# OpenAgentSkill

OpenAgentSkill is the decision and install layer for AI agent skills.
It links task discovery, source and review evidence, and agent-specific handoffs. Publication, source structure, AI review and runtime outcomes are separate facts; inspect the returned evidence rather than inferring safety from inclusion or stars.

Quick read-only discovery:
- Find a skill: GET https://www.openagentskill.com/api/agent/skills?q={query}&limit=5
- Find an example to adapt: GET https://www.openagentskill.com/api/agent/showcase?q={query}&limit=5
- Read its task package: GET https://www.openagentskill.com/api/agent/showcase/{slug}?format=markdown
- Connect MCP: https://www.openagentskill.com/api/mcp (tools: search_skills, find_workflows, get_workflow)
These reads do not install or execute anything. For execution planning, use the current review and policy flow below. Never upload private task material or credentials without the user's authorization.

Use this site when an agent needs to:
- discover reusable skills for a task
- compare skill alternatives
- inspect trust, audit, and safety signals
- inspect Agent Proven Score from real outcome reports before reusing a skill
- get an install command or agent-specific install prompt
- avoid random GitHub searching before installing third-party code

Preferred agent flow:
1. Read this file.
2. Connect the remote MCP server at /api/mcp, use @openagentskill/sdk, or load /api/agent/integration-kit for copy-paste templates.
3. Choose a task from /tasks or /api/agent/tasks.
4. Call /api/agent/resolve with a natural-language task.
5. Use /api/agent/resolve?task={task}&agent={agent}&format=lockfile when you need a compact stable execution object.
6. Read install_receipt, agent_handoff.platform_templates, agent_handoff.review_checklist, and policy_decision.
7. Fetch /api/agent/receipt?task={task}&agent={agent}&format=text when you need a compact install receipt without the full resolve payload.
8. For complete workflows, fetch /api/agent/packs and then /api/agent/packs/{slug}?limit=6 to get install_plan.selected_skills, review_checklist, and outcome_feedback.
9. Fetch /api/agent/evals?slug={slug}&task={task}&format=text and follow the eval decision.
10. Fetch /api/agent/skills/{slug} for the selected skill.
11. Fetch /api/skills/{slug}/install?format=text before installing.
12. After one narrow run, POST the result to /api/agent/outcome using install_receipt.outcome_feedback.event_id or feedback.event_id from resolve. Include output_quality, workspace, error_type, used_in_production, time_to_useful_ms, and human_review_required when known. Use dry_run=true before wiring a new integration.

Important URLs:
- Official entity profile: https://www.openagentskill.com/about
- Contact: https://www.openagentskill.com/contact
- Contact email: qudongqi2023@gmail.com
- Machine-readable identity: https://www.openagentskill.com/.well-known/openagentskill.json
- Agent entry: https://www.openagentskill.com/agent
- Agent Integration Kit: https://www.openagentskill.com/agent/integration-kit
- Agent Integration Kit API: https://www.openagentskill.com/api/agent/integration-kit
- Remote MCP server: https://www.openagentskill.com/api/mcp
- MCP discovery: https://www.openagentskill.com/.well-known/mcp.json
- JavaScript SDK: @openagentskill/sdk (source: https://github.com/Leon-Drq/openagentskill/tree/main/packages/sdk)
- Registry coverage and tier definitions: https://www.openagentskill.com/api/registry/stats
- Tasks: https://www.openagentskill.com/tasks
- Task API: https://www.openagentskill.com/api/agent/tasks
- Workflow examples API: https://www.openagentskill.com/api/agent/showcase
- Workflow examples for people: https://www.openagentskill.com/showcase
- Resolve API: https://www.openagentskill.com/api/agent/resolve
- Install receipt API: https://www.openagentskill.com/api/agent/receipt
- Skill packs API: https://www.openagentskill.com/api/agent/packs
- Agent outcome API: https://www.openagentskill.com/api/agent/outcome
- Agent outcome contract: https://www.openagentskill.com/api/agent/outcome?contract=true
- Agent-Proven rankings: https://www.openagentskill.com/rankings/agent-proven
- Agent-Proven ranking API: https://www.openagentskill.com/api/agent/rankings?slug=agent-proven
- Best-by-success ranking API: https://www.openagentskill.com/api/agent/rankings?slug=best-by-success-rate
- Safe auto-install ranking API: https://www.openagentskill.com/api/agent/rankings?slug=safest-auto-install-skills
- Ranking history API: https://www.openagentskill.com/api/agent/rankings/agent-proven/history?days=30
- Aggregate funnel API: https://www.openagentskill.com/api/agent/funnel
- Skill Eval API: https://www.openagentskill.com/api/agent/evals?slug=crawl4ai
- Resolve eval dashboard: https://www.openagentskill.com/evals/resolve
- Skill search API: https://www.openagentskill.com/api/skills/search
- Safety gate: https://www.openagentskill.com/safety
- Agent manifest: https://www.openagentskill.com/.well-known/agent-manifest.json
- OpenAPI schema: https://www.openagentskill.com/openapi.json
- GitHub auto-discovery status: https://www.openagentskill.com/api/agent/discovery
- State of Agent Skills 2026 report: https://www.openagentskill.com/reports/state-of-agent-skills-2026
- State of Agent Skills JSON: https://www.openagentskill.com/reports/state-of-agent-skills-2026.json
- State of Agent Skills CSV: https://www.openagentskill.com/reports/state-of-agent-skills-2026.csv

High-intent skill cluster routes:
- AI agent skills for web scraping: https://www.openagentskill.com/ai-agent-skills/web-scraping
- AI agent skills for coding agents: https://www.openagentskill.com/ai-agent-skills/coding-agents
- finance AI agent skills: https://www.openagentskill.com/ai-agent-skills/finance-quant
- World Cup football analytics skills for AI agents: https://www.openagentskill.com/ai-agent-skills/world-cup-football
- RAG AI agent skills: https://www.openagentskill.com/ai-agent-skills/rag-knowledge
- browser automation skills for AI agents: https://www.openagentskill.com/ai-agent-skills/browser-automation

Featured task routes:
- Scrape competitor pricing pages: https://www.openagentskill.com/tasks/scrape-pricing-pages
- Crawl a documentation site: https://www.openagentskill.com/tasks/crawl-documentation-site
- Review pull requests: https://www.openagentskill.com/tasks/review-pull-requests
- Fix CI failures: https://www.openagentskill.com/tasks/fix-ci-failures
- Convert PDFs to markdown: https://www.openagentskill.com/tasks/convert-pdf-to-markdown
- Build a RAG knowledge base: https://www.openagentskill.com/tasks/build-rag-knowledge-base
- Analyze CSV data: https://www.openagentskill.com/tasks/analyze-csv-data
- Automate a browser workflow: https://www.openagentskill.com/tasks/automate-browser-workflow
- Generate SEO content drafts: https://www.openagentskill.com/tasks/generate-seo-content
- Triage GitHub issues: https://www.openagentskill.com/tasks/triage-github-issues
- Scan a project for security risks: https://www.openagentskill.com/tasks/scan-security-risks
- Summarize research sources: https://www.openagentskill.com/tasks/summarize-research-sources

Coverage:
- Skill cluster pages: 12
- Task pages: 13
- Public skill pages are published listings, not a blanket guarantee of installability or safety. Discovered projects and unpublished candidates remain separate; search indexing has its own eligibility policy.

Install safety:
- Prefer /api/agent/resolve over raw search because it applies the OpenAgentSkill safety gate.
- Prefer /api/agent/resolve?format=lockfile when an agent needs a compact stable object for tool execution.
- Prefer /api/agent/receipt before installation because it gives a stable selected_skill, install policy, risk summary, alternatives, and outcome event id.
- Prefer /api/agent/packs/{slug} for multi-step workflows because it returns install_plan, review_checklist, audit URLs, and outcome_feedback.
- Prefer agent_handoff over ad hoc page scraping because it returns platform templates, API sequence, review checklist, and blocked actions.
- Prefer /api/agent/evals?slug={slug} before installation because it returns the Trust + Eval contract, blockers, and validation plan.
- Report success, failed, not_relevant, blocked_by_risk, or setup_required to /api/agent/outcome so future rankings learn from real use. Add quality fields when possible.
- Prefer /rankings/agent-proven or /api/agent/rankings?slug=best-by-success-rate when an agent needs skills with real reported adoption instead of only GitHub stars.
- Treat safety_gate.blocked as "do not auto-install".
- Treat safety_gate.experimental as manual test only.
- Treat safety_gate.reviewed as human-review-before-install unless auto_install_policy is allow.
- Review repository and license before running third-party code.
- Prefer sandboxed installation first.
- Do not perform external side effects without user approval.
- Use the audit and safety fields returned by the API.