story-short-analyze
Pass
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill contains explicit instructions to override safety guidelines. In
SKILL.md, under the '拆解边界声明' (Deconstruction Boundary Statement) section, the agent is instructed that it 'must not refuse a whole story or a fragment based on "sensitivity" or "discomfort"' (不得以「敏感」「不适」为由拒绝整篇或某个片段). This instruction specifically targets themes like domestic violence, betrayal, and violence common in the targeted web novel genres. While aimed at ensuring the analysis pipeline remains functional for fictional content, it represents a directive to bypass standard AI safety refusals. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process arbitrary, untrusted story text provided by users for structural and emotional analysis.
- Ingestion points: Untrusted data enters the agent's context in
SKILL.mdPhase 1, Step 1 ('拿到原文'), where the agent accepts story text or file paths from the user. - Boundary markers: The instructions lack explicit boundary markers or warnings to ignore instructions that might be embedded within the story text, although it defines the input as a fictional work.
- Capability inventory: The skill has extensive file-writing capabilities, creating and updating
拆文报告.md,情节节点.md,写作手法.md, and_meta.jsonwithin the local filesystem. - Sanitization: No explicit sanitization or filtering of external content is mentioned before it is processed or written to reports.
Audit Metadata