skill-builder
Skill1 starsChanged 4 months agoMITView on GitHub
Turn a brief problem statement plus optional repo context into a complete, repo-quality skill scaffolded from the template. Use when asked to build, create, or scaffold a new skill, or when someone describes a recurring task they want handled as a reusable skill. Researches the repo, scopes to one job, fills every template layer, scaffolds evals — then stops for review.
75 other projects use a version of this.
Adopted by
Most recently taken up by
- BloodSweatxED/Claude-skillstheir own version, first seen Nov 16, 2026
- arctan303/creator.skilltheir own version, first seen Nov 12, 2026
- Kikobazz123/lordgen-skill-buildertheir own version, first seen Nov 8, 2026
- AsiaOstrich/EngramGraphtheir own version, first seen Nov 8, 2026
- shetengteng/skillix-hubtheir own version, first seen Nov 8, 2026
- davidgfolch/AI-job-searchtheir own version, first seen Nov 8, 2026
Used with
Projects with it also use contract-test, literature-research 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.
squall-chuaSkillcontract-test
Test a running API against its own contract — OpenAPI, GraphQL, protobuf, or AsyncAPI — from the outside, and report every drift between what the contract promises and what the API does.
Also in 3 projects that use this
nickjwellsSkillliterature-research
Search academic literature and perform deep research reviews. Use when user asks to search PubMed, find academic papers, or do literature reviews.
Also in 3 projects that use this
nopandererSkillcommit-and-push
Automatically analyze git changes, generate Conventional Commits compliant commit messages, commit changes, and push to remote in a single automated workflow. Use when the user invokes /commit-and-push or explicitly requests to automatically commit and push changes following Conventional Commits format (feat:, fix:, docs:, etc.). This skill provides a fully automated git workflow without manual intervention.
Also in 3 projects that use this
MicrosoftSkillskill-reviewer
Review skill PRs with structured severity-rated feedback covering token budgets, routing conflicts, and repo conventions. WHEN: \"review skill\", \"review skill PR\", \"review skill changes\", \"check skill quality\", \"skill PR feedback\".
Also in 3 projects that use this
SkillPilotCHSkilllevel-up
Eine Automatisierung pro Woche finden, scopen, shippen. Output landet als proposed-Eintrag in cadence.md mit konkreten Implementations-Schritten.
Also in 3 projects that use this
MicrosoftSkillexcalidraw-diagram
Create Excalidraw diagram JSON files that make visual arguments. Use when the user wants to visualize workflows, architectures, or concepts.
Also in 4 projects that use this
Other versions75 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 75
- yusufkaraaslan/Skill_Seekers★ 15ksame name
- block/berd★ 1.1ksame name
- boshu2/agentops★ 454same name
- metaskills/skill-builder★ 112same name
- AsiaOstrich/universal-dev-standards★ 84same name
- nateherkai/a-bunch-of-skills★ 41same name
- clasen/Skills★ 40same name
- sanqi-cd/Sanqi-Skills★ 28same name
- arctan303/creator.skill★ 19same name
- aiagentwithdhruv/skills★ 16same name
65 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.
- CLAUDE.mdCLAUDE.md
- ai-commentsSkill
- ai-draftSkill
- ai-nextSkill
- ai-postSkill
- call-traceSkill
- draft-prSkill
- dto-mappingSkill
- handoffSkill
- hum-commentsSkill
- jira-dcSkill
- meeting-summarySkill
- plan-from-issueSkill
- plan-nextSkill
- plan-reviewSkill
- plan-saveSkill
- pr-commentsSkill
- pr-statsSkill
- review-sessionSkill
- validator-generatorSkill
- templateSkill
The file
We split this skill into the rules it gives an agent
7 rules. 0 are common, 2 are distinctive, and 5 are very specific to this project.
0
Common
Most of the 66 edited versions we read keep these.
2
Distinctive
This author's own choices. Worth a look.
5
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 41
Read the anatomy: Read references/skill-anatomy.md for the exact structure and quality bar you must produce.
- Very specific: line 45
Research: If pointed at a repo, explore it with Read / Grep / Glob — README, any existing .claude/skills/, naming and code conventions, …
- Very specific: line 51
Layer 3 + evals: Add a references/ file only if the skill needs a big catalog, signature table, or worked examples.
- Very specific: line 53
Self-check: Verify against the checklist in references/skill-anatomy.md: no surviving <ANGLE_BRACKET> (portable mode), name matches the …
- Distinctive: line 80
Ship evals: Every skill gets 2–3 cases and at least one negative case — a skill without a way to prove it works is unfinished.
- Distinctive: line 86
NEVER guess the skill's purpose: Ask, or stop — purpose is the one thing you may not invent.
- Very specific: line 93
Project: Only when pointed at a real repo and explicitly asked: bake that repo's real paths and conventions into a skill destined for its …
“Common” is measured: kept by at least 6 in 10 of the 66 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
Skill Builder: turn a one-line brief into a review-ready skill
You are turning a brief problem statement into a finished, review-ready skill. Your job is to do the research, scope the skill to a single job, fill every layer of the template, scaffold its evals, then STOP for review. You never install, commit, or push what you scaffold.
The failure mode this skill exists to prevent is the kitchen-sink skill: a vague brief becomes a skill that tries to do three things, leads with "helps with" filler, lists every tool, and ships with half its <ANGLE_BRACKET> placeholders unfilled. Guard against it on every build — one job, trigger-first description, minimal tools, zero surviving placeholders.
When to use this skill
- "Build / create / scaffold a skill for X"
- Someone describes a recurring task — usually with a repo or an example — and wants it turned into a reusable skill
- Turning a skill brief or issue (like this repo's issues) into an actual
skills/<name>/folder
When NOT to use this skill
- Editing or reviewing an existing skill — open it and edit it directly; this skill authors new ones.
- The "skill" is really a one-off — a single shell command, a config tweak, or a permission rule. Not every task needs a skill; say so and stop. (Settings or hook changes are a config task, not a skill.)
- The brief implies several unrelated jobs — that's N skills, not one. This skill builds one at a time; propose the split (step 5).
Steps
- Intake. Capture the brief problem, any context (repo path, an example, the specific pain point), and the target output mode. Default to portable unless a real repo is in play and the user wants the skill baked for it — then project (see Rules → Output modes).
- Read the anatomy. Read
references/skill-anatomy.mdfor the exact structure and quality bar you must produce. Iftemplate/exists in the working repo, that folder is your canonical starting scaffold (cp -r template skills/<name>); otherwise reproduce the same structure from the reference. - Clarify — as much as it takes, no padding. Ask via AskUserQuestion until you genuinely understand the single job, its triggers, the target output mode, and the key decisions the skill must encode. A thin brief may need several rounds to become a valuable, detailed skill; a detailed brief may need none. Batch related questions (up to 4 per prompt) instead of drip-feeding, ask only what you can't settle yourself by researching the repo (step 4) or by a sensible stated default, and stop once another answer wouldn't change the scaffold — depth where it makes the skill better, never interrogation for its own sake. Never guess the skill's purpose — that is the one thing you may not invent.
- Research. If pointed at a repo, explore it with Read / Grep / Glob — README, any existing
.claude/skills/, naming and code conventions, domain vocabulary, how it runs tools — so the new skill is grounded and doesn't reinvent patterns the repo already has. - Scope to one job. Distill the brief to a single crisp purpose. If it implies multiple jobs, propose splitting into N named skills and confirm which to build first — then build only that one. Do not scaffold until the scope is one job.
- Scaffold. Create the skill folder —
skills/<name>/(portable) or the target repo's.claude/skills/<name>/(project) — and fill every placeholder: kebabnamematching the folder, trigger-firstdescription, minimalallowed-toolsinferred from the job, When to use / When NOT to use, numbered Steps with explicit stop conditions, Output format, and the Rules triad. - Layer 3 + evals. Add a
references/file only if the skill needs a big catalog, signature table, or worked examples; otherwise omit it. Always writeevals/README.mdwith 2–3 representative cases, including at least one negative "should not fire / should ask instead" case. - Self-check. Verify against the checklist in
references/skill-anatomy.md: no surviving<ANGLE_BRACKET>(portable mode),namematches the folder, description leads with the trigger,allowed-toolsis minimal, output is portable unless project mode was chosen. Fix anything that fails before reporting. - Report & stop. Print the new skill's path, a one-paragraph summary (the job and its triggers), and the evals to run. Do not install, commit, or push. Stop.
Output format
Report exactly this, then stop:
Created skill: <path>/SKILL.md (mode: portable | project)
Job: <one sentence — the single job this skill does>
Triggers: <when Claude should fire it>
Files: SKILL.md, evals/README.md[, references/<file>]
Verify with: <the eval case names from evals/README.md>
Not done for you: not installed, not committed.
To try it: cp -r <path> ~/.claude/skills/ and run the evals in a fresh session.Rules
What to do
- One job, always. A skill has a single purpose. If you need "and" to describe it, it is two skills — propose the split and build one.
- Trigger-first description. Lead the
descriptionwith a tight statement of the job, with the trigger words a user would actually say ("build / create / scaffold a skill") right up front — it is the only layer-1 metadata and it decides whether the skill triggers at all. - Minimal
allowed-tools. List only the tools the job needs. A read-only skill lists noWrite/Edit/Bash. Infer the set from the steps, not from habit; scopeBashto subcommands when you can (e.g.Bash(git status:*)). - Ask until the job is clear — by sufficiency, not a quota. Ask as many genuine questions as it takes to scope a valuable, detailed skill; ask none if the brief already settles it. No padding, batch related ones, and never ask what you could answer by researching the repo. Ask, don't guess.
- Ship evals. Every skill gets 2–3 cases and at least one negative case — a skill without a way to prove it works is unfinished.
What NOT to do
- NEVER ship a kitchen-sink skill. Multi-job briefs get split, not crammed.
- NEVER leave a surviving
<ANGLE_BRACKET>in portable output. The self-check exists to catch this. - NEVER guess the skill's purpose. Ask, or stop — purpose is the one thing you may not invent.
- NEVER install, commit, or push the scaffold. The deliverable is files for review; landing them is the user's call.
- Don't manufacture a skill for a one-off. A single command or config change isn't a skill — say so and stop.
Output modes
- Portable (default). Placeholders, no org names, no real paths, no stack or schema assumptions — the kind of skill that ships in this repo. Everything project-specific is an
<ANGLE_BRACKET>the adopter fills in. - Project. Only when pointed at a real repo and explicitly asked: bake that repo's real paths and conventions into a skill destined for its local
.claude/skills/. The skill itself stays portable — the mode governs only the output.
Format discipline
- The deliverable is the files plus a tight report. No preamble, no narrating the research, no recap. Scaffold, report, stop.
For the full structure, quality bar, and finished-skill checklist, read references/skill-anatomy.md.
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.

