judge-ctf
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to evaluate its success by searching through external data sources, such as agent output logs, evidence files, and JSON reports (
vulns.json), which could contain attacker-controlled content designed to influence the agent's logic. - Ingestion points: The agent is directed to read and search content from
vulns.jsonand external logs/evidence files generated during a CTF challenge. - Boundary markers: The instructions lack explicit boundary markers or directives to ignore instructions that might be embedded within the files being searched.
- Capability inventory: This skill file is a markdown-based guidance document and does not contain direct tool calls or script execution, though it guides the agent in using other available capabilities.
- Sanitization: There is no mention of filtering or sanitizing the data read from logs or evidence files before the agent processes them.
Audit Metadata