topic-deep-dive
Pass
Audited by Gen Agent Trust Hub on Oct 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill researches topics by ingesting data from untrusted external sources including Reddit, Twitter, and general web search results.
- Ingestion points: External data is fetched via
sandbase_call_toolusing sources likeexa_search,reddit_app_dynamic_search, andtwitter_web_search_timeline(SKILL.md). - Boundary markers: The instructions tell the agent to "Synthesize findings into a clear, evidence-backed answer" and "separate factual observations from analysis," but do not provide specific delimiters or instructions to ignore embedded commands within the retrieved text.
- Capability inventory: The skill possesses the capability to call specific search and content retrieval tools via
sandbase_call_tool(SKILL.md). - Sanitization: No explicit sanitization or filtering of external content is mentioned.
- [SAFE]: The skill uses tools and APIs belonging to the author's infrastructure (
sandbase_call_tool,sandbase_describe_tool). No evidence of credential theft, obfuscation, or unauthorized command execution was found. The workflow is restricted to read-only research as per the internal guidelines.
Audit Metadata