expo-review
Pass
Audited by Gen Agent Trust Hub on Nov 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external data including pull request content and git diffs, which provides a surface for indirect prompt injection.
- Ingestion points: GitHub pull request metadata, body text, and file diffs (SKILL.md).
- Boundary markers: The instructions explicitly command the agent to treat all source data and PR text as untrusted and never to obey instructions contained within them.
- Capability inventory: The skill uses Bash to run a local reviewer script and git commands.
- Sanitization: The skill requires summarizing progress and results rather than echoing raw, potentially malicious content.
- [COMMAND_EXECUTION]: The skill executes local repository scripts and git tools using user-supplied arguments.
- Evidence: It uses
./scripts/expo-code-reviewandgit rev-parse. - Mitigation: It implements strict validation logic for PR numbers, URLs, and git refs, including the use of
--end-of-optionsto prevent option injection.
Audit Metadata