audio-to-text-and-video-to-text Security Report — Low Risk | ClawSafe

10 /100

audio-to-text-and-video-to-text

Transcribe audio and video files into text using OpenAI's Whisper API

Standard audio/video transcription skill with legitimate subprocess and API usage; no malicious behavior detected.

Skill Nameaudio-to-text-and-video-to-text

Duration34.4s

Enginepi

✓

Safe to install

Pin pip dependencies to specific versions to mitigate supply chain risks. Otherwise safe to use.

Findings 1 items

Severity	Finding	Location
Low	Dependencies installed without version pinning Supply Chain The pip install commands for openai and pydub do not specify version constraints, potentially allowing installation of tampered or vulnerable versions. `subprocess.check_call([sys.executable, '-m', 'pip', 'install', 'openai', '--break-system-packages', '-q'])` → Pin to known-safe versions: openai>=1.0.0,<2.0.0 pydub>=0.25.0	`scripts/transcribe.py:103`

Resource	Declared	Inferred	Status	Evidence
Filesystem	`WRITE`	`WRITE`	✓ Aligned	scripts/transcribe.py reads media input and writes transcript output
Network	`READ`	`READ`	✓ Aligned	scripts/transcribe.py calls OpenAI Whisper API only
Shell	`WRITE`	`WRITE`	✓ Aligned	subprocess calls for ffmpeg and pip install are declared in SKILL.md
Environment	`READ`	`READ`	✓ Aligned	Reads OPENAI_API_KEY from environment as documented

2 findings

🔗

Medium External URL 外部 URL

https://platform.openai.com/api-keys

SKILL.md:96

🔗

Medium External URL 外部 URL

https://ffmpeg.org/download.html

scripts/transcribe.py:96

4 files · 39.6 KB · 1152 lines

Python 2f · 852L Markdown 2f · 300L

├─ ▾ 📁 scripts

│ └─ 🐍 transcribe.py Python 426L · 14.4 KB

├─ ▾ 📁 transcription

│ ├─ ▾ 📁 scripts

│ │ └─ 🐍 transcribe.py Python 426L · 14.4 KB

│ └─ 📝 SKILL.md Markdown 150L · 5.4 KB

└─ 📝 SKILL.md Markdown 150L · 5.4 KB

Package	Version	Source	Known Vulns	Notes
`openai`	`*`	pip	No	Version not pinned
`pydub`	`*`	pip	No	Version not pinned

✓ All subprocess usage (ffmpeg, pip) is documented in SKILL.md

✓ API key is used only for Whisper API calls; not harvested or exfiltrated

✓ No access to sensitive paths (~/.ssh, ~/.aws, .env)

✓ No obfuscation techniques (base64, eval, encoded commands)

✓ No persistence mechanisms (cron, startup hooks)

✓ No credential theft or data exfiltration

✓ No reverse shell or C2 communication

✓ No hidden instructions or prompt injection