two-axis-review
Pass
Audited by Gen Agent Trust Hub on Oct 14, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute several git commands (rev-parse, diff, log, show) to analyze the repository state, retrieve changes, and examine commit history. It also uses search tools like 'grep' and patching tools to verify the exact locations of code changes in the source files.
- [INDIRECT_PROMPT_INJECTION]: The skill operates on untrusted external inputs including git diffs, requirement specifications, and coding standard documents. There is a potential risk that these inputs could contain instructions designed to deceive the reviewer or suppress valid findings.
- Ingestion points: Reads contents from 'git diff', specification files (e.g., 'spec.md'), and local repository documentation ('CLAUDE.md', 'CODING_STANDARDS.md').
- Boundary markers: The process is strictly partitioned into separate axes (Standards and Spec) with specific prompts and a required findings template to maintain structured analysis.
- Capability inventory: The agent utilizes shell commands ('git', 'grep'), file read operations, and file write operations to an optional knowledge vault.
- Sanitization: The skill includes a mandatory 'Self-check' step that requires every finding to be supported by explicit evidence (quotes from the code or docs) and verified line numbers, which helps prevent the agent from following instructions embedded in the analyzed data.
Audit Metadata