writing-skills
Skill1 starsChanged 2 months agoMITView on GitHub
Author and maintain Claude Code skills that trigger reliably and actually change behavior. Use when creating a new skill, editing an existing SKILL.md or its description, deciding what moves to references/, or verifying a skill works before deployment ("write a skill", "improve this skill", "why doesn't this skill trigger"). NOT for authoring project convention rules in .claude/rules/ or agent definitions in agents/.
191 other projects use a version of this.
Adopted by
Most recently taken up by
- 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
- alexuraxin/retainertheir own version, first seen Nov 15, 2026
- crbnos/carbontheir 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
frankxaiSkilldefense-in-depth
Use when invalid data causes failures deep in execution, requiring validation at multiple system layers - validates at every layer data passes through to make bugs structurally impossible
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★ 1.8ksame 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 repository27 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- design-craft/CLAUDE.mdCLAUDE.md
- android-emulatorSkill
- audit-branchSkill
- brainstormSkill
- create-devtoolSkill
- create-featureSkill
- create-infrastructureSkill
- execute-tasksSkill
- finish-featureSkill
- frontend-testingSkill
- generate-feature-docSkill
- generate-prdSkill
- git-commitSkill
- ios-simulatorSkill
- modify-featureSkill
- pipeline-helpSkill
- plan-implementationSkill
- prd-clarifierSkill
- prd-to-uxSkill
- prototypeSkill
- react-clean-architectureSkill
- receiving-code-reviewSkill
- refactoring-uiSkill
- setup-daher-skillsSkill
- systematic-debuggingSkill
- test-mobile-appSkill
The file
We split this skill into the rules it gives an agent
5 rules. 1 is common, 4 are distinctive, and 0 are very specific to this project.
1
Common
Most of the 181 edited versions we read keep these.
4
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 38
Start with "Use when..." and list concrete triggers, symptoms, and situations
- Common: line 39
Write in third person (it is injected into the system prompt)
- Distinctive: line 59
Run one subagent with a realistic prompt that should trigger the skill
- Distinctive: line 60
Check the outcome against what an agent does without the skill
- Distinctive: line 63
If the agent ignored or misapplied the skill, fix the description or the ambiguous section and re-run.
“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
Overview
A skill is a reference guide that teaches future Claude instances a reusable technique, pattern, or workflow — not a narrative about how you solved a problem once. Keep SKILL.md lean, write the description for discovery, and verify the skill actually changes behavior before trusting it.
Canonical methodology: references/anthropic-best-practices.md — Anthropic's full skill-authoring guide (conciseness, progressive disclosure, workflows, scripts, evaluation). Read it when authoring anything non-trivial.
Writing the Description (Most Important Part)
Claude reads only the frontmatter description when deciding whether to load a skill. It must answer: "Should I read this skill right now?"
Rule: the description states WHEN to use the skill — triggers, symptoms, situations. It NEVER summarizes the skill's workflow.
Why: when a description summarizes the process, Claude tends to follow the summary and skip the skill body entirely. A description saying "code review between tasks" caused Claude to run ONE review even though the skill required two; rewriting the description to triggers-only ("Use when executing implementation plans with independent tasks") made Claude read the body and follow both reviews.
# ❌ Summarizes workflow — Claude may follow this instead of reading the skill
description: Use when executing plans - dispatches subagent per task with code review between tasks
# ❌ Too abstract, no trigger
description: For async testing
# ✅ Triggering conditions only
description: Use when executing implementation plans with independent tasks in the current session
# ✅ Symptoms, not process
description: Use when tests have race conditions, timing dependencies, or pass/fail inconsistentlyGuidelines:
- Start with "Use when..." and list concrete triggers, symptoms, and situations
- Write in third person (it is injected into the system prompt)
- Describe the problem (race conditions, flaky tests), not language-specific symptoms — unless the skill is technology-specific, in which case name the technology explicitly
- Include keywords Claude would search for: error messages, tool names, synonyms
Structure Checklist
- Frontmatter has
name(letters, numbers, hyphens only) anddescription(max 1024 chars) - A skill that takes an argument declares
argument-hint: <what-to-pass>— it is what the user sees on/skill <tab>; without it they guess (the validator requires the<placeholder>form) - SKILL.md stays lean — well under 500 lines; move depth to
references/files loaded on demand - One canonical copy of any shared reference — never duplicate guidance across skills; cross-reference by skill name instead
- One excellent code example beats many mediocre ones in multiple languages
- No
@force-load links to other files (they consume context immediately); use plain paths or skill names - Skills that manage long-lived processes (dev servers, watchers, emulators, browsers) define ownership — reuse a healthy process, never kill one the session didn't start, read ports/URLs from real output — and end with a symptom→fix troubleshooting list
- Written for Claude 5, not against an older model's habits: no
CRITICAL/MUST/NEVERwithout the failure it prevents (bare emphasis now causes over-triggering); no "double-check", "be thorough", "think step by step" (the model does these unprompted — the lines only add re-verification cost); no request to reproduce internal reasoning in the reply (it can be declined and trip safety flags — ask for the conclusion plus the evidence instead); review prompts ask for everything and let a triage step filter (a "be conservative" reviewer withholds real findings). - Task sections state outcome + guardrails + a runnable finish line ("the tests pass", "every endpoint is migrated" — not "make it work"), and keep a numbered protocol only where the order itself is the deliverable; documents the skill writes carry a length calibration
Test Before Trusting
For any meaningful new skill (or meaningful edit), verify it works before deploying:
- Run one subagent with a realistic prompt that should trigger the skill
- Check the outcome against what an agent does without the skill
- Confirm the skill actually changed behavior — the agent found it, read it, and followed it
If the agent ignored or misapplied the skill, fix the description or the ambiguous section and re-run. Trivial edits (typos, link fixes) don't need this.
Common Mistakes
| Mistake | Fix |
| Description summarizes the workflow | Rewrite to triggers/symptoms only |
| Everything inline in SKILL.md | Move heavy reference material to references/ |
| Same guidance pasted into multiple skills | Keep one canonical copy, cross-reference it |
| Narrative "how I solved it once" storytelling | Distill into a reusable technique |
| Deployed without testing | Run one subagent check first |
Done when: gate that self-attests ("you understand X") | Gate on a runnable check — a command plus its expected output |
Verdicts without evidence — a reviewer contract or triage step that accepts Superseded / Looks fine / Rejected as a reason | Require the evidence and the alternative in the reason: superseded by X, which covers A and B; nothing unique remains — a bare label cannot be checked, so it cannot be wrong |
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.

