The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts scripts/render_captions.py and scripts/validate_caption_render.py invoke ffmpeg and ffprobe via subprocess.run to perform media processing tasks such as metadata extraction, image sequence generation, and video encoding. These operations are functional components of the skill's purpose.
[COMMAND_EXECUTION]: The scripts/render_captions.py script contains a self-re-execution block that uses os.execv to switch to a specific bundled Python interpreter located at ~/.cache/codex-runtimes/codex-primary-runtime/dependencies/python/bin/python3 if the required Pillow library is missing. This process replacement targets a fixed local path for runtime compatibility.
[PROMPT_INJECTION]: The skill processes untrusted text from external caption files to render video overlays, creating an Indirect Prompt Injection surface (Category 8).
Ingestion points: Reads text from captions.json or .srt files within scripts/render_captions.py.
Boundary markers: Absent; the text is interpolated directly into the rendering logic without delimiters or warnings.
Capability inventory: The skill executes shell commands (ffmpeg/ffprobe) and replaces the current process via os.execv.
Sanitization: Absent; the text is measured and wrapped for visual layout, but no content validation or instruction filtering is performed.

skill-captions