Skillv0.1.0

ClawScan security

Voice To Protocol Transcriber · ClawHub's context-aware review of the artifact, metadata, and declared behavior.

Scanner verdict

ReviewMar 13, 2026, 9:46 AM

Verdict: Review
Confidence: medium
Model: gpt-5-mini
Summary: The skill's purpose and code mostly line up with a local voice transcription tool, but there are notable inconsistencies (dependency/install instructions vs. requirements.txt, and ambiguous network usage for speech recognition) that warrant caution before installing in a lab environment.
Guidance: What to check before installing: - Dependency mismatch: SKILL.md asks you to install speechrecognition, pyaudio, pydub, python-docx but requirements.txt does not include them. Expect to manually install the missing packages (use a virtualenv). Verify the skill runs after installing exactly what it needs. - Network/privacy risk: The speechrecognition library typically uses online services by default (for example, Google Web Speech API). Confirm whether the code uses an offline recognizer or sends audio to external servers. If the code calls cloud recognizers, audio (potentially sensitive experimental data) may leave your machine. Ask the author or inspect the full main.py to see how Recognition is configured. - Inspect the rest of main.py: only a truncated portion was provided. Review the remainder for any hard-coded endpoints, hidden subprocess/network calls, or unexpected file access before trusting this in regulated labs. - Install safely: install inside a virtual environment on a non-production/test machine first. Run with microphone permissions in an isolated environment. Check that saved files go to a directory you control and do not inadvertently overwrite other files. - Operational caution: don't run this against sensitive or PHI-containing experiments until you confirm where audio/text is processed (locally vs. remote) and audit third-party dependencies. If you can provide the full main.py (untruncated) or confirm whether speech recognition is configured for an offline engine, I can raise or lower the confidence and give more specific advice.

Review Dimensions

Purpose & Capability: noteName/description, SKILL.md and the included Python code all align with a voice-driven experiment recorder that saves files to the user's Documents directory. However, SKILL.md lists external Python packages (speechrecognition, pyaudio, pydub, python-docx) while the included requirements.txt only contains 'dataclasses', 'enum', and 'wave' — a clear mismatch. Also the bundle contains code but no install spec, which is inconsistent with the advertised dependency installation steps.
Instruction Scope: concernSKILL.md says 'No external API calls' in the risk table, yet it recommends the 'speechrecognition' package and notes Chinese recognition needs a good network connection. The speechrecognition library commonly uses cloud/online recognizers (e.g., Google) unless explicitly configured for an offline engine; this can send audio to third-party servers. The instructions otherwise operate on local files and a local config path (~/.openclaw/...), which is expected for the stated purpose.
Install Mechanism: concernThere is no install spec in the registry (instruction-only), yet the package includes executable Python code and a requirements.txt that does not reflect the SKILL.md 'pip install' list. This mismatch may lead to missing runtime dependencies or hidden manual install steps. No remote download URLs are present in files examined, which reduces high-risk install behavior, but the missing/incorrect dependency declarations are a practical risk.
Credentials: okThe skill requests no environment variables or credentials and only reads/writes local files (default save dir ~/Documents/Experiment-Protocols or ~/.openclaw config). That is proportionate for a local transcription tool. Users should note audio data and saved protocol files may contain sensitive lab data.
Persistence & Privilege: okNo elevated privileges requested, always=false, and the skill does not declare modifications to other skills or global agent settings. It writes its own config under ~/.openclaw and saves files to a user-owned Documents folder — expected for this functionality.