The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources, which presents a surface for indirect prompt injection attacks.
Ingestion points: According to SKILL.md, the skill ingests Pull Request bodies, issues, and retrieved text for analysis.
Boundary markers: The agent is explicitly instructed to treat PR bodies and issues as untrusted evidence and never as execution authority (SKILL.md line 39).
Capability inventory: The skill's capabilities include reading codebase files, writing HTML artifacts to a dedicated directory (.tigerkit/), and executing local Git commands.
Sanitization: Reference instructions in references/html-output.md mandate escaping source and code text for use within HTML or script contexts.
[COMMAND_EXECUTION]: The skill executes Git commands to ensure its generated artifacts are correctly ignored by the version control system.
Evidence: The file references/artifact-paths.md instructs the agent to run git ls-files and git check-ignore to verify tracking status and potentially append the .tigerkit/ path to the .gitignore file.

tk-explain-diff