story-long-write
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes several internal utility scripts (e.g.,
scripts/storyctl.py,scripts/check-ai-patterns.js,scripts/check-degeneration.js) to perform automated linting, word count validation, and structural verification. These scripts are executed locally using the system'snodeandpythonbinaries. The usage is strictly scoped to the skill's primary function of maintaining novel quality and project structure. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-generated story content and external novel data for 'benchmarking.' This creates a typical surface for indirect prompt injection attacks.
- Ingestion points: Untrusted novel data and outlines are read from the file system (
正文/,大纲/,对标/). - Boundary markers: Present. The instructions include explicit directives to use boundary markers and to 'ignore embedded instructions' when processing external data.
- Capability inventory: The skill possesses file-writing capabilities (creating story artifacts) and local execution via subprocesses for its internal tools.
- Sanitization: The skill uses structural validation scripts (
check-outline-contract.js) and prose checkers (check-ai-patterns.js) to ensure data remains within the expected novel format.
Audit Metadata