The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses the ghq tool to download external codebases from GitHub, a well-known service, to enable automated analysis and documentation.
[COMMAND_EXECUTION]: The skill executes several shell commands to manage its workflow, including ghq for repository cloning, ln for creating symlinks, mkdir for directory structure, and text processing tools like sed, grep, and sort for manifest management.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process content from external, untrusted codebases, which creates an inherent surface for indirect prompt injection attacks where malicious code or comments could attempt to override the instructions of the analyzing sub-agents.
Ingestion points: Source code files are read from the origin/ symlink directory which points to the cloned external repository as described in SKILL.md.
Boundary markers: The instructions provide distinct READ and WRITE literal paths to separate source code from generated documentation, though the prompt templates do not explicitly include "ignore embedded instructions" warnings for the sub-agents.
Capability inventory: The skill uses file system management tools (mkdir, ln), repository management (ghq), and basic text utilities (sed, find, sort, grep). It does not execute the target code itself.
Sanitization: There is no evidence of content sanitization or filtering applied to the source files before they are read by the Haiku agents.

learn