research-prompt-guide
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill provides structured templates designed to ingest untrusted external data (URLs, text, and documents) for fact-checking and research purposes.
- Ingestion points: Templates in
references/full.md(e.g.,FactCheckPrompt,ResearchPrompt) define specific fields like<Source>,<Link>, and<RawText>for external content. - Boundary markers: The use of XML-style tagging provides a structural boundary between the agent's instructions and the untrusted data, although this does not inherently prevent sophisticated injection attacks.
- Capability inventory: The skill's workflows (e.g.,
EvidenceGathering,DataCollection) require the agent to execute web searches and cross-reference information, which could be influenced by malicious instructions embedded in the source data. - Sanitization: While the skill emphasizes IFCN principles for source reliability, it does not include explicit instructions or methods to sanitize or detect prompt injection attempts within the processed data.
Audit Metadata