Docs · API
Agent Skills OS API
A small public REST API over the skill library. Search for a skill, read its full instructions, or ask in plain language which one fits. No key, no sign-up, JSON in and out. The machine-readable description is /openapi.json (OpenAPI 3.1).
Overview
| Operation | Request | Returns |
|---|---|---|
listSkills | GET /api/skills?q=&limit= | Skills matching the query, best first (or all of them) |
getSkill | GET /api/skills/{name} | One skill with its full SKILL.md and usage snippets |
askSkills | POST /api/ask | A short answer plus cards for the skills it means |
- Base URL:
https://skills.rakibulism.space - Authentication: none. The read endpoints allow cross-origin requests (
Access-Control-Allow-Origin: *). - Format: JSON with UTF-8. Every error is JSON too (see Errors).
- Limits:
POST /api/askallows about 12 questions a minute per client. The read endpoints are cached for a few minutes and have no fixed limit.
When to use it
- An agent needs instructions for a task (a UI review, an accessibility audit) and wants to check whether a ready-made skill exists: call
listSkillswith a few keywords. - A tool wants to load a skill at runtime without installing the npm package: call
getSkilland use themarkdownfield as the system prompt. - A user asks a fuzzy question ("how do I use this with LangChain?"): call
askSkillsand show its answer and cards.
If you write JavaScript or TypeScript and can install a package, the npm package is faster and works offline. For shell scripts, use the CLI.
Search or list skills
GET /api/skills returns every skill sorted by name. Add q to keep only skills whose name, description, tags or text contain every word, best match first.
curl "https://skills.rakibulism.space/api/skills?q=ui&limit=5"{
"query": "ui",
"total": 1,
"count": 1,
"skills": [
{
"name": "super-ui",
"title": "super-ui",
"description": "…",
"version": "0.1.0",
"tags": ["design", "ui"],
"url": "https://skills.rakibulism.space/skills/super-ui",
"markdownUrl": "https://skills.rakibulism.space/skills/super-ui.md",
"apiUrl": "https://skills.rakibulism.space/api/skills/super-ui"
}
]
}| Parameter | Type | Meaning |
|---|---|---|
q | string, up to 200 characters | Keywords for the task. Optional. |
limit | integer, 1 to 100 | Most skills to return. Defaults to 100. |
Get one skill
GET /api/skills/{name} returns the same fields plus markdown (the whole SKILL.md), usage snippets and sourceUrl.
curl https://skills.rakibulism.space/api/skills/super-uiOnly need the file? Request /skills/super-ui.md directly, or send Accept: text/markdown to the skill page.
Ask a question
POST /api/ask with a JSON body. The answer is grounded in the skills search ranks highest, and cards point at skills, install commands or usage snippets the site can render.
curl -X POST https://skills.rakibulism.space/api/ask \
-H "Content-Type: application/json" \
-d '{"question": "Which skill reviews UI?"}'{
"answer": "super-ui is the closest skill …",
"cards": [{ "type": "skill", "slug": "super-ui" }],
"followups": ["How do I use it with LangChain?"]
}Errors
Errors use normal HTTP status codes and always return the same JSON shape, so an agent can read what went wrong and what to do next.
{
"error": "There is no skill named \"nope\".",
"code": "skill_not_found",
"hint": "GET /api/skills?q=<keywords> to find the right name.",
"docs": "https://skills.rakibulism.space/docs/api"
}| Status | code | What to do |
|---|---|---|
| 400 | invalid_name, invalid_limit, query_too_long, missing_question | Fix the request as the hint says. |
| 404 | skill_not_found, not_found | Search with listSkills, or check the path against /openapi.json. |
| 405 | method_not_allowed | Use the method in the Allow header. |
| 429 | rate_limited, ai_busy | Wait for Retry-After seconds, or use listSkills. |
| 502, 503, 504 | ai_unavailable, empty_answer, index_unavailable, skill_file_unavailable, not_configured, ai_timeout | Try again shortly. |
Markdown for agents
Every page on this site has a Markdown version. Send Accept: text/markdown to any page URL and you get Markdown with Content-Type: text/markdown and Vary: Accept; browsers keep getting HTML. You can also add .md to a path, for example /docs.md, or use /index.md for the home page.
curl -H "Accept: text/markdown" https://skills.rakibulism.space/Start from /llms.txt for a map of the site written for language models, and /sitemap.xml for every page.