autoresearch
Pass
Audited by Gen Agent Trust Hub on May 7, 2026
Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill is designed to autonomously execute a
verifycommand specified in abrief.mdconfiguration file. This is the core functionality of the 'autoresearch' pattern, allowing the agent to run arbitrary shell commands to measure metrics. - [COMMAND_EXECUTION]: A command injection vulnerability exists in the example script
examples/prose-readability/autoresearch.sh. The script iterates over filenames and interpolates them directly into a Python command string usingpython3 -c "... open('$f').read() ...". If a file in the repository has a malicious name (e.g., containing a single quote followed by Python code), it could lead to arbitrary code execution when the verification script is triggered. - [SAFE]: The skill implements protective 'frozen' scopes to prevent the agent from modifying sensitive configuration files like
package.jsonor lockfiles during its autonomous cycles.
Audit Metadata