patch-sweep
Pass
Audited by Gen Agent Trust Hub on Oct 31, 2026
Risk Level: SAFEDATA_EXFILTRATIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The skill accesses project configuration files and environment variable settings (including
.envand platform-specific config vars) to identify runtime versions and apply mitigation flags. The instructions include a safeguard requiring the agent to list only keys and never print secret values. - [EXTERNAL_DOWNLOADS]: The skill fetches vulnerability data and exploitation signals from well-known security services, including the CISA Known Exploited Vulnerabilities (KEV) feed, the FIRST EPSS API, and the OSV vulnerability database.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted metadata from external source repositories and cloud platform APIs to generate a centralized status board, creating an attack surface for indirect instructions.
- Ingestion points: Project names, repository content, deployment logs, and domain metadata from GitHub, Vercel, Heroku, and Render APIs.
- Boundary markers: Not explicitly defined for the data interpolated into the status board report.
- Capability inventory: The skill possesses significant capabilities including
git push,git merge,vercel deploy, andheroku config:set. - Sanitization: No explicit sanitization or escaping is performed on the ingested application data before processing.
- [COMMAND_EXECUTION]: The skill executes local Node.js scripts (
enrich-advisories.mjsandvalidate-findings.mjs) included in the skill package to process and validate security findings data.
Audit Metadata