forget
Fail
Audited by Gen Agent Trust Hub on Oct 26, 2026
Risk Level: HIGHCREDENTIALS_UNSAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [CREDENTIALS_UNSAFE]: The skill's troubleshooting guide (
TROUBLESHOOTING.md) directs the agent to retrieve authentication secrets by reading sensitive local files:~/.agentmemory/.envand~/.agentmemory/secret. While it includes warnings against displaying or exfiltrating these secrets, instructing an AI agent to access local credential stores poses a significant risk of exposure. - [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied text to search through stored memories, which could lead to the agent ingesting and acting upon malicious instructions hidden within those memories.
- Ingestion points: User input provided via the
$ARGUMENTSvariable is used as thequeryfor thememory_smart_searchtool inSKILL.mdandEXAMPLES.md. - Boundary markers: The skill does not define specific delimiters or boundary markers to isolate the untrusted user query when it is passed to the search tool.
- Capability inventory: The skill possesses the capability to permanently delete data using the
memory_governance_deletetool as described inSKILL.md. - Sanitization: The skill mitigates risk by requiring "explicit confirmation" from the user before any deletion and instructing the agent to "Redact secret values" before displaying search results to the user.
Recommendations
- AI detected serious security threats
Audit Metadata