learn
Pass
Audited by Gen Agent Trust Hub on Feb 27, 2026
Risk Level: SAFEPROMPT_INJECTIONNO_CODE
Full Analysis
- [PROMPT_INJECTION]: The skill is designed to ingest and process untrusted information from external web sources and local files, which is a known vector for indirect prompt injection.
- Ingestion points: The procedure uses
WebFetchandWebSearchto gather data from the internet andReadto explore codebases in Step 1 (Survey) and Step 3 (Explore). - Boundary markers: The skill lacks specific instructions to treat external data as untrusted or to disregard commands found within analyzed materials.
- Capability inventory: The agent can read files, search the web, and write to local state files like
MEMORY.mdto consolidate findings. - Sanitization: No validation or filtering is applied to the retrieved content before it is integrated into the agent's mental model.
- [NO_CODE]: The skill consists entirely of natural language instructions and metadata, with no executable scripts or external code dependencies provided.
Audit Metadata