The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill contains a hardcoded API key and Bot ID for the Coze platform.
Evidence in SKILL.md: export COZE_BOT_ID="7572557757883383858" and export COZE_API_KEY="cztei_qHZQ0A5OSJjsmfZWmVb8bqu2BTbtB240YGbDYLhZpsIr8jER4aL4Aevyii8rnKfNs".
Evidence in scripts/coze_bot_client.py: The main() function sets default values for bot_id ("7572557757883383858") and api_key ("cztei_qHZQ0A5OSJjsmfZWmVb8bqu2BTbtB240YGbDYLhZpsIr8jER4aL4Aevyii8rnKfNs") if environment variables are not present.
[INDIRECT_PROMPT_INJECTION]: The skill has a significant attack surface for indirect prompt injection via video content or vision model responses.
Ingestion points: Untrusted data enters the context via video file paths or URLs in scripts/video_frame_extractor.py, and via model responses from external APIs in scripts/visual_analyzer.py and scripts/coze_bot_client.py.
Boundary markers: There are no delimiters or instructions to ignore embedded commands within the processed visual analysis or video data.
Capability inventory: The skill has the ability to write files to the local system (cv2.imwrite in scripts/video_frame_extractor.py) and perform network requests to external APIs (requests.post in scripts/visual_analyzer.py and scripts/coze_bot_client.py).
Sanitization: There is no evidence of sanitization, filtering, or validation performed on the text descriptions returned by the vision models before they are consumed by the agent.

video-frame-extractor