story-long-write
Pass
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill invokes local tools via
subprocess.runin scripts likescripts/storyctl.pyandscripts/build_writer_prompt.pyto perform word counting and execute Node.js-based style checkers (check-ai-patterns.js,check-degeneration.js). These executions are limited to the skill's local directory and standard system environments. - [DYNAMIC_EXECUTION]: Scripts such as
scripts/storyctl.pyandscripts/author_memory_commit.pyutilizeimportlib.utilto dynamically load local helper modules (wordcount_core.py,tracking_commit.py). This is used for modular code reuse within the skill package. - [INDIRECT_PROMPT_INJECTION]: The skill's primary function involves processing user-authored novel content.
- Ingestion points: Novel drafts are read from the
正文/directory and injected into the narrative writer's prompt viascripts/build_writer_prompt.py. - Boundary markers: The skill uses clear section headers (e.g.,
——— 本节速记 ———) to delineate user data within the prompt. - Capability inventory: Subprocess execution of local scripts for linting and file writing for state tracking.
- Sanitization: Standard text normalization and cleaning are performed, though specific adversarial injection filtering of the novel content is minimal.
Audit Metadata