The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple shell commands to manage video production, including ffmpeg for video encoding/probing and node for running local rendering scripts.
[EXTERNAL_DOWNLOADS]: Utilizes npx playwright install chromium to ensure the required browser environment is available for recording the ad mockup.
[REMOTE_CODE_EXECUTION]: Executes local Node.js scripts (record-master.js, render-end-card.js) while dynamically setting NODE_PATH to include required atom modules for the rendering process.
[DATA_EXFILTRATION]: Performs legitimate data transfers by uploading rendered MP4 files and thumbnails to the vendor's backend using get_upload_url as part of the ad publishing workflow.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from brand research files (*.md) and assets to generate conversation threads. While this presents an ingestion surface for untrusted data, the skill operates within a controlled project environment and utilizes structured JSON for processing conversation beats.
Ingestion points: Reads brand research from agent-config/brands/<slug>/brand-research/*.md and project scripts from thread.json.
Boundary markers: Not explicitly defined in the prompt interpolation; relies on the agent's logic to map source beats to brand voice.
Capability inventory: Full shell access for ffmpeg, npx, and node execution as described in Phase 2.
Sanitization: Relies on internal agent formatting for the thread.json structure.

remix-imessage-ad-from-sample