---

name: skill-authoring description: "Create, register, test, package, and maintain ClawBio skills and their intent descriptors." disable-model-invocation: true metadata: disco-role: operating license: NOASSERTION

ClawBio Skill Authoring

Use this sub-skill when a contributor must scaffold or modify a ClawBio skill, its SKILL.md, executable implementation, tests, demo fixtures, CLI registration, INTENTS.json, catalog metadata, or packaging contribution. Keep this sub-skill focused on the authoring lifecycle; domain algorithms and scientific-result interpretation belong to the domain skill and validation-safety.

Operating route

1. Read references/authoring.md before changing a skill.
2. Define a one-task scope, explicit trigger and do-not-trigger cases, input/output contract, safety boundary, synthetic demo, and registration plan.
3. Scaffold from the repository template; write tests before implementation and demonstrate the red/green cycle.
4. Register only a working executable. Keep the CLI alias, skill directory, script path, descriptor, and catalog distinctions explicit.
5. Use scripts/check_skill_assets.py with an explicit target-checkout root, then run focused tests and the appropriate public CLI/demo in that checkout.
6. Regenerate the catalog with the repository generator after frontmatter changes; inspect the diff rather than hand-editing generated JSON.
7. Hand off validation and scientific evidence to validation-safety.

Contracts to preserve

• Contributor skill folders are lowercase hyphenated identifiers; Python modules and tests use lowercase underscores.
• A contributor SKILL.md follows templates/SKILL-TEMPLATE.md; its name matches the folder, metadata is complete, triggers are loud and discriminating, and documented output is honest.
• Executable skills accept the stable --input, --output, and --demo contract, use synthetic data for demos, warn before overwriting, and include the exact ClawBio medical disclaimer in reports.
• clawbio.py's static SKILLS entry is the CLI registration and its allowed_extra_flags is the security boundary. An INTENTS.json descriptor is deterministic routing metadata, not a shell command or a privilege grant.
• Catalog JSON is generated output. Never repair a stale catalog by blindly editing the artifact; fix source metadata/registration, run the generator, and review.

References

• references/authoring.md — scaffold-to-tested workflow, trigger and I/O quality, demos, tests, registration, and contributor expectations.
• references/intent-descriptors.md — schema, route planning, slot extraction, path confinement, and blocked descriptor args.
• references/validation-and-packaging.md — checker/test commands, catalog generation, hatch package inclusion, and PR gates.
• references/troubleshooting.md — recovery for stale catalogs, missing scripts, invalid frontmatter, ignored descriptors, and package/test failures.

Static checker

From the repository root, run:

python sub-skills/skill-authoring/scripts/check_skill_assets.py \
  --root <ClawBio-checkout> --skill <ClawBio-checkout>/skills/<skill-name> \
  --catalog <ClawBio-checkout>/skills/catalog.json

The checker is read-only, reports actionable errors, validates static registry and catalog distinctions, and exits nonzero on a contract violation. Use --help to inspect options; it never generates or overwrites artifacts.