Back to skill
Skillv1.0.1
ClawScan security
抖音视频转文字 · ClawHub's context-aware review of the artifact, metadata, and declared behavior.
Scanner verdict
BenignMar 10, 2026, 9:28 AM
- Verdict
- Benign
- Confidence
- medium
- Model
- gpt-5-mini
- Summary
- The skill's code and runtime instructions are internally consistent with its stated purpose (extract audio from Douyin and transcribe via Groq + ffmpeg), but it asks users to paste an API key into chat and will run local commands and write files — pay attention to secret handling and local execution.
- Guidance
- This skill appears to do what it claims: open a browser (via the agent) to capture the Douyin audio URL, use ffmpeg to download/convert audio, then call Groq to transcribe and post-process. Before installing or using it, consider the following: - Do not paste your Groq API key into a public chat if you are uncomfortable: the SKILL.md instructs users to send their GROQ_API_KEY in chat so the agent can write it into .env. That will expose the key in conversation logs. Prefer to manually edit the .env file locally (or use a platform secret store) instead of sending the key in chat. - The skill runs local binaries (ffmpeg, ffprobe, optionally yt-dlp) and will execute them via child_process; run it in an environment you trust. It will create files under the skill folder (douyin-transcripts/ and temp/). - The agent will open a browser profile (OpenClaw browser) and run a small page-eval script to extract media resource URLs. That is necessary for the intended flow, but be aware the browser action may access pages you open to extract resource URLs. - If you are concerned about secrets or provenance, review scripts/transcribe.js yourself (it is provided) and prefer to configure GROQ_API_KEY directly in .env rather than sending it to the agent. Rotate the API key after testing if you accidentally shared it. Overall the skill is internally coherent for its purpose; the main practical risk is secret handling (user-provided API key in chat) and execution of local commands — not hidden network endpoints or unexpected credential requests.
Review Dimensions
- Purpose & Capability
- okName/description (Douyin → transcription) match what the code and SKILL.md do: open a browser to extract an audio stream, use ffmpeg to download/convert audio, and call Groq (or optionally OpenAI) for STT and punctuation. Declared requirements (ffmpeg, Groq API) are appropriate.
- Instruction Scope
- noteInstructions are explicit about actions the agent will take (read/copy .env, start browser, run ffmpeg, run scripts/transcribe.js, read/write skill-local temp and output directories). A notable UX/security detail: the SKILL.md asks the user to paste their Groq API key into chat so the agent can update .env — this exposes a secret in conversation logs and is a potential privacy/security concern (expected for configuration but worth caution).
- Install Mechanism
- okNo install spec — instruction-only plus an included Node script. Nothing is downloaded from arbitrary URLs during install. The runtime executes local binaries (ffmpeg, optionally yt-dlp) which is expected for this task.
- Credentials
- noteThe only required credential is a Groq API key (GROQ_API_KEY), which is proportionate to the described use. The skill also accepts optional OPENAI_API_KEY and will look for local cookies (douyin-cookies.txt) if falling back to yt-dlp; these are plausible but the SKILL.md's recommendation that users paste API keys into chat is a risk. No unrelated cloud credentials or broad environment access are requested.
- Persistence & Privilege
- okalways is false; the skill requires browser/exec/read tools (expected). It writes output and temp files under its own directories and does not request modifying other skills or system-wide configs.