The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill documentation (SKILL.md and BOOTSTRAP.md) references mattpocock/skills as a dependency. It provides instructions for the user to install these skills into the repository's skills directory, specifically recommending that they be pinned to a reviewed revision to ensure security.
[INDIRECT_PROMPT_INJECTION]: The skill contains explicit guardrails in the 'Agent Rules' section of SKILL.md. It instructs the agent to treat external content (issue bodies, pull requests, comments, etc.) as untrusted data that cannot override system instructions or user requests.
Ingestion points: External data enters the context via issue trackers, PR comments, and research files (documented in SKILL.md section 7).
Boundary markers: The skill instructs the agent to explicitly route work through authoritative documentation and treat all other inputs as data.
Capability inventory: The skill can perform file writes, symbolic link creation (ln -s), and Git operations (git add, git commit) as part of its bootstrap and governance workflow (documented in BOOTSTRAP.md).
Sanitization: The agent is instructed to stop before any mutation if a conflict is detected and to hand the decision to a human product owner.
[COMMAND_EXECUTION]: BOOTSTRAP.md contains standard shell commands for creating symbolic links and managing Git state. These are used for project initialization and do not involve execution of untrusted or remote code.

docs-governance