weekly-research-summary
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, specifically clipboard pastes from other LLM sessions ("GPT paste") and external research links, to generate summaries in Notion. This creates a potential surface where malicious instructions hidden in the pasted content could attempt to influence the agent's behavior.
- Ingestion points: Processes user-provided "GPT paste," multi-session dumps, and arXiv links (referenced in SKILL.md under "Writing Flow" and "GPT Paste / Normalization Rules").
- Boundary markers: The instructions do not specify any delimiters or warnings for the agent to ignore instructions embedded within the ingested data.
- Capability inventory: The skill utilizes the
notion-doc-workflowtool to perform durable writes and updates to the user's Notion workspace. - Sanitization: While the skill performs formatting cleanup (removing trackers, fixing TeX), it lacks security-focused sanitization to filter out prompt injection attacks from the input data.
- [SAFE]: The skill references an official Notion hub URL for document storage. Notion is a well-known service, and this reference is consistent with the skill's primary function of research document management.
Audit Metadata