The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes transcript data from external assets to perform synchronization, identification, and editing tasks. This creates a vulnerability surface where malicious instructions could be embedded in the transcript files (e.g., subtitles or ASR output) to influence the agent's behavior during the analysis phase.
Ingestion points: The skill reads transcripts for every asset (Part 1) and accepts an utterances.json file as input for the scripts/transcript-offset.mjs script (Part 2).
Boundary markers: None detected. The instructions do not define delimiters or provide specific commands for the agent to ignore instructions embedded within the processed data.
Capability inventory: The agent has the capability to execute a local Node.js script, invoke the multicam_sync tool, modify timelines, and rename tracks and files.
Sanitization: The scripts/transcript-offset.mjs script performs text normalization (removing non-alphanumeric characters) for matching purposes, but this does not sanitize the data against logical or instructional prompt injections.
[COMMAND_EXECUTION]: The skill instructions explicitly direct the agent to execute a local script file (scripts/transcript-offset.mjs) using the Node.js runtime to calculate synchronization offsets. While the script is part of the skill's distribution, this involves spawning a subprocess with an external file path passed as an argument.

multicam-sync