writing-skills
Skill0 starsChanged 36 days agoApache-2.0View on GitHub
Use when writing, editing, reviewing, or pruning a skill (SKILL.md) — the craft reference for making skills predictable: invocation choice, description writing, information hierarchy, leading words, and the named failure modes (premature completion, duplication, sediment). Also use for the periodic pruning pass over ~/.claude/skills/.
191 other projects use a version of this.
Adopted by
Most recently taken up by
- AtilaMedeiros/skillstheir own version, first seen Nov 16, 2026
- Floppa2003/skillstheir own version, first seen Nov 16, 2026
- dragoshont/architravetheir own version, first seen Nov 16, 2026
- rundesk-ai/skillstheir own version, first seen Nov 15, 2026
- draekien/skillstheir own version, first seen Nov 15, 2026
- mintunitish/tandemtheir own version, first seen Nov 15, 2026
Used with
Projects with it also use chinese-code-review, chinese-commit-conventions and 10 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
Leodorareluctant259Skillchinese-code-review
中文代码审查规范——在保持专业严谨的同时,用符合国内团队文化的方式给出有效反馈
Also in 8 projects that use this
Leodorareluctant259Skillchinese-commit-conventions
中文 Git 提交规范 — 适配国内团队的 commit message 规范和 changelog 自动化
Also in 8 projects that use this
obraSkilldiagnosing-superpowers
Use when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.
Also in 20 projects that use this
kntismSkillcondition-based-waiting
Use when tests have race conditions, timing dependencies, or inconsistent pass/fail behavior - replaces arbitrary timeouts with condition polling to wait for actual state changes, eliminating flaky tests from timing guesses
Also in 3 projects that use this
aaddrickSkillwriting-agents
Use when creating new agents, editing existing agents, or defining specialized subagent roles for the Task tool
Also in 3 projects that use this
pedronauckSkilllesson-learned
Analyze recent code changes via git history and extract software engineering lessons. Use when the user asks 'what is the lesson here?', 'what can I learn from this?', 'engineering takeaway', 'what did I just learn?', 'reflect on this code', or wants to extract principles from recent work.
Also in 3 projects that use this
- +3 more from Leodorareluctant259
Other versions191 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 191
- obra/superpowers★ 314ksame name
- PostHog/posthog★ 41ksame name
- jnMetaCode/superpowers-zh★ 8.5ksame name
- openai/plugins★ 7.8ksame name
- farm-fe/farm★ 5.6ksame name
- compozy/compozy★ 2.9ksame name
- crbnos/carbon★ 2.8ksame name
- lingyired/status-trio★ 2ksame name
- GanyuanRan/Aegis★ 1.4ksame name
- pcvelz/superpowers★ 1.3ksame name
181 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository21 other files it gives its agents.
- browserbaseSkill
- cairnSkill
- developmentSkill
- dispatchSkill
- house-styleSkill
- kimi-subagentSkill
- lean-design-standardSkill
- mergeSkill
- nexus-authSkill
- nexus-jiraSkill
- orchestrateSkill
- orchestratorSkill
- planningSkill
- releaseSkill
- remote-opsSkill
- reviewSkill
- securitySkill
- specSkill
- ticket-pipelineSkill
- vision-reviewSkill
- workflow-basicsSkill
The file
We split this skill into the rules it gives an agent
1 rule. 0 are common, 1 is distinctive, and 0 are very specific to this project.
0
Common
Most of the 181 edited versions we read keep these.
1
Distinctive
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 23
Put the skill's leading word first.
“Common” is measured: kept by at least 6 in 10 of the 181 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Writing skills well
A skill exists to wrangle determinism out of a stochastic system. The root virtue is predictability — the agent taking the same process every run (not producing the same output). Every rule below serves it.
(Adapted from mattpocock/skills writing-great-skills, fitted to our conventions. Our prior rule stands: skills = procedures + the tools to run them; facts/state/decisions → memory; needed-every-turn → CLAUDE.md. A skill that is mostly facts is a memory file wearing the wrong hat.)
Invocation: two costs, pick one
- Model-invoked (has a rich
description): the agent can fire it autonomously and other skills can reach it. Costs context load — the description sits in the window every turn, forever. - User-invoked (
disable-model-invocation: true, human-facing one-line description): zero context load, but costs cognitive load — the operator is the index that must remember it exists.
Pick model-invocation only when the agent must reach it on its own or another skill must reach it. Our per-role SkillAllowlist work sharpens the stakes: every model-invoked description a role can see is context that role pays for on every turn — scope tightly.
The description
Two jobs: state what the skill is, and list the branches that trigger it. It earns the hardest pruning in the file:
- Put the skill's leading word first.
- One trigger per branch — synonyms restating one branch are duplication; collapse them.
- Cut identity that's already in the body; keep triggers + any "when another skill needs…" reach clause.
Information hierarchy
Two content types — steps (ordered actions) and reference (rules/definitions consulted on demand) — mix freely. Place each piece on the ladder by how immediately the agent needs it:
- In-skill step — ends on a completion criterion that is checkable (done vs not-done decidable) and, where it matters, exhaustive ("every modified file accounted for", not "produce a list"). A vague criterion invites premature completion. This is the same rule as the broker's observable-acceptance-criteria doctrine — one discipline, two homes.
- In-skill reference — a flat peer-set of rules on one rung is fine, not a smell.
- External reference — pushed to a sibling file behind a context pointer, loaded only when it fires. The pointer's wording decides whether the agent reaches it.
Progressive disclosure = moving material down the ladder so the top stays legible. The branching test: inline what every branch needs; push behind a pointer what only some branches reach. Co-location: keep a concept's definition, rules, and caveats under one heading — reading one part brings its neighbours.
Leading words
A leading word is a compact concept already in the model's pretraining that anchors a region of behaviour in one token (tracer bullet, fog of war, red, seal, fold). It serves predictability twice: in the body it anchors execution; in the description it anchors invocation. Hunt for restatements begging to collapse: "fast, deterministic, low-overhead" → tight; "a loop you believe in" → red. Fewer tokens AND a sharper hook. (Our own examples already work this way: cairn's express/fold/seal, the gate work's judge-the-diff, the cognitive-core's pool/belief.)
When to split
Each cut spends one of the two loads — split only when the cut earns it:
- By invocation: a distinct leading word should trigger independently, or another skill must reach it. Pays context load.
- By sequence: the steps still ahead tempt the agent to rush the current one — hide the post-completion steps. Use only when the completion criterion is irreducibly fuzzy AND the rush is observed; sharpening the criterion is cheaper and comes first.
Failure modes (diagnose by name)
- Premature completion — ending a step before it's genuinely done. Defence in order: sharpen the completion criterion (cheap, local); only then split the sequence.
- Duplication — one meaning in two places. Costs maintenance + tokens, and inflates the meaning's rank on the ladder. Keep a single source of truth per meaning.
- Sediment — stale layers that settle because adding feels safe and removing feels risky. The default fate of any skill without a pruning discipline. Our verify-the-deployed-artifact lesson generalizes: a skill's text drifts from practice exactly like a manifest drifts from a deploy.
The pruning pass (run this over a skill, or the whole library)
For each skill:
- Hat check — is any section facts/state/decisions? Move it to memory; leave a pointer if a branch needs it.
- Description audit — one trigger per branch, leading word front-loaded, no body-identity restated.
- No-op hunt — test each sentence in isolation; a sentence that changes no behaviour is deleted whole, not trimmed.
- Duplication sweep — every meaning has one home; collapse restatements into the leading word.
- Ladder check — anything inline that only one branch reaches → push behind a pointer; anything behind a pointer that every run needs → pull up.
- Criterion check — every step ends on a checkable, exhaustive-where-it-matters completion criterion.
- Sediment date — if a line describes a practice we no longer follow, delete it now; "might need it" is what memory and git history are for.
Completion criterion for the pass itself: every skill in scope has had all seven checks applied and each resulting edit either made or explicitly declined with a reason.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

