The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses Python's subprocess module for orchestration. scripts/doctor.py checks the installation and login status of the required codex CLI tool. scripts/render_codex.py executes sibling scripts (check_prompts.py, build_prompts.py, make_page.py) and calls the codex exec command to perform image generation. These calls are documented as the core mechanism for processing tasks.
[INDIRECT_PROMPT_INJECTION]: As the skill generates prompts from user-provided text, it contains an indirect prompt injection surface. This is mitigated by structured templates and a validation script (scripts/check_prompts.py) that uses regular expressions to enforce boundary markers (e.g., 【必须逐字放入的文字（唯一真值）】) and verify that only intended text is passed to the generation tool.
[DATA_EXPOSURE]: The skill accesses local configuration files in ~/.scene-ppt for persistence across updates. It also reads session logs from the external codex tool (~/.codex/sessions) to perform an automated verification that the generated image content matches the original prompt verbatim. This data access is localized and serves a functional quality-assurance role.

scene-ppt