground and the headline count. 1176 is what a person can browse; 1178 is what an installer can resolve.
I'm not a product manager. Is this for me? PM stands for Professional here. Most of the library is decoders for leases and medical bills, salary-negotiation practice, career-moment kits, life admin, and 35 professions from teaching to veterinary. The product-management corner is just where it started.
Will this mess with my existing setup? No. Skills are inert text files in a folder; your assistant reads them when relevant. Remove the folder and it's like they were never there.
How do I know these are any good? Every skill passes a structural gate (SkillSpec L3) and a security scan in CI; 208 outputs are eval-scored in the open (avg 4.8/5), and the benchmark report publishes the negative findings too. When something is machine-translated or unscored, it's labelled.

🗺️ Roadmap

  • ✅ Exports to 12 platforms and one install command
  • ✅ MCP server, subagents and slash commands
  • ✅ Quality gates in CI: structure, security, duplicates, drift
  • ✅ The decision layer and the promote loop
  • ✅ The design taste pipeline
  • ⚪ Published eval scores for more skills
  • ⚪ A promotion round for the Production-Ready tier
  • ⚪ Merging the near-duplicate skills still marked as pending
  • ⚪ Translated skill descriptions, done at scale or retired

Direction, not a contract. Details and good first issues: ROADMAP.md.

🔒 Privacy and telemetry

Off by default. A skill is a text file and sends nothing.

WhatDefaultDetail
Skills, CLI, exportsNothing is sentPlain files on your machine
Usage counterOffOpt in with PM_SKILLS_TELEMETRY=1. It sends the skill name and nothing else
Playground and hosted MCPOnly if you use themYour prompt goes to the model provider you choose

The full statement, including the handler you can read: docs/TELEMETRY.md and SECURITY.md.

🤝 Contributing

The library grows a skill at a time — plant one of your own. One markdown file, one PR.

Add a skill via PR (the standard, CONTRIBUTING), request one via issue, or publish your own repo to the community index. Translations follow skills-i18n/.

❤️ Support

If a skill saved you real money or a real mistake, star the repo — it's how others find it. Sponsors fund the playground's free runs and get naming rights, not influence: become a sponsor. Organisations: a clearly-disclosed logo and link only, never product integration — email me.

📄 License

MIT — use them, fork them, ship them at work. Skills are judgment, and judgment wants to be free.


Built by Mohit with Claude. 1176 skills · 134 bundles · 35 professions · every commit gated. The long version — every feature, wave, and frontier bet — lives in the Showcase.

内容与创作Agent / MCP / Skill 创作

低风险

  • 来源需自行核对维护者身份。
  • 未检测到明显脚本安装指令。
  • 未检测到明显外部权限要求。
  • 未检测到高风险命令。
  • 扫描发现:0 条。

Codex — Git Clone 安装

  1. 安装前请先查看来源仓库和风险报告。
  2. 克隆仓库:git clone https://github.com/mohitagw15856/pm-claude-skills.git
  3. 将 "skills/executing-plans" 文件夹复制到 Codex 的 skills 目录中。
  4. 重启 Codex 让新的 skill 生效。

Codex — 手动复制安装

  1. 安装前请先查看来源仓库和风险报告。
  2. 从源仓库下载 SKILL.md 及相关文件。
  3. 在 Codex 的 skills 目录中创建新文件夹。
  4. 将所有 skill 文件复制到新文件夹中。
  5. 重启 Codex 让新的 skill 生效。

Claude Code — Git Clone 安装

  1. 安装前请先查看来源仓库和风险报告。
  2. 克隆仓库:git clone https://github.com/mohitagw15856/pm-claude-skills.git
  3. 将 "skills/executing-plans" 文件夹复制到 Claude Code 的 skills 目录中。
  4. 重启 Claude Code 让新的 skill 生效。

Claude Code — 手动复制安装

  1. 安装前请先查看来源仓库和风险报告。
  2. 从源仓库下载 SKILL.md 及相关文件。
  3. 在 Claude Code 的 skills 目录中创建新文件夹。
  4. 将所有 skill 文件复制到新文件夹中。
  5. 重启 Claude Code 让新的 skill 生效。

Cursor — Git Clone 安装

  1. 安装前请先查看来源仓库和风险报告。
  2. 克隆仓库:git clone https://github.com/mohitagw15856/pm-claude-skills.git
  3. 将 "skills/executing-plans" 文件夹复制到 Cursor 的 skills 目录中。
  4. 重启 Cursor 让新的 skill 生效。

Cursor — 手动复制安装

  1. 安装前请先查看来源仓库和风险报告。
  2. 从源仓库下载 SKILL.md 及相关文件。
  3. 在 Cursor 的 skills 目录中创建新文件夹。
  4. 将所有 skill 文件复制到新文件夹中。
  5. 重启 Cursor 让新的 skill 生效。

GitHub Copilot — Git Clone 安装

  1. 安装前请先查看来源仓库和风险报告。
  2. 克隆仓库:git clone https://github.com/mohitagw15856/pm-claude-skills.git
  3. 将 "skills/executing-plans" 文件夹复制到 GitHub Copilot 的 skills 目录中。
  4. 重启 GitHub Copilot 让新的 skill 生效。

GitHub Copilot — 手动复制安装

  1. 安装前请先查看来源仓库和风险报告。
  2. 从源仓库下载 SKILL.md 及相关文件。
  3. 在 GitHub Copilot 的 skills 目录中创建新文件夹。
  4. 将所有 skill 文件复制到新文件夹中。
  5. 重启 GitHub Copilot 让新的 skill 生效。

Windsurf — Git Clone 安装

  1. 安装前请先查看来源仓库和风险报告。
  2. 克隆仓库:git clone https://github.com/mohitagw15856/pm-claude-skills.git
  3. 将 "skills/executing-plans" 文件夹复制到 Windsurf 的 skills 目录中。
  4. 重启 Windsurf 让新的 skill 生效。

Windsurf — 手动复制安装

  1. 安装前请先查看来源仓库和风险报告。
  2. 从源仓库下载 SKILL.md 及相关文件。
  3. 在 Windsurf 的 skills 目录中创建新文件夹。
  4. 将所有 skill 文件复制到新文件夹中。
  5. 重启 Windsurf 让新的 skill 生效。
查看 SKILL.md 原文
name: executing-plans
description: "Execute a written plan with discipline — verify each step before advancing, surface deviations instead of improvising around them, and keep a visible execution log. Use when working through a plan (yours or another agent's), resuming multi-session work, or when execution keeps drifting from what was agreed. Produces completed work plus an execution log showing what matched the plan, what deviated and why, and what the plan got wrong. Pairs with writing-plans."

Executing Plans Skill

A plan's value is realised or destroyed at execution time. The two failure modes: rigid execution (following a plan reality has invalidated) and drift (quietly improvising until the work no longer resembles the plan and nobody decided that). The discipline is the same for both: deviations are decisions, made visibly.

What This Skill Produces

  • The work, executed step-by-step with per-step verification actually run
  • An execution log: step → result → verification outcome → any deviation with its reason
  • A plan feedback note: what the plan got wrong (feeds the next plan)

Execution Method

  1. Load the plan and check it's still true. Before step 1: do the plan's assumptions still hold (the branch, the data, the constraint)? A plan written yesterday can be stale today; two minutes of validation beats an hour of executing a fiction.
  2. One step, then its verification — actually run. The verification isn't decoration: run the command, check the observable, record the result. Advancing on "that probably worked" is how step 6 fails mysteriously because step 3 silently didn't.
  3. Classify every divergence out loud. When reality disagrees with the plan, stop and classify:
    • Plan-preserving detail — the plan's intent holds, the mechanics differ slightly → note it in the log, continue
    • Plan deviation — the approach must change for this step → amend the plan visibly (strikethrough + new step), state why, continue
    • Plan invalidation — a stop condition hit, or the goal itself is now wrong → STOP; report; replan with the human before another line of work The cardinal sin is treating an invalidation as a detail because stopping feels like failure.
  4. Respect the stop conditions absolutely. They were written calm; you are now in flow and biased toward momentum. The "must not do" list doesn't bend for convenience — if it should, that's a visible plan amendment, decided, not slid into.
  5. Checkpoint on schedule. At each planned checkpoint (or ~every 30-45 min of work): where am I vs the plan, what's the log show, is the remaining plan still right? Multi-session work ends each session with a state note: done through step N, next action, open questions — the resume beats re-derivation.
  6. Close with the feedback loop. At completion: run the plan's DONE test (not your feeling of doneness). Then write the plan feedback: which estimates were off, which risk fired, which verification caught something. Plans improve only if execution reports back.

Output Format

Per step (in the log): Step N: [action] → [result] · verify: [check run → outcome] · [deviation? classified + reason]

On completion:

Execution report: [plan name]

Done test: [the plan's test → passed/failed, evidence] Deviations: [each, with classification and reason — or "none"] The plan was wrong about: [feedback for the next plan] Follow-ups discovered (not done, not forgotten): […]

Quality Checks

  • Every step's verification was actually executed, result recorded
  • Every divergence was classified (detail / deviation / invalidation) — none absorbed silently
  • Stop conditions were honoured; any override was a visible, stated decision
  • Completion was declared by the plan's done-test, not by fatigue
  • The plan-feedback note exists

Anti-Patterns

  • Do not improvise around a broken plan — amend it visibly or stop; silent drift is unaccountable work
  • Do not skip verifications when steps "obviously worked" — the mysterious step-6 failure was born at step 3
  • Do not push through a stop condition on momentum — it was written calm precisely because you wouldn't be
  • Do not declare done without running the done-test — feeling-finished and being-finished diverge exactly when it matters
  • Do not end a session without the state note — re-derivation is the tax on every resumed task

发现问题?提交给管理员复核

评分:

评论 (0)

暂无评论,成为第一个评论者吧!