Scaffold and review Agent Skill directories that pass repository quality gates.
npx skills add https://github.com/dhanesh/agent-skills --skill repo2skill-dhanesh
Please help me install this Agent Skill. Skill: repo2skill Source: https://github.com/dhanesh/agent-skills/tree/main/repo2skill Command: npx skills add https://github.com/dhanesh/agent-skills --skill repo2skill-dhanesh
assets/scaffold_skill.py generates a complete skill directory (SKILL.md, README.md, references stub, smoke-test suite, eval stub) that passes validate-skill.sh, prompting-playbook.sh, and frontmatter-standard.sh on the first run. It refuses non-kebab-case or over-long names and never overwrites an existing directory. - Guided authoring: references/authoring-guide.md walks file-by-file from skeleton to shippable, replacing every TODO(repo2skill) marker with real content. - Semantic review: references/semantic-review.md provides the reviewer checklist for substance the mechanical gates cannot judge — orthogonal sections, real deliverable contracts, justified absolutes, and honest evals. - Use Case: You want to add a new skill to an agent-skills repository. Run the scaffolder, confirm make gate-skill SKILL=my-new-skill is green, author the content under gate cover, run the semantic review, and iterate until both the gate and the review pass. ## Quick Start Ask the agent to scaffold a new skill named my-new-skill at the repo root and iterate the quality gates until they pass.High-intent search queries and answers about installing and using this skill.
Run python3 assets/scaffold_skill.py <name> --dir <repo-root> to generate a skeleton that is already green against the repo's gate scripts. Then replace every TODO(repo2skill) marker following the authoring guide and re-run make gate-skill SKILL=<name> until green.
The scaffolder emits SKILL.md with the full standard frontmatter block — kebab-case name, description under 1024 characters, license, compatibility, and author/version/tags metadata. You edit values, not structure, so the frontmatter-standard gate stays green.
Skill names must be kebab-case matching ^[a-z0-9]([a-z0-9-]*[a-z0-9])?$ and at most 64 characters, identical to the directory name. The scaffolder refuses invalid names like Bad_Name and writes nothing when it refuses.
The scaffolder runs anywhere python3 does, stdlib only and offline. The gate loop additionally needs make and a POSIX shell, and full gate coverage requires a repo that vendors the same scripts/gates quality checks.
A fresh scaffold passes the gates, so any red check after that points at your own edit — commonly a dangling references/ or assets/ path, an over-long description, or broken body structure. Fix the named check and re-run make gate-skill.
Do not use it to audit whether a whole repo is ready for coding agents (use agent-ready-rails), to fact-check a codebase's claims (use base-in-reality), or to design a single agent loop (use crafting-self-prompting-loops). It owns only one skill directory versus the repo standard.
RorySullivan1Guides authoring of Claude Code skill bundles with triggering descriptions and structured SKILL.md bodies.
aaronjmarsPackages an existing skill as a public GitHub community pack and submits a registry PR.
megaeth-labsVerify documentation claims against source code, configs, and deployments.
sonthanhValidate brain-os skill eval definitions against SKILL.md content.