The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because its 'DISCOVER' phase involves reading untrusted content from the repository's documentation directory, configuration files, and performing external web research to gather implementation context.
Ingestion points: Files in docs/ (e.g., OVERVIEW.md, GETTING_STARTED.md), .codedocs.json, and results from web-based research agents.
Boundary markers: None explicitly defined in the instructions to prevent the agent from following instructions embedded within these external sources.
Capability inventory: Extensive capabilities including modifying project source code, creating files, and executing shell commands (test, lint, build) via the agent's environment.
Sanitization: No sanitization or validation of the ingested documentation or web content is specified before the data is used to generate execution plans.
[COMMAND_EXECUTION]: The skill instructs the agent to execute various shell commands to detect project conventions and verify task completion. This includes running directory listings on sensitive agent configuration paths (e.g., ~/.claude/skills/) and executing project-specific scripts found in package.json or Makefile (e.g., npm test, pytest, make build).
[EXTERNAL_DOWNLOADS]: The skill documents its installation using the npx skills add command and contains logic to automatically detect missing recommended companion skills and suggest their installation from the same vendor.

super-human