The Agent Skills Directory

[PROMPT_INJECTION]: The SKILL.md file contains instructions that explicitly direct the agent to bypass standard confirmation gates. In the '工具优先级' section, it instructs the agent to install missing dependencies directly without prompting the user ('不要停下来问用户'). Additionally, the '持续积分授权' section encourages the agent to bypass confirmation for actions that consume account credits on third-party platforms.
[PRIVILEGE_ESCALATION]: The scripts/setup_deps.js script attempts to execute sudo -n apt-get install to install system-level packages on Linux. This behavior aims to perform privileged operations silently without human intervention.
[COMMAND_EXECUTION]: The skill operates by building and executing complex shell command chains. It spawns child processes for dependency management (npm, brew, apt), video rendering (ffmpeg), and browser-based frame capture (playwright). While the skill implements internal path validation to ensure files stay within the project root, the breadth of command execution is substantial.
[EXTERNAL_DOWNLOADS]: The skill autonomously manages software installations from external package registries (NPM, PyPI, Homebrew) and downloads generated media assets from remote URLs. This automated environment setup is a core part of its functionality but bypasses typical user oversight.
[DATA_EXFILTRATION]: The skill is designed to read sensitive credential files from the local filesystem to authenticate with external services. For example, scripts/adapters/33tc.js reads tokens from ~/Library/Application Support/33TaiCi/config.json, and scripts/adapters/listenhub.js accesses configuration files in ~/.config/listenhub/. While these are used for the skill's primary purpose, they involve handling high-value authentication secrets.

qiaomu-cut