Reference — generated from the toolkit

Source of truth: toolkit/skills/README.md. Edit it there; this page is regenerated on build.

Skills library

Repeatable agent capabilities we invoke, not reinvent. Every skill declares a spec and ships with golden test scenarios — no spec and tests, not published. This is what separates "capabilities we invoke" from a folder of scripts.

The spec template is seeded, plus two STOP-critical skill specs: code-archaeology/ (understand a legacy repo fast) and characterisation-tests/ (pin legacy behaviour before changing it). The rest of the seed set grows from demonstrated repetition.

Per-skill contents:

• Spec: trigger conditions, inputs/outputs, procedure, guardrails and permissions required, failure modes, owner, version.
• Golden scenarios: the fixed tests the skill is validated against in the evaluation harness.