The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The shell script bin/om dynamically executes npx --yes huashu-motion, which downloads and runs external code from the NPM registry whenever the skill's primary commands are invoked. While the package belongs to the skill author, it introduces a reliance on unpinned external code at runtime.
[COMMAND_EXECUTION]: The skill makes extensive use of local CLI tools, including git, node, npx, and ffmpeg. Specifically, it uses git to conduct network-based version checks against the remote repository on GitHub.
[DATA_EXFILTRATION]: The references/experience.md file documents a technique for using yt-dlp with the --cookies-from-browser chrome flag to bypass bot detection on video platforms. This constitutes an exposure of sensitive user session data (browser cookies).
[PRIVILEGE_ESCALATION]: In SKILL.md, the instructions explicitly direct the agent to bypass environment restrictions: "If your sandbox blocks starting a browser or a local server... run them outside it — ask the user to approve them." This is a social engineering technique used to escape the security constraints of the agent's execution sandbox.
[PROMPT_INJECTION]: The Version Check logic in SKILL.md causes the agent to fetch remote configuration (git tags) from a network source at load time. This external data influences the agent's behavior and allows the author to inject instructions into the user's session via update notifications.

huashu-motion