story-review
Pass
Audited by Gen Agent Trust Hub on Oct 10, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes local Python and Node.js scripts (scripts/normalize-punctuation.js, scripts/check-ai-patterns.js, scripts/check-degeneration.js, scripts/author_memory_commit.py, scripts/tracking_commit.py) to perform linguistic analysis and manage state files. It also uses git diff to identify modification ranges in the workspace.
- [DYNAMIC_EXECUTION]: In scripts/tracking_commit.py, the skill dynamically loads its companion module wordcount_core.py using the importlib library. This is a modular programming technique used to share logic between its internal tools and targets a static relative path within the skill's own directory.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted novel text as its primary input, creating a surface for indirect prompt injection.
- Ingestion points: Reads story content from user-specified file paths for review and state updates.
- Boundary markers: Present; the instructions in style-resolution.md and references/quality-rubric.md define strict interpretation rules and output schemas.
- Capability inventory: File system access (reading/writing state files in 追踪/ and .story/), execution of local scripts, and spawning sub-agents via platform-provided tools.
- Sanitization: The skill uses regex-based linguistic patterns to flag issues rather than following instructions contained within the analyzed text.
Audit Metadata