The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests external code, diffs, and project files to perform reviews, creating a surface where malicious instructions embedded in the code could influence the behavior of the reviewer subagents.
Ingestion points: SKILL.md (Step 1) identifies and packages diffs (git diff) and file contents from the user's environment to be reviewed.
Boundary markers: The prompt template in references/reviewer-prompt.md uses Markdown headers to delimit sections, but does not provide explicit instructions or delimiters to prevent subagents from obeying instructions found within the code being reviewed.
Capability inventory: The skill utilizes the Agent tool to spawn subagents, reads local configuration files (~/.claude/pstack-models.md), and executes shell commands for version control scope.
Sanitization: There is no evidence of filtering, escaping, or sanitization of the code content before it is interpolated into the reviewer prompts.

interrogate