The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool to perform extensive file system operations and validation routines.
It executes shell loops involving sed, grep, and cut to parse YAML frontmatter from various SKILL.md files.
It automates the execution of local build and synchronization scripts, such as npm run sync:skills, bash scripts/generate-og-image.sh, and npx tsx scripts/syncSkillSubpages.ts.
[PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection (Category 8) due to its core function of ingesting and processing content from other skills.
Ingestion points: The agent reads markdown content and metadata from the .claude/skills/ directory, including subfolders for references, templates, examples, and guides.
Boundary markers: The prompt lacks explicit delimiters or instructions to prevent the agent from obeying commands that might be embedded in the metadata of the skills it documents.
Capability inventory: The skill is granted Bash, Write, and Edit permissions, providing a high-impact execution environment if an injection occurs.
Sanitization: While the skill mentions escaping angle brackets to ensure MDX compatibility, it does not implement comprehensive sanitization against adversarial instructions in the source skill files.

skill-documentarian