capability-atlas

Identify the canonical function first, then render it in the target host's native form while keeping the trigger meaning and input/output contract equivalent. Record any host-specific differences explicitly and verify with parity tests.

Use the capability matrix template with fields for function, trigger, inputs, outputs, host renderings, canonical location, parity tests, and drift risks. Fill one row per capability to track what is stable versus host-specific.

A skill covers procedural interactive behavior, an agent covers autonomous role-based behavior, a hook handles event-driven actions, and a command handles explicit user invocation. Choose the form matching how the capability is triggered.

Run parity tests confirming the same trigger phrase activates both renderings, the same input produces the same outcome, and the same guardrails and failure modes apply. Sample test prompts are provided in the parity tests reference.

Renaming is safe only when the behavior contract, trigger meaning, and verification method stay intact. If a host wrapper would silently change the meaning or drop test coverage, document the drift instead of claiming parity.