autoreview
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONCREDENTIALS_UNSAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses
subprocess.runto interact with Git, GitHub CLI, and various LLM command-line tools (Codex, Claude, Pi). This is essential for retrieving code changes and performing automated reviews. - [DYNAMIC_EXECUTION]: The test suite (
scripts/autoreview_test.py) utilizesSourceFileLoaderto dynamically load and execute the mainautoreviewscript during unit testing. - [CREDENTIALS_UNSAFE]: The skill contains logic for detecting hardcoded secrets in code patches and includes an opt-in 'testbox' mode (
OPENCLAW_TESTBOX=1) to stage specific credentials for authenticated testing in isolated environments. - [INDIRECT_PROMPT_INJECTION]: As the tool processes untrusted code changes, it implements significant isolation strategies, such as empty workspaces and stripped environment variables, to prevent malicious instructions in the code from influencing the agent's behavior.
Audit Metadata