Skillv1.0.0

ClawScan security

Ai Meeting Helper · ClawHub's context-aware review of the artifact, metadata, and declared behavior.

Scanner verdict

ReviewMar 15, 2026, 7:07 PM

Verdict: Review
Confidence: medium
Model: gpt-5-mini
Summary: The skill's code matches its stated purpose (audio→transcript→LLM summary) and only needs an OpenAI API key, but there are metadata inconsistencies and some install choices that warrant caution.
Guidance: What to consider before installing: - Metadata mismatch: the skill.json and SKILL.md require OPENAI_API_KEY, but the top-level registry metadata claimed no required env vars. Treat that inconsistency as a red flag — confirm which metadata is authoritative before trusting automated installers. - Review code yourself (you have the files): the Python script uses only the OpenAI client and local file I/O; no hard-coded external endpoints or obfuscated code were found. If you don't review, install only in an isolated environment. - Install in a virtual environment or sandbox rather than system-wide (the install.sh runs pip3 install without a venv). This avoids contaminating global Python packages. - Backups: the tool will copy original audio into .ai_meeting_backup/. Those files can contain sensitive conversations — ensure you control that directory, or disable backup if not needed, and securely delete backups when appropriate. - API key scope and monitoring: provide an OpenAI key with appropriate usage controls; monitor usage and billing. If possible use an account with limited permissions or monitoring alerts. - Trust and provenance: owner is unknown and there's no homepage. If you need high assurance, request more provenance or prefer a skill from a known source. If you decide to proceed: run install.sh only after inspecting it, execute the tool inside a dedicated virtualenv or container, and consider auditing network traffic or logs during first runs. If you need help verifying specific lines of code or testing in a sandbox, provide the environment and I can help step through it.

Review Dimensions

Purpose & Capability: noteThe code, SKILL.md, and skill.json all implement a meeting-transcription-and-summary tool that uses OpenAI Whisper and an LLM — requesting an OPENAI_API_KEY is coherent with that purpose. However, the top-level registry metadata provided to you (which claimed no required env vars / no primary credential) contradicts the included skill.json and SKILL.md that require OPENAI_API_KEY. The mismatch is unexpected and should be resolved before trusting the package metadata.
Instruction Scope: okRuntime instructions and the Python code stay within the declared purpose: they read local audio files, call OpenAI APIs for transcription and summarization, format results, optionally backup original audio to a local .ai_meeting_backup/ directory, and write outputs. The instructions do not reference unrelated system paths, other credentials, or external endpoints beyond the OpenAI client.
Install Mechanism: noteInstallation is a local install.sh that runs pip3 install openai python-dotenv (no virtualenv) and creates a .env template and backup/log directories in the skill folder. This uses a public PyPI package (openai) — not high risk by itself — but the script installs packages system-wide (no venv) which may be undesirable. No downloads from arbitrary URLs or obfuscated installers were found.
Credentials: noteThe code requires a single credential (OPENAI_API_KEY), which is proportionate to the stated functionality. However, the provided registry metadata (at the top of this report) claims no required env vars/primary credential while skill.json and SKILL.md declare OPENAI_API_KEY required — that inconsistency is suspicious and could indicate sloppy packaging or metadata manipulation.
Persistence & Privilege: okThe skill does not request 'always: true' or elevated agent privileges. It creates and writes files only within its own skill directory (.env, .ai_meeting_backup/, .ai_meeting_logs/) and does not modify other skills or global agent config. Uninstall.sh offers to remove these files.