openrouter-debug-bundle
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_CONTEXT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from an external API (OpenRouter) which could contain malicious instructions if the upstream provider or model is compromised.
- Ingestion points: In
SKILL.md, thedebug_requestfunction capturesresponse.choices[0].message.contentinto aDebugBundle. Similar ingestion occurs inreferences/examples.mdandreferences/request-logging.md. - Boundary markers: The instructions do not define boundary markers or instruct the agent to ignore instructions embedded within the API responses.
- Capability inventory: The skill has access to
Bash(includingpython3andnodeexecution) andWritetools, which could be targeted by a successful injection. - Sanitization: There is no evidence of sanitization or filtering of the response content before it is stored or displayed.
- [DYNAMIC_CONTEXT_INJECTION]: The skill uses the dynamic context injection syntax to gather environment information at load time.
- Evidence:
SKILL.mdcontains!node --version 2>/dev/null || echo 'N/A'and `!`python3 --version 2>/dev/null || echo 'N/A'. These commands are benign as they only report software versions for troubleshooting purposes.
Audit Metadata