All skills
boshu2 avatar

/skill-builder

@860fce8
by Boboshu2/agentops446 stars
42

Create, adapt, consolidate or repair skill packages and projections. Use when: authoring guidance, descriptions or structure; Skill Eval measures behavioral benefit.

Use this Skill: https://skilld.dev/gh/boshu2/agentops/skill-builder

This session only. Nothing lands on disk.

referencesskill-template.md

≈641 tokens on demand. Your agent reads this file only when SKILL.md points to it.

Skill source authoring contract

Choose the smallest shape that communicates the actual behavior. No heading, helper, reference directory, role, output file or scoring target is mandatory. Canonical source uses AgentOps host metadata; generated Codex packages use the portable contract. See Codex parity.

A completed skill must make these meanings unambiguous, in prose or examples:

  • applicability and required inputs;
  • authority and possible effects;
  • the operation and its inline result or necessary artifact;
  • how completion is established;
  • what happens with missing inputs, failed operations or unavailable authority.

For example, a read-only adapter can say in one paragraph: “For a request to inspect the current branch, run git status --short in the caller-selected repository. Report the changed paths inline. Do not alter files or Git state. Finish after the command succeeds and the paths are reported; if the directory is not a repository or Git fails, report that error and stop.” This needs no validator script or output-file section.

The initializer creates SKILL.md with metadata defaults, placeholders and metadata.authoring_state: scaffold. Complete the behavior and remove that state before strict source checking. Deleting a marker cannot prove the prose complete: fresh semantic review is still necessary.

heal.sh --check --strict skills/<slug> checks identity, source fields, explicit scaffold state and required linked resources. It never mutates. heal.sh --fix skills/<slug> projects structurally valid explicit targets; it does not invent missing behavior. Invalid targets are refused before mutation.

audit.sh defaults to separate conformance, effects, behavior and located non-gating authoring evidence with no aggregate rank. audit.sh --legacy retains eight legacy lexical check IDs and its wire schema. For repo-runtime, their findings are advisory WARNs, including under --strict. Only source defects block that audit. Other consumers of the shared profile, including trigger CI, keep their existing policy. The external-observation profile keeps its legacy strict WARN behavior. Static PASS/WARN/FAIL describes only that selected mechanical operation, not semantic completion, safety, or effectiveness.

The package-readiness and craft scores are legacy advisory compatibility data. Do not add content to raise them. Use authoring doctrine and context density guidance as judgment aids.

Source: SKILL.md on GitHub

No alerts3d3 checks · Risk SAFE
  • Gen Agent Trust Hub3d

    The skill-builder is a robust developer tool for managing AI agent skill packages. It incorporates strong security practices, including clean-room verification to prevent indirect prompt injection, containment guards to prevent accidental file deletion, and safe parsing of YAML metadata.

  • Socket3d

    No alerts

  • Snyk3d

    Risk: LOW · No issues

Signed by skilld at 860fce8. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub yesterday.

Activeupdated last week

README badge

README badge for boshu2/agentops/skill-builder