iterative-academic-writing
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DATA_EXPOSURE]: The skill instructions direct the agent to access local project files, such as
config.yamlandresults.json, to extract parameters and metrics for the writing process. While this access is inherent to its function as a research paper or report writing assistant, it involves processing local environment data. - [INDIRECT_PROMPT_INJECTION]: The skill ingests data from external web searches to verify citations and claims, creating a potential surface for indirect prompt injection.
- Ingestion points: WebSearch results and local project files (Phase 0: Build Fact Base).
- Boundary markers: The process uses a strict 'Fact Base' format to isolate and track verified versus unverified claims.
- Capability inventory: The skill performs natural language drafting, iterative evaluation, and uses standard web search/file reading tools provided by the agent environment.
- Sanitization: The 'Phase 2: Evaluate' step acts as a sanitization layer, requiring every claim to be verified against the Fact Base and flagging 'hallucinations' as critical issues that block completion.
Audit Metadata