regression-sweep
Warn
Audited by Gen Agent Trust Hub on Jun 13, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [COMMAND_EXECUTION]: The skill's primary workflow involves the automated execution of 'python3 poc.py' for every entry discovered in the 'findings/' directory tree. This results in arbitrary code execution on the agent's host system whenever the sweep is triggered.
- [REMOTE_CODE_EXECUTION]: Because the skill discovers and runs scripts dynamically from the filesystem (globbing 'findings/finding-NNN/'), it is vulnerable to code injection. If an attacker can influence the contents of the findings directory (e.g., through a separate task or repository write access), they can achieve code execution during the next scheduled sweep.
- [DATA_EXFILTRATION]: The Proof of Concept (PoC) scripts are designed to interact with network targets to verify vulnerabilities. The skill does not implement network isolation or egress filtering, which could allow a malicious or compromised PoC to exfiltrate local environment variables, credentials, or sensitive files from the agent's environment under the guise of an exploit verification.
- [COMMAND_EXECUTION]: The skill utilizes vendor-specific utility scripts such as 'tools/nvd-lookup.py' from transilienceai. The overall security of the re-validation process is dependent on the integrity of these local tools, which are executed as part of the automated workflow.
Audit Metadata