The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill utilizes npx hyperframes transcribe, which downloads and runs the hyperframes package from the public NPM registry at runtime.
[REMOTE_CODE_EXECUTION]: By using npx, the skill executes remote code fetched from an external package registry, which could be compromised or contain malicious logic.
[COMMAND_EXECUTION]: The skill relies heavily on shell commands for forensic analysis, including ffmpeg for frame extraction and cut detection, ffprobe for metadata probing, and whisper-cli for audio transcription.
[CREDENTIALS_UNSAFE]: The instructions in references/method.md suggest asking the user to "provide their logged-in browser session" to bypass platform download restrictions. This could lead to the exposure or exfiltration of sensitive session cookies and authentication tokens.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests and analyzes untrusted user-provided video files and scripts.
Ingestion points: Reference video file (ref.mp4) and user-provided script.
Boundary markers: None identified to delimit processed data from agent instructions.
Capability inventory: Shell command execution (ffmpeg, ffprobe, whisper-cli), remote package execution (npx), and subagent tasking.
Sanitization: No sanitization or validation of the content extracted from frames or transcripts is described.
[DYNAMIC_EXECUTION]: The skill dynamically loads and executes the hyperframes tool via npx to perform transcription, representing a runtime code loading pattern.

reel-style-clone