crossframe-critical
Pass
Audited by Gen Agent Trust Hub on Oct 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted user-supplied data (facts, events, and materials) to generate critique dossiers and essays. This creates an attack surface for instructions embedded in input data.
- Ingestion points: User input prompts and provided materials are processed in
SKILL.mdandprotocols/critical-article-protocol.md. - Boundary markers: The skill implements conceptual boundary markers by requiring a 'Fact Boundary' (事实边界) section that explicitly distinguishes between user-provided facts, interpretations, and hypotheses.
- Capability inventory: The skill is limited to text generation within the agent's context. No subprocess execution, network operations, or filesystem write capabilities were found in the analyzed files.
- Sanitization: The skill employs a strict 'source-ledger-workflow' (来源台账) and evidence rules in
references/example-and-evidence-rules.mdto validate and label external content.
Audit Metadata