The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches template files and configuration manifests from the patinaproject/skills repository on GitHub using git clone and the GitHub API. These downloads are from the official organization of the skill author.
[COMMAND_EXECUTION]: The skill executes several system commands to manage repositories, including git for cloning and staging, gh for interacting with the GitHub API to read and update repository settings, and pnpm for dependency management and running local tools.
[PERSISTENCE]: The skill sets up automated hooks, including Husky Git hooks (commit-msg, pre-commit) and Claude Code SessionStart hooks in .claude/settings.json. These mechanisms are standard for the Patina Project to automate linting and environment setup upon opening the repository.
[INDIRECT_PROMPT_INJECTION]: The skill has an indirect prompt injection surface as it ingests untrusted data from existing files during repository audits. This is documented with a clear evidence chain:
Ingestion points: The skill reads existing repository files and configuration manifests like core-baseline.txt and audit-checklist.md (found in SKILL.md and audit-checklist.md).
Boundary markers: The skill instructions explicitly require showing unified diffs for every proposed change.
Capability inventory: The skill has file-write access and can execute shell commands via pnpm, git, and gh (documented in SKILL.md).
Sanitization: The skill implements a strict human-in-the-loop requirement, asking the user to accept, skip, or defer each individual recommendation before applying changes.

scaffold-repository