Compositions

Skills + Project Instructions

The repository contract routes a task to a skill; the skill carries the conditional procedure. This keeps always-loaded context small without making the workflow optional.

Skills + Commands

Expose a manually invoked command as a thin entrypoint to the same skill body. Do not maintain separate command and skill procedures that can drift.

Skills + Subagents

A skill can delegate bounded research, implementation, or review to specialized agents. Define ownership and result schemas before fan-out; aggregate evidence in the parent.

Skills + Hooks

Hooks detect a lifecycle condition and point to a remediation skill. Keep automatic invocation bounded so a failed check cannot create an infinite model loop.

Skills + MCP

The skill defines when and how to use an external capability; MCPMCPUSB-C for AI — one standard plug so any agent can connect to any data source or tool.The Model Context Protocol: an open standard (Anthropic, 2024) for connecting AI systems to external tools and data without building a custom integration for each one.Full definitionClose supplies the tool. Credential scope and service authorization remain outside the skill.

Skills + Evidence

Require the workflow to emit a machine-readable summary: inputs, decisions, checks run, failures, and artifact paths. Validate that summary before a downstream gate accepts it.

The strongest composition keeps knowledge in skills, deterministic work in scripts, authority in policy, and final judgment with an accountable human.