The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted data from external sources, which constitutes an indirect prompt injection surface. \n
Ingestion points: Data is ingested from the repository's local files (SKILL.md, references/*), internal knowledge base search results (mcp__kb__kb_search), GitHub API responses (mcp__surf__surf_github_get), and external web documentation accessed via WebFetch.\n
Boundary markers: The skill implements explicit 'GATE 1' and 'GATE 2' banners that require affirmative user response. Instructions to subagents are provided 'verbatim' with specific formatting requirements.\n
Capability inventory: The skill can perform file writes (Edit, Write), repository management (git commit, git push, gh pr create), and local task execution (just check).\n
Sanitization: Human approval is required before any side effects (edits or commits). Subagents are strictly instructed to return verbatim quotes from sources to prevent the introduction of malicious or hallucinated instructions.\n- [COMMAND_EXECUTION]: The skill executes local commands and utilizes project-specific tools to manage the repository state.\n
It uses git for repository status, diffing, and version control operations.\n
It uses gh (GitHub CLI) for interacting with Pull Requests and monitoring CI workflows.\n
It runs just check to validate project integrity and regenerate documentation files.

update-skill