1 · pick
categoryDev ✓
DevelopmentFrontendContentData & MLProjectCustom
2 · answer
answers5 / 5
› Which dirs are off-limits?
› Named or default exports?
› Commit policy never to main
› Never modify __tests__/
› Strict null checks on
3 · generate
SKILL.md0 / 100
--- name: ts-code-review description: Reviews TS & Next.js — type safety, naming, commit rules. --- ## Hard stops - Never modify __tests__/ · never commit to main ## Verification - [ ] No implicit any · named exports only
SKILL.md generator Security scanned

Your agent.
Your rules.
Every session.

Answer a few targeted questions about your workflow. Get a production-ready, quality-scored SKILL.md for every major AI agent.

$npx skilldraft install <id>auto-installs to every detected agent

Skills people are shipping right now.

Browse all community skills
owasp-security-audit95
security@SkillDraft
sql-query-guardrails95
Data@SkillDraft
github-actions-ci94
devops-infrastructure@SkillDraft
a11y-accessibility-audit94
Frontend@SkillDraft
ts-code-review93
Development@SkillDraft
fastapi-endpoint-builder93
backend-frameworks@SkillDraft
owasp-security-audit95
security@SkillDraft
sql-query-guardrails95
Data@SkillDraft
github-actions-ci94
devops-infrastructure@SkillDraft
a11y-accessibility-audit94
Frontend@SkillDraft
ts-code-review93
Development@SkillDraft
fastapi-endpoint-builder93
backend-frameworks@SkillDraft
release-notes88
Project@wei
api-contract89
Development@noa
design-tokens94
Frontend@ivy
migration-runner86
Data@tom
changelog-writer87
Content@rae
test-scaffold90
Development@os
release-notes88
Project@wei
api-contract89
Development@noa
design-tokens94
Frontend@ivy
migration-runner86
Data@tom
changelog-writer87
Content@rae
test-scaffold90
Development@os
Built on the Anthropic API/ Security scanned before publishing/ Scored on 7 dimensions/ agentskills.io open standardCommunity skills

Most instructions get ignored.
A skill doesn't.

A one-line ask your agent forgets by the third message becomes a structured, scored, scanned SKILL.md it follows every session.

✕ vague prose
ignored

"be careful with the tests, use good naming, and please don't break anything this time"

— pasted into chat · forgotten by msg #3
structure score scan
✓ production-ready
ts-code-review.md0 / 100
--- name: ts-code-review description: Reviews TS & Next.js — type safety, naming, rules. Use when: "review this", "check PR". Not for greenfield. --- ## Hard stops - Never modify __tests__/ · never commit to main ## Verification - [ ] No implicit any · named exports only
input: 1 line → output: 7 sections · hard stops · anti-patterns · checklist · scored 92/100
One file · written natively for every agent
Claude CodeCursorWindsurfCodex CLIGemini CLIGitHub CopilotChatGPT

Skills that actually work.

Most agent instructions are vague prose that the model quietly ignores. SkillDraft enforces structure, then proves it.

01

Quality scored

Every skill is rated 0–100 across seven dimensions — description density, why-annotations, when-not-to-use, output template, hard stops, anti-patterns, and a verification checklist. Nothing vague ships.

02

Security scanned

Checked for prompt injection, dangerous shell commands, data-exfiltration patterns, and obfuscation before anything is published. Community skills you can actually trust.

03

One-command install

The CLI auto-detects which AI agents you have installed and writes the skill to each. Claude Code, Cursor, Windsurf, Codex CLI — one command covers all of them.

Any community skill, one command.

Auto-detects Claude Code, Cursor, Windsurf, Codex CLI, and Gemini CLI — then writes the file to each agent directory automatically. No copy-paste, no wrong paths.

zsh — skilldraft
$ npx skilldraft install ts-code-review › detecting agents… Claude Code ~/.claude/skills/ Cursor ~/.cursor/rules/ Windsurf ~/.windsurf/ Codex CLI ~/.codex/ installed to 4 agents · 92/100

How it works.

Five steps, under two minutes. The specificity of the questions is what separates production-ready output from generic boilerplate.

01

Choose a category

Six workflow types — Development, Frontend, Content, Data, Project, or Custom. Each has its own tailored question set.

02

Pick your agent

Claude Code, Cursor, Copilot, ChatGPT, Windsurf, Codex, Gemini, or Universal. Platform drives format, path, and triggers.

03

Answer the questions

Three to five domain-specific questions — no generic text boxes. Specificity is what makes the output production-ready.

04

Watch it generate

Assembled in real time via the Anthropic API — instructions, hard stops, anti-patterns, checklist — in about ten seconds.

05

Install or copy

Drop it in your skills folder, or run skilldraft install to write it to every detected agent.

Stop re-explaining yourself to your agent.

Build a SKILL.md in under two minutes. Free, no login required — save and share with an account.