The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process and manage user-generated creative content such as scripts and visual settings, which are ingested into the agent's context.
Ingestion points: Creative documents (e.g., 剧本.md, 视觉设定.md) located in the project workspace are read and parsed by scripts/creator_views.py and scripts/dashboard_server.py.
Boundary markers: Document structures are defined in references/creator-documents.md, though explicit warnings to ignore instructions inside screenplay text are not present.
Capability inventory: The skill possesses capabilities for recursive file-system operations within the workspace (scripts/project_tool.py) and hosts a local HTTP server (scripts/dashboard_server.py).
Sanitization: The dashboard interface (assets/dashboard/app.js) implements HTML escaping for text values before they are rendered in the structured views.
[COMMAND_EXECUTION]: The skill spawns subprocesses to facilitate background operation of its dashboard.
Evidence: In scripts/dashboard_server.py, the _detached_child function uses subprocess.Popen to re-launch the script using sys.executable. The execution is restricted to the script's own path with hardcoded arguments, mitigating the risk of arbitrary command injection.

short-drama