The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions and workflow utilize git diff and gh repo view to gather information about local branch changes. These are standard development tools used within the expected scope of code auditing and do not indicate malicious intent.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external, untrusted data in the form of source code diffs and files during the audit process.
Ingestion points: Untrusted data enters the agent context through the execution of git diff and individual file reads as described in the "Phase 1: Complete Input Gathering" section of SKILL.md.
Boundary markers: The instructions do not define specific delimiters or "ignore embedded instructions" warnings to prevent the agent from accidentally following instructions hidden within the code it is auditing.
Capability inventory: The skill is configured to use tools such as codex-cli, claude-code, and gh (GitHub CLI), providing the agent with the ability to interact with the repository and potentially the network.
Sanitization: There is no evidence of sanitization, escaping, or filtering of the external code content before it is interpolated into the agent's prompt for review.

find-bugs