The Agent Skills Directory

[SAFE]: The skill provides instructional content for a benign scriptwriting workflow. No malicious behavior, obfuscation, or safety filter bypass attempts were found.
[PROMPT_INJECTION]: The skill processes external data (sidecar files and codebase) which presents a surface for indirect prompt injection. This is evaluated as safe.
Ingestion points: Milestone timestamps in JSON 'sidecar' files and claims verification against the user's codebase.
Boundary markers: Absent; the skill relies on data integrity rather than delimiters.
Capability inventory: File system reads and execution of video metadata tools (ffprobe).
Sanitization: Includes a check for a timestamps_suspect flag in the input data to prevent processing unreliable timestamps.
[COMMAND_EXECUTION]: The skill workflow involves using standard CLI tools like ffprobe and ffmpeg for video synchronization tasks. This usage is transparent and aligns with the skill's stated purpose of creating timed scripts.

pitch-craft