The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple shell scripts (check-cdp.sh, check-agent-browser.sh) to prepare the environment and uses the agent-browser CLI to interact with a web browser.
Evidence: SKILL.md triggers sh scripts/check-cdp.sh and sh scripts/check-agent-browser.sh. It also executes agent-browser --cdp 9222 eval "$(cat scripts/order-analysis.js)".
[EXTERNAL_DOWNLOADS]: The installation script automatically downloads and installs Node.js via Homebrew and the agent-browser package from a third-party NPM mirror if they are not present.
Evidence: scripts/check-agent-browser.sh executes brew install node@22 and npm install -g agent-browser --registry=https://registry.npmmirror.com.
[DATA_EXFILTRATION]: The skill is designed to extract potentially sensitive internal ticket data from an authenticated session into a local file for subsequent AI analysis.
Evidence: SKILL.md captures JSON data from a fetch request in scripts/order-analysis.js and saves it to $OUTPUT_DIR/order.json.
[REMOTE_CODE_EXECUTION]: The skill uses agent-browser --cdp 9222 eval to execute the content of a local JavaScript file within the browser context. While the file is local, this mechanism could be used to execute arbitrary logic in the browser.
Evidence: SKILL.md contains the command agent-browser --cdp 9222 eval "$(cat scripts/order-analysis.js)".

order-analysis