The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes execFileSync in scripts/link-transcript.js to run local installations of ffmpeg and ffprobe. These tools are used to extract audio from video files and verify media durations before uploading them for transcription. This behavior is directly aligned with the skill's stated purpose of processing local media files.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by processing data from external sources.
Ingestion points: Data enters the agent's context through scripts/sync.js (fetching notes from biji.com) and scripts/link-transcript.js (fetching transcription results from third-party media platforms and the Get Notes API).
Boundary markers: Notes are delimited using standard Markdown and YAML frontmatter blocks in scripts/format.js. There are no specific instructions added to the content to ignore embedded prompts.
Capability inventory: The skill has the capability to write to the local file system (fs.writeFileSync in scripts/sync.js) and execute subprocesses (execFileSync in scripts/link-transcript.js).
Sanitization: The safeName function in scripts/format.js sanitizes titles for use in filenames by stripping illegal characters and limiting length, but the content of the notes is interpolated into Markdown files without additional escaping or filtering.

get-biji