research-doc-workflow
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external research documents, paper cards, and PDF translations (referencing tools like
pdf2zh-next), which constitute untrusted data sources. - Ingestion points: Content is pulled from Feishu/Lark URLs, Notion pages, local Obsidian vaults, and external PDF translation outputs.
- Boundary markers: The instructions focus on formatting and hierarchy but do not define delimiters or specific warnings to ignore instructions that might be embedded within the research text.
- Capability inventory: The skill orchestrates data writes, page migrations, and complex modifications across multiple third-party cloud platforms using referenced workflow adapters.
- Sanitization: While the skill performs 'platform residue' stripping (e.g., removing platform-specific HTML or artifacts), it does not explicitly sanitize the semantic content against adversarial instructions designed to influence the agent.
Audit Metadata