The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute git diff main...HEAD to retrieve the current changeset for review. This is a legitimate part of the code review workflow but involves executing shell commands based on branch names provided in the context.\n- [INDIRECT_PROMPT_INJECTION]: The skill has a potential surface for indirect prompt injection as it ingests untrusted data (git diffs or file contents) and incorporates them into prompts for multiple subagents without explicit sanitization or specific boundary instructions.\n
Ingestion points: Diff contents and file source code are gathered in SKILL.md (Step 1).\n
Boundary markers: The references/reviewer-prompt.md template places the untrusted content under a header but lacks explicit warnings for the subagent to ignore embedded instructions within the code.\n
Capability inventory: The skill uses the Agent tool to spawn reviewers and the main agent performs synthesis and judgment, potentially being influenced by adversarial subagent output. The agent also has shell execution capabilities (git diff).\n
Sanitization: No escaping or validation is applied to the code content before it is processed by the subagents.

interrogate