light-paper-polishing
Pass
Audited by Gen Agent Trust Hub on Jun 23, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [SAFE]: The script 'scripts/polish.py' communicates with 'api.languagetool.org' to perform grammar and style checks. This is a well-known technology service, and its integration is transparently documented in the skill instructions. The script uses this endpoint solely for its intended purpose of text refinement.
- [SAFE]: The 'scripts/mechanical_check.py' and 'scripts/style_fingerprint.py' tools are designed for local execution. They use standard Python libraries to perform statistical analysis of writing style and to scan for academic tone issues (such as overclaims and AI-generated filler phrases) without requiring network access.
- [PROMPT_INJECTION]: The skill processes untrusted user-supplied data in the form of academic papers, which constitutes an indirect prompt injection surface.
- Ingestion points: Text inputs via stdin or local files (e.g., 'paper.txt') processed by the provided Python scripts and the agent.
- Boundary markers: Explicit delimiters for the processed text are absent in the prompt templates.
- Capability inventory: The agent can perform network requests via 'scripts/polish.py' and has general file-system access typical of agent environments.
- Sanitization: No explicit sanitization or filtering of the user-provided paper content is performed before processing. However, the risk is inherent to the skill's primary function and is mitigated by the structured workflow (distill, critique, polish, audit) defined for the agent.
Audit Metadata