The Agent Skills Directory

[COMMAND_EXECUTION]: The skill frequently executes command-line tools to process media. Specifically, scripts/analyze_replication.py utilizes subprocess.check_output to invoke ffmpeg for audio analysis and frame extraction. Additionally, mcp/server.mjs uses execFile to run the skill's own CLI commands.
[PRIVILEGE_ESCALATION]: The environment setup script scripts/bootstrap-host.sh uses sudo apt-get to install required system packages on Linux distributions. These operations are explicitly declared to the user and require interactive authorization before proceeding.
[EXTERNAL_DOWNLOADS]: The skill fetches voice synthesis models from external repositories, including Hugging Face and ModelScope. These downloads are performed to support local AI narration features and are routed to established AI community hubs.
[INDIRECT_PROMPT_INJECTION]: The skill has a large attack surface for indirect prompt injection as it ingests and processes external video files, project metadata (project.json), and transcript data.
Ingestion points: Reads .timeline archives (ZIP files) containing project.json and external media assets.
Boundary markers: Instructions mandate the use of project.diff for a semantic dry run and a structural validator (validate_edit_plan.mjs) to verify plans before execution.
Capability inventory: Includes file system writes, CLI tool execution, and network API calls to video generation services.
Sanitization: Implements strict schema validation using the zod library for the MCP server interface and JSON-based command contracts.
[DYNAMIC_EXECUTION]: The skill manages a local runtime environment where it dynamically executes Python and Node.js scripts from its own scripts/ directory to perform specialized tasks like media replication analysis and environment verification.

edit-timeline-studio