cross-review
Pass
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill invokes external CLI tools including
claude,codex,jq,python3, andteeto perform code reviews and process data streams. - Evidence: Found in
SKILL.md(invocation commands) andscripts/run-claude-review.sh. - [DYNAMIC_EXECUTION]: The utility script
scripts/run-with-deadline.pyusessubprocess.Popento launch and manage the lifecycle of external commands, specifically theclaudeCLI in this context. - Evidence:
scripts/run-with-deadline.pyat line 34. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted repository content such as diffs, branches, and untracked files, which are then passed to a secondary reviewer agent.
- Ingestion points: Files in the repository, git diffs, and pull request metadata as defined in the handoff section of
SKILL.md. - Boundary markers: The handoff template uses structured sections (e.g., 'Intent and acceptance criteria'), but the secondary agent is instructed to 'Read the whole diff', which may contain malicious instructions targeting the reviewer.
- Capability inventory: The skill can execute shell commands via scripts and interact with the filesystem.
- Sanitization: No explicit sanitization or filtering of the repository content is performed before passing it to the reviewer agent.
Audit Metadata