The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes a dedicated runtime management script (scripts/codex_ppt_runtime.py) that utilizes subprocess.run to create a Python virtual environment and install required dependencies. This functionality is part of a standard bootstrap process to ensure the skill operates within a controlled and isolated environment.
[EXTERNAL_DOWNLOADS]: During its setup phase, the skill downloads verified libraries from the Python Package Index (PyPI). It also communicates with the OpenAI API, a well-known service, to perform its core image generation functions and to verify API connectivity via the doctor command.
[PROMPT_INJECTION]: The skill ingests user-supplied content to generate presentation materials, creating a surface for indirect prompt injection. This is addressed by a robust architecture that uses structured JSON jobs (prompts/slide_XX.json) and clear prompt templating to isolate user data from the agent's core instructions.
Ingestion points: User source materials such as articles, reports, or outlines are ingested into the project workspace.
Boundary markers: The workflow employs structured templates and multi-phase validation (outline to spec to job) which separates data from instruction.
Capability inventory: The skill's capabilities are limited to project file management, image generation API calls, and the execution of its bundled Python scripts.
Sanitization: Raw user input is processed through a planning phase that converts unstructured text into a technical specification before it is used in generative prompts.

codex-ppt