Skip to content

Latest commit

 

History

367 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

OpenAgentSkill

OpenAgentSkill

The skill layer for AI agents.

Find, compare, audit, and install the right reusable Agent Skill before an agent acts.

CI GitHub Release GitHub Stars License

Try Resolve → · Install the CLI →


OpenAgentSkill task-to-skill resolver and registry

Why OpenAgentSkill?

Agent Skills are easy to publish and increasingly hard to evaluate. A repository can be popular, recently updated, or well documented without being the right choice for a specific agent task.

OpenAgentSkill adds the decision layer between discovery and execution:

Capability What an agent gets
Task-to-skill resolve One recommended Skill plus relevant alternatives
Trust and audit signals License, maintenance, install safety, permission, and quality context
Install receipt A stable, target-specific handoff for Codex, Claude Code, Cursor, or CLI
Outcome loop Evidence from successful, failed, blocked, or setup-required agent runs
Machine-readable surfaces JSON, text, OpenAPI, manifests, rankings, and public research datasets

OpenAgentSkill is not a static list and does not claim third-party code is safe. It helps agents make a better, inspectable decision before installation.

30-second demo

Resolve a real task:

curl "https://www.openagentskill.com/api/agent/resolve?task=extract+tables+from+PDF+reports&agent=codex&max_risk=medium&format=text"

Example response:

OpenAgentSkill Resolve
Task: extract tables from PDF reports
Best skill: <recommended skill>
Trust Score v5: <score and decision>
Install: <reviewed install command>
Risk: <risk level>
Alternatives: <ranked alternatives>
Receipt: <stable install handoff>

The same resolver is available as JSON, plain text, and a compact lockfile format.

Install the CLI

Run the pinned, dependency-free GitHub Release:

npx --yes https://github.com/Leon-Drq/openagentskill/releases/download/cli-v0.2.1/openagentskill-0.2.1.tgz resolve "extract tables from PDF reports" --agent codex

Useful commands:

# Search without installing
npx --yes https://github.com/Leon-Drq/openagentskill/releases/download/cli-v0.2.1/openagentskill-0.2.1.tgz search "browser automation"

# Inspect a reviewed install plan
npx --yes https://github.com/Leon-Drq/openagentskill/releases/download/cli-v0.2.1/openagentskill-0.2.1.tgz install <skill-slug> --agent codex --dry-run

# Read the outcome reporting contract
npx --yes https://github.com/Leon-Drq/openagentskill/releases/download/cli-v0.2.1/openagentskill-0.2.1.tgz outcome-contract

The CLI refuses blocked Skills, requires explicit confirmation before execution, and supports --no-telemetry. Source and usage details live in packages/cli.

Give this to an agent

Before installing a third-party Skill, call:

GET https://www.openagentskill.com/api/agent/resolve
  ?task={TASK}
  &agent={AGENT}
  &max_risk=medium
  &format=json

Inspect the recommended skill, alternatives, Trust Score v5, audit URL,
risk level, install command, and install receipt. Use a sandbox first.

After one narrow run, report the outcome to:
POST https://www.openagentskill.com/api/agent/outcome

Copy-ready templates for Codex, Claude Code, and Cursor are available in the Agent Integration Kit.

How it works

flowchart LR
    A["Agent task"] --> B["Resolve API"]
    B --> C["Candidate retrieval"]
    C --> D["Task fit + Trust Score v5"]
    D --> E["Best Skill + alternatives"]
    E --> F["Audit + install receipt"]
    F --> G["Sandboxed agent run"]
    G --> H["Outcome feedback"]
    H --> D
Loading

Ranking combines task relevance with repository evidence, install readiness, maintenance, license clarity, risk signals, and real agent outcomes. See Resolve Evals for the public evaluation surface.

Core interfaces

Interface Purpose
Resolve Workbench Turn a task into a recommended Skill and install plan
Skill Registry Search and filter indexed Skills
Rankings Compare trending, trusted, and Agent-Proven Skills
Audit Index Inspect install, maintenance, license, and risk evidence
Skill Packs Compose reviewed Skills into reusable workflows
Creator Kit Submit, claim, badge, and share a Skill
API Docs Integrate registry and agent endpoints

Machine-readable entry points:

TypeScript SDK

The dependency-free SDK source lives in packages/sdk:

import { OpenAgentSkill } from './packages/sdk/openagentskill.mjs'

const client = new OpenAgentSkill({
  baseUrl: 'https://www.openagentskill.com',
})

const plan = await client.resolve('audit a repository', {
  agent: 'codex',
  maxRisk: 'medium',
})

await client.reportOutcome({
  event_id: plan.feedback.event_id,
  skill_slug: plan.recommendation.best_skill.slug,
  task: plan.task,
  outcome: 'success',
  dry_run: true,
})

The SDK package is prepared for npm publication but is not presented here as published until the public registry release is verifiable.

For Skill authors

Paste a repository, subdirectory, or SKILL.md URL into Submit Skill. Submissions are saved first and reviewed asynchronously; GitHub stars, a predefined category, and perfect metadata are not required.

After indexing, authors receive:

  • A canonical Skill page and machine-readable metadata.
  • Trust, audit, quality, and Agent-Proven badge endpoints.
  • Claim and verified-maintainer paths.
  • Alternatives and use-case pages that can send qualified traffic back to the source.

Browse the curated GitHub Skill Index for domain and scenario maps.

Trust and security

OpenAgentSkill never treats popularity as proof of safety. Trust Score and audits are decision-support signals, not certifications or guarantees.

Before executing third-party code:

  1. Read the source and install scripts.
  2. Review permissions, network calls, dependencies, and required secrets.
  3. Start in a sandbox or low-risk workspace.
  4. Keep production credentials and customer data out of unreviewed Skills.

Report vulnerabilities privately through the repository's Security tab. Report risky or misleading third-party listings through the Skill data issue form. See SECURITY.md for scope and disclosure rules.

Local development

Requirements: Node.js 22 and pnpm 10.28.2.

git clone https://github.com/Leon-Drq/openagentskill.git
cd openagentskill
corepack enable
pnpm install
cp .env.example .env.local
pnpm dev

Quality checks:

pnpm run lint
pnpm run typecheck
pnpm test
pnpm run build

The public website requires Supabase configuration. Pure parsing, resolver-contract, SDK, CLI, and circuit-breaker regression tests run without production credentials.

Repository map

app/          Next.js pages, APIs, manifests, reports, and agent surfaces
components/   Product UI and client interactions
lib/          Resolve, ranking, audit, indexer, database, and growth logic
packages/     Publishable CLI and SDK packages
scripts/      Regression tests, migrations, importers, and maintenance jobs
skills/       Curated GitHub Skill index by domain and scenario
supabase/     Database functions and migrations

Contributing

Contributions are welcome across product, APIs, indexing, trust, documentation, and Skill metadata.

Project decisions and maintainer responsibilities are documented in GOVERNANCE.md. Releases are tracked in CHANGELOG.md, and planned work lives in ROADMAP.md.

License

MIT © OpenAgentSkill contributors.

Releases

Packages

Used by

Contributors

Languages