story-long-analyze
Warn
Audited by Gen Agent Trust Hub on Jul 26, 2026
Risk Level: MEDIUMPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The instructions explicitly direct the agent to override safety and sensitivity filters for fictional content, specifically mentioning that elements like violence and domestic abuse must be processed and that the agent must not refuse due to 'sensitivity' or 'discomfort'.
- [COMMAND_EXECUTION]: The skill uses shell commands and pipelines (
ls,sed,sort,cut,cat) to merge and process summary files. It also dynamically executes a Python script via a heredoc in the shell to calculate sentence length and punctuation density from sample text. - [DATA_EXPOSURE]: The skill attempts to access and evaluate agent configuration files stored in hidden project directories including
.claude/agents/,.opencode/agents/, and.codex/agents/to determine subagent compatibility. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted novel text provided by users. While it uses a specific '[材料声明]' (Material Declaration) boundary marker to scope the content, it possesses significant capabilities including shell execution and file writing that could be targeted by adversarial text embedded in the source novels.
Audit Metadata