Agent Skills

Runtime for composable AI agent skills — deterministic execution of capability-based workflows.

Quick navigation

• New here? Start with Installation and the 10-Minute Onboarding
• Build a skill: Tutorial: First Skill → Skill Authoring Guide
• Integrate: Environment Variables · MCP Integration · OpenAPI Construction
• Troubleshoot: Troubleshooting · Error Taxonomy
• Architecture: Target Architecture (Canonical) · Runner Guide · Scheduler · CognitiveState
• Durability: Durability Contract
• Policy/Tenancy: Policy Decision Contract
• Policy Bundles: OPA Policy Bundle Lifecycle
• Governance Ops: GitHub Ruleset Runbook
• Production Ops: Production Readiness Playbook
• Deployment Flow: Deployment Product Flow
• Release Use Cases: Public Release Use Cases
• 100% Plan: Product 100% Completion Plan
• Release Exceptions: Release Exceptions Policy
• Release Lineage: Release Lineage Model
• Release Gate SLO Policy: Release Gate SLO Policy
• Target Architecture RFCs: RFC-0001 · RFC-0002 · RFC-0003 · RFC-0004 · RFC-0005 · RFC-0006 · RFC-0007
• Progress: Target Architecture Progress Snapshot
• Readiness: Target Architecture Merge Readiness

What is this?

Agent Skills is a deterministic execution engine for AI agent capabilities. It lets you:

1. Define atomic capabilities (summarize, classify, extract, translate…)
2. Compose them into multi-step skills (workflows)
3. Execute skills via CLI, embedded SDK, MCP, HTTP REST, or LLM tool integrations
4. Observe every step via structured traces, webhooks, and metrics

See the README for the full overview.