The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill relies on ai_workspace/project-snapshot.json as the "single source of truth" for the timeline state, creating a surface for indirect prompt injection if the project data is sourced from untrusted external inputs.
Ingestion points: ai_workspace/project-snapshot.json (referenced in SKILL.md).
Boundary markers: None mentioned in the documentation for separating user-provided data from instructions.
Capability inventory: The skill enables file system modifications (archiving, switching, and deleting project folders) and shell command execution via local Python scripts.
Sanitization: No sanitization or schema validation processes are described for the snapshot data beyond basic type checks.
[DATA_EXFILTRATION]: The skill documents a local API endpoint that serves files from the local filesystem to the browser using absolute paths.
Evidence: "Always use absolute paths wrapped in the /api/media/serve?path= API. This ensures files from any drive (F:, D:, etc.) work correctly."
Risk: This utility provides a mechanism to read and serve any file on the host system if the agent is directed to an absolute path of a sensitive file (e.g., credentials or configuration files).
[COMMAND_EXECUTION]: The skill facilitates the execution of various Python scripts located in the scripts/ directory to perform video editing tasks.
Evidence: Provides usage examples for scripts/aicut_tool.py, scripts/fix_subtitle_pos.py, and scripts/project_manager.py.
Context: These scripts are part of the skill's internal tooling for media import and project management.

aicut-editing