using-agent-skills

Route AI agent tasks by matching delivery phase and intent to repository-defined skill names. This meta-skill directs the session to the appropriate engineering workflow based on specific contexts like implementing, testing, reviewing, or shipping.

Phase-based skill routing in OpenCode maps user intents such as define, plan, build, verify, review, and ship to specific workflow skills. It ensures the AI follows repository process guardrails rather than improvising ad-hoc execution.

Set up skill discovery by installing or linking skills into OpenCode discovery paths. This enables the meta-routing rules to correctly identify and invoke the best matching engineering workflow based on evidence and verification.

Yes, you can enforce evidence verification by applying meta-routing rules that require evidence rather than ad-hoc execution. This consolidates repo-wide operating behaviors like verifying evidence and making surgical changes.

Skill routing works for detailed scenarios including mobile, web, shader, testing, docs, and artifacts. It maps these specific intents to the appropriate engineering skill to maintain consistent guardrails across different domains.

Use a meta-skill at session start or whenever you need to rediscover the best workflow for a current problem. It prevents improvisation by ensuring the AI correctly matches the task to the repository's predefined process guardrails.