The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill's primary workflow involves cloning a third-party repository (git@github.com:heimanba/oss-video-skill.git) and then explicitly instructing the agent to load and execute the instructions contained in {WORKSPACE_DIR}/oss-video-skill/.qoder/skills/oss-video/SKILL.md. This pattern allows an external, unverified source to provide the logic and commands that the agent will follow, including package installation and rendering operations.
[EXTERNAL_DOWNLOADS]: The skill fetches code and configuration files from an external GitHub account (heimanba) that is not identified as a trusted vendor. This introduces a supply chain risk as the content of the repository can be modified at any time to include malicious instructions.
[COMMAND_EXECUTION]: The skill uses shell commands (cd and git clone) to download external resources and navigate the file system.
[INDIRECT_PROMPT_INJECTION]: The skill accepts an {owner}/{repo} parameter which is used to identify the GitHub repository for video generation. This ingestion point allows untrusted data to enter the agent's context. If the targeted repository contains malicious metadata or data specifically crafted to influence the rendering and data-gathering logic of the triggered skill, it could lead to unintended agent behavior.

oss-styles