Clawify Documentation

Quick Start

For AI Agents (zero install)

One command to teach your AI agent how to use the registry:

curl -sL https://clawify.world/api/agent/bootstrap | bash

After this, tell your agent: "帮我从 clawify 找个 code review 的 skill 装上"

For Developers (CLI)

pip install clawify
clawify match "help me write unit tests"
clawify install subagent-dev

For Applications (API)

curl -X POST https://clawify.world/api/agent/match \
  -H "Content-Type: application/json" \
  -d '{"task":"code review","limit":5}'

Bootstrap for Agents

The bootstrap script installs a meta-skill into your AI tool (Claude Code, Codex, OpenClaw). After installation, the agent can autonomously search and install skills from the registry.

GET /api/agent/bootstrap

Returns a shell script that auto-detects installed AI tools and writes the clawify meta-skill.

curl -sL https://clawify.world/api/agent/bootstrap | bash
# Output:
#   ✓ Installed to Claude Code (~/.claude/commands/clawify.md)
#   ✓ Installed to Codex CLI (~/.codex/skills/clawify/SKILL.md)

Agent Discovery

GET /.well-known/clawify.json

Standard discovery endpoint for agents to find the registry automatically.

{
  "registry": "https://clawify.world/api/agent",
  "version": "1.0",
  "capabilities": ["search", "match", "install", "download", "publish", "versions"],
  "auth": { "type": "api_key", "prefix": "clw_" },
  "skill_count": "..."
}

Roles & Tags

GET /api/agent/roles

~200 tokens. Layer 0 discovery.

{"roles": [
  {"name": "operations", "count": 108},
  {"name": "research", "count": 79},
  {"name": "engineering", "count": 30},
  ...
]}

GET /api/agent/tags?limit=20

List Skills

GET /api/agent/skills?q=&role=&detail=summary&page=1&per_page=20&sort=quality

Progressive Disclosure

The detail parameter controls response density, saving agent context window:

Level	Fields	~Tokens	Use case
`minimal`	name, slug, role	~15	Listing, autocomplete
`summary`	+ description, tags, quality_score	~30	Selection, comparison
`full`	+ version, signals, install commands	~80	Decision, install

Examples

# Minimal
curl 'https://clawify.world/api/agent/skills?detail=minimal&per_page=50'

# Summary
curl 'https://clawify.world/api/agent/skills?q=testing&detail=summary'

# By role
curl 'https://clawify.world/api/agent/skills?role=engineering&detail=summary'

Smart Match

POST /api/agent/match

Server-side intelligent matching. The core agent endpoint.

Describe a task in natural language. The server matches skills using keyword relevance + quality signals, returning the top results. Typically ~400 tokens total.

curl -X POST https://clawify.world/api/agent/match \
  -H "Content-Type: application/json" \
  -d '{"task": "help me optimize database queries", "limit": 5, "detail": "summary"}'

Skill Detail

GET /api/agent/skills/{name}?detail=full

Resolve by human_name, source-{id} slug, or source_key.

Skill Content

GET /api/agent/skills/{name}/content

Returns raw SKILL.md (text/markdown). Increments download count.

Download with Tracking

POST /api/agent/skills/{name}/download

Returns JSON with content + metadata. Set X-Client header for tracking.

Publish a Skill

POST /api/agent/skills

Requires authentication. Creates a new skill in the registry.

SKILL.md Frontmatter

---
name: my-code-review
role: engineering
tags: [review, quality, testing]
depends: [git-conventional-commits]
requires: [tool:git]
---
# My Code Review
...

Update a Skill

PUT /api/agent/skills/{name}

Only the original author or admin can update. Auto-bumps patch version.

Version History

GET /api/agent/skills/{name}/versions

Dependencies

GET /api/agent/skills/{name}/dependencies

API Keys

POST /api/agent/auth/api-keys

Create an API key. The full key (clw_...) is returned only once.

Auth Methods

Method	Format	Use case
API Key	`Authorization: Bearer clw_...`	Agent/CLI access, publishing
Session Token	`Authorization: Bearer tkn-...`	Web UI, trial jobs

Most read endpoints work without authentication. Auth is required for publishing and API key management.

CLI Commands

pip install clawify

Command	Description
`clawify match "task"`	Find best skills for a task
`clawify search --role eng`	Search by keyword/role/tag
`clawify install <name>`	Install to Claude/Codex/OpenClaw
`clawify uninstall <name>`	Remove skill
`clawify update [name]`	Update installed skills
`clawify list`	Show installed skills
`clawify info <name>`	Skill details + quality signals
`clawify publish <file>`	Publish SKILL.md to registry
`clawify versions <name>`	Version history
`clawify deps <name>`	Show dependencies
`clawify roles`	List categories
`clawify login`	Authenticate

clawify match

clawify match "help me design a REST API"
clawify match "帮我写小红书文案" --limit 10
clawify match "database optimization" --json

clawify search

clawify search "code review"
clawify search --role engineering --limit 20

clawify install

clawify install subagent-dev                 # all detected tools
clawify install subagent-dev --target claude  # Claude Code only

clawify publish

clawify publish ./SKILL.md --role engineering
clawify publish ./SKILL.md --update --changelog "Added security checks"

Quality Score

Each skill has a composite quality score (0-1) computed from multiple signals. The ranking function is pluggable.

quality_score = f(
  text_relevance,      // query match
  trial_rating,        // user ratings (1-5)
  usage_frequency,     // log(trial + download count)
  success_rate,        // completed / total trials
  community_signal     // log(github_stars + likes)
)

Usage Signals

Signals are collected automatically as skills are used:

Signal	Source	Updated when
trial_count	Trial worker	Job completes or fails
trial_avg_rating	Rating endpoint	User rates a job
download_count	Content endpoint	GET /content called
install_count	Download endpoint	CLI/agent installs
view_count	Web UI	Skill card viewed

More usage data = better recommendations. This is the core feedback loop.

Getting Started

Agent API

Publishing

Auth

CLI

Concepts