Explicit invocation

Run these checks in fresh Claude Code, Codex, and Cursor sessions after installing the skills and reloading discovery. These are live checks; file validation alone does not establish that a harness enforces its invocation settings.

Ordinary request

Ask: “Fix the typo in this README.”

Expect ordinary assistance without loading HolyCow, starting Refine, or opening a planning workflow. Check the harness’s context or skill-use display where available. Listing a skill in its picker is not evidence of automatic invocation.

Explicit request

Invoke /hc-refine in Claude Code or Cursor, or $hc-refine in Codex, with a small new idea. Expect the Refine instructions to load and recommendations to accompany the questions. Repeat with each of the other installed HolyCow skills.

Selected workflow with companions

Invoke Build for a small code change that needs review. Expect Build to read the installed Review instructions as needed within that requested task. It must not load every HolyCow skill, publish a PR without authorization, or treat a manual-only companion as missing merely because automatic invocation is disabled.

Repeat with Compass and a discussion question: Refine can be read as supporting guidance; Lab and Plan should remain unloaded unless the requested work needs them.

Fresh session after use

Start another session and make an ordinary request. HolyCow should not become an automatic default because it was used previously. Within the original conversation, already loaded instructions may remain in context; do not claim they were removed.