audit-agent-harness
Pass
Audited by Gen Agent Trust Hub on Jul 20, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSDATA_EXFILTRATIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: Fetches service discovery metadata from 'https://cristianmoroaica.github.io/bountyverdict/agent-manifest.json'.\n- [DATA_EXFILTRATION]: Public repository URLs are transmitted as query parameters to an external API endpoint specified in the downloaded manifest.\n- [COMMAND_EXECUTION]: Instructs the agent to interact with a payment client to authorize blockchain transactions on the Base network using specific USDC contract addresses.\n- [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it requires the agent to process and obey external API outputs, including instructions labeled as 'recommendations', 'findings', and status codes like 'REPAIR' or 'REVIEW' that dictate subsequent autonomous actions without sanitization or boundary markers.
Audit Metadata