The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions direct the agent to execute various shell commands and scripts for repository maintenance, including npm run sync:skills, bash scripts/generate-og-image.sh, and directory traversal loops using sed and grep for validation.
[EXTERNAL_DOWNLOADS]: The documentation recommends the installation of system dependencies like imagemagick via Homebrew and references assets from well-known services such as Google Fonts. These are standard development practices and target trusted sources.
[INDIRECT_PROMPT_INJECTION]: The skill operates by ingesting and processing documentation (SKILL.md, guides, references) from multiple other skill directories to sync them with a central website. This represents an attack surface where a malicious skill could include hidden instructions in its markdown or frontmatter to influence the documentarian's behavior.
Ingestion points: .claude/skills/*/SKILL.md and associated subdirectories (references/, guides/, templates/, examples/).
Boundary markers: The skill uses basic escaping for angle brackets during MDX conversion, but lacks comprehensive delimiters to isolate processed content from its own instructions.
Capability inventory: The skill has broad capabilities including file read/write access and the ability to execute shell commands.
Sanitization: There is minimal sanitization beyond preventing MDX compilation errors, leaving the agent potentially susceptible to instructions embedded in the processed markdown data.

skill-documentarian