story-long-analyze
Pass
Audited by Gen Agent Trust Hub on Aug 20, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes shell commands such as
grep,sed,ls, andcatinSKILL.mdandreferences/style-profile-generator.mdto identify chapter boundaries, merge summary files, and manage progress tracking. - [COMMAND_EXECUTION]: In
references/style-profile-generator.md, the skill dynamically executes a Python script via a shell heredoc to calculate sentence length and punctuation density from sampled text for stylistic profiling. - [PROMPT_INJECTION]: Instructions in
SKILL.mddirect the agent to disregard potential safety refusals related to sensitive fictional content (e.g., violence, mature themes) to prevent interruptions in the automated analysis pipeline, framing this as a boundary for literary criticism. - [PROMPT_INJECTION]: The skill processes user-provided novel text and passes it to sub-agents (
chapter-extractor). This presents a surface for indirect prompt injection, which the skill attempts to mitigate by employing a[材料声明](Material Declaration) prefix to provide context and reduce the risk of the agent following instructions embedded in the analyzed fiction.
Audit Metadata