Skip to main content
Reference — generated from the toolkit

Source of truth: toolkit/skills/_skill-spec-template.md. Edit it there; this page is regenerated on build.

Skill: {name}

  • Owner: {team or person}
  • Version: 0.1.0
  • Lifecycle: draft | trial | proven | deprecated
  • Phase: definition | design | tech-selection | build | refinement | flow-state | production | cross-cutting

Trigger

When this skill should be invoked (the repeated task it replaces). If a prompt is being pasted for the third time, that's the signal to make it a skill.

Inputs → outputs

  • Inputs: what the skill needs (files, parameters, context).
  • Outputs: what it produces, in what shape (be precise — downstream steps depend on it).

Procedure

The steps the skill performs. Deterministic where it can be; note where it uses judgement.

Guardrails & permissions

  • Permissions required (filesystem/network/tools) — least privilege per the guardrail standard.
  • Actions it must not take (anything irreversible is out of scope — it prepares, a human authorises).

Failure modes

Known ways it goes wrong and how they're detected/handled.

Golden scenarios

The fixed test cases the skill is validated against in the evaluation harness. List them here; store the fixtures in golden/.

ScenarioInputExpected outputPass criterion