| name | skillet-authoring |
| description | Authors, improves, or migrates agent skills with the Skillet CLI; use when asked to create or write a skill, improve its instructions or evals, diagnose failing evals, or migrate a legacy SKILL.md, uppercase SPEC.md, or spec.yaml skill, but not when merely using an existing skill. |
| spec_hash | 3b86b9ec11a6 |
Skillet Authoring
Run every Skillet command through the current package: npx -y @sentry/skillet@latest, or pnpx @sentry/skillet@latest in a pnpm environment. Never prefer a bare skillet executable from PATH; if one reports an update, rerun that command through the explicit latest package before continuing. Treat spec.md as the behavior contract, SKILL.md as the agent instructions, and eval cases as repeatable runs of the spec scenarios. Never write these files from a remembered format.
Steps
- Run
npx -y @sentry/skillet@latest status <dir> --json and do what next says. For a brand-new skill, run npx -y @sentry/skillet@latest new <name> first. Never guess a skill's state or start over when artifacts already exist.
- When status reports uppercase
SPEC.md, preserve or rename that legacy document before creating lowercase spec.md; never parse it as the active Skillet spec.
- When status marks lowercase
spec.md invalid, preserve or rename legacy content and derive a valid Skillet spec before rendering SKILL.md or adding coverage.
- When migrating, inventory behavior-bearing material before drafting: triggers, ordered workflow, exact lists, protocols and output formats, thresholds, failure and stop rules, constraints, runtime references, and maintenance docs that describe active behavior. Every accepted behavioral rule must land in ; verbose execution detail may additionally remain in a linked runtime reference after the spec defines the observable contract. Explicitly supersede or reject non-behavior content.
