Semantic Consistency Auditor

Use semantic consistency auditor for academic writing workflows that need structured execution, explicit assumptions, and clear output boundaries.

This skill appears to implement a semantic-evaluation tool, but there are two important cautions: (1) domain mismatch — SKILL.md sometimes frames the tool as for academic writing while the code and examples reference clinical notes; clarify intended domain and thresholds before relying on results. (2) privacy and network activity — the script will attempt to import/initialize BERTScore and COMET and will download model artifacts at runtime (network access). Don’t run it on sensitive patient data (PHI) without ensuring compliance and isolation. Recommended next steps before installing or running: inspect the full scripts/main.py (the provided snippet was truncated here), run python -m py_compile scripts/main.py in an isolated virtualenv, pin dependency versions, restrict outbound network access if you want to avoid automatic model downloads, and test with non-sensitive sample data. If you need higher assurance, ask the author for a clear statement of intended domain (academic vs clinical), the full source, and an audit of network endpoints used for model downloads.

Type: OpenClaw Skill Name: semantic-consistency-auditor Version: 1.0.0 The skill is a legitimate utility designed to evaluate semantic consistency between AI-generated text and gold standards using BERTScore and COMET metrics. The implementation in `scripts/main.py` is well-structured, uses `yaml.safe_load` for security, and contains no evidence of data exfiltration, unauthorized execution, or malicious prompt injection. A minor discrepancy exists where `requirements.txt` lists 'comet' (the experiment tracker) instead of 'unbabel-comet' (the evaluation library), but this appears to be a functional bug rather than a malicious dependency confusion attempt.