anti-deception
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONNO_CODE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructions facilitate the ingestion of user-provided requests to generate queries for the
anti-deceptiontool, whose output is then incorporated into the agent's reasoning. - Ingestion points: User-supplied requests are processed and summarized as tool input in
SKILL.md. - Boundary markers: The skill does not define specific structural delimiters for tool inputs but relies on the agent to create a sanitized summary.
- Capability inventory: The skill calls the
anti-deceptiontool from theejentumMCP server to process the query. - Sanitization: The instructions include a clear mandatory warning prohibiting the transmission of private project details, identities, credentials, or sensitive claims to the external server, recommending a sanitized summary instead.
- [NO_CODE]: The skill consists exclusively of markdown instructions and metadata, without any accompanying scripts, executables, or configuration files that could execute logic independently of the agent's core processing.
Audit Metadata