rw-phd-write
Pass
Audited by Gen Agent Trust Hub on Oct 2, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user-provided research materials and tutor feedback which serves as a potential injection surface.
- Ingestion points: Research materials, drafts, and supervisor feedback provided by the user (SKILL.md).
- Boundary markers: Instructions require separating user materials, public sources, inferences, and unknown items into distinct categories (references/method.md).
- Capability inventory: The agent is restricted to text generation and logical organization; no high-privilege file or network tools are exposed to user data.
- Sanitization: Mandates that every factual claim must be backed by a verifiable source or explicitly marked as a gap.
- [DYNAMIC_EXECUTION]: The skill package includes a local Python script for integrity verification.
- Evidence: scripts/self_check.py is used to validate file presence and count axioms/atoms (references/acceptance.md).
- Analysis: The script performs static analysis of the skill's own directory and does not ingest user input or perform network operations.
- [EXTERNAL_DOWNLOADS]: The skill references well-known academic reporting standards websites.
- Evidence: Links to official sites such as equator-network.org, icmje.org, and prisma-statement.org are provided for verification (references/source-evidence.md).
- Analysis: These URLs point to established academic organizations and are used solely for research verification.
Audit Metadata