The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill generates animation logic in anim.js at runtime based on the interpretation of a math problem. This script is subsequently executed within a Playwright-controlled browser environment during the rendering process. Furthermore, build_audio.py performs dynamic loading by importing the pron.py module from a computed path relative to the project workspace.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill processes user-supplied math problem text and screenshots, which are added to the agent's context during the initial problem-solving phase in SKILL.md (Step 1).
Boundary markers: The skill lacks explicit delimiters or specific instructions for the agent to ignore potentially malicious embedded commands within the ingested math problem data.
Capability inventory: The skill possesses capabilities including system command execution via subprocess.run and child_process.spawn for media processing, along with extensive file system read/write access.
Sanitization: No dedicated sanitization or validation mechanisms were found to filter input math problem content before it influences the animation generation logic.
[COMMAND_EXECUTION]: The skill utilizes subprocess.run in Python and child_process.spawn in Node.js to invoke system utilities, primarily ffmpeg for audio and video multiplexing and the macOS say command as a fallback for voice synthesis. These operations are conducted using argument lists, which mitigates simple shell injection but confirms the skill's ability to execute external binaries.
[EXTERNAL_DOWNLOADS]: The skill interacts with the Zhipu AI official API (https://open.bigmodel.cn) for voice synthesis and speech recognition services. This is an established AI service, and the network communication is necessary for the skill's primary function of generating high-quality audio narration.
[CREDENTIALS_UNSAFE]: To access the Zhipu AI services, the skill requires a GLM_API_KEY. It follows secure practices by instructing the user to store this sensitive credential in a local .env file (~/.config/math-problem-video/.env) with restricted file permissions (chmod 600), rather than hardcoding it or exposing it in logs.

edu-math-video