recap
Fail
Audited by Gen Agent Trust Hub on Oct 26, 2026
Risk Level: HIGHCREDENTIALS_UNSAFEDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [CREDENTIALS_UNSAFE]: The TROUBLESHOOTING.md file provides explicit instructions for the agent to search for and read sensitive credential files on the local filesystem, specifically ~/.agentmemory/.env and ~/.agentmemory/secret, to handle tool failures.
- [DATA_EXFILTRATION]: The troubleshooting workflow involves reading local secrets and including them in HTTP requests to a target specified by the AGENTMEMORY_URL environment variable. This logic creates a potential path for credential exfiltration to unauthorized endpoints.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted session data. 1. Ingestion points: session titles from memory_sessions and observations from memory_recall (SKILL.md). 2. Boundary markers: absent. 3. Capability inventory: tool calls (SKILL.md) and network requests/file reads via REST fallback (TROUBLESHOOTING.md). 4. Sanitization: absent.
Recommendations
- AI detected serious security threats
Audit Metadata