The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an inherent attack surface for indirect prompt injection.
Ingestion points: The scripts/read-source.mjs script ingests untrusted content from external URLs (X/Twitter, YouTube, and generic web pages), which is then used to generate scripts in Source.md.
Boundary markers: The instructions lack explicit boundary markers or delimiters to protect the agent's prompt context from instructions potentially embedded in the scraped source data.
Capability inventory: The skill possesses significant local capabilities, including subprocess execution of ffmpeg, ffprobe, and yt-dlp, as well as browser automation through playwright.
Sanitization: No evidence of sanitization or filtering of the ingested external content was found in the scripts before it is processed by the AI.
[COMMAND_EXECUTION]: Multiple scripts (scripts/decursor.py, scripts/join-takes.py, scripts/tighten-voice.py, scripts/verify-export.mjs) use Python's subprocess module to execute ffmpeg and ffprobe binaries.
Evidence: These calls are used for specialized video processing tasks such as inpainting cursors, audio normalization, and merging video segments, which align with the skill's primary purpose.
[EXTERNAL_DOWNLOADS]: The skill connects to several external services to fetch data and media.
Evidence: scripts/vo-words.py interacts with the ElevenLabs API (api.elevenlabs.io) for voice generation. scripts/read-source.mjs uses the fxtwitter API to scrape social media content. scripts/preview-reel.mjs downloads audio assets from fil.design. These are legitimate service integrations for the tool's use case.

html-reel