The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes ls ~/.claude/skills/ to determine which specialized skills are currently installed, allowing it to provide accurate installation advice.
[EXTERNAL_DOWNLOADS]: The skill is configured to fetch skill definitions and metadata from the vendor's GitHub repository and other specified third-party sources to ensure routing prompts are accurate and up-to-date.
[PROMPT_INJECTION]: The skill processes external metadata from various sales tool repositories to construct ready-to-use prompts. This constitutes an indirect prompt injection surface.
Ingestion points: External SKILL.md files are retrieved via WebFetch from GitHub in SKILL.md (Step 2.5).
Boundary markers: No explicit boundary markers or instructions to ignore embedded commands are present in the routing prompt generation logic.
Capability inventory: The skill can list local files (ls) and write new instruction files (do-*.md) to the filesystem as part of its routing functionality.
Sanitization: There is no evidence of sanitization or filtering of the content retrieved from external repositories before it is incorporated into prompts.

sales-do