The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local shell and Python scripts to perform its tasks. Specifically, 'scripts/start-review-chrome.sh' launches Google Chrome with remote debugging flags and manages temporary profiles, while 'scripts/build-review-page.py' uses subprocess.run to call the macOS 'sips' utility for image resizing. \n- [CREDENTIALS_UNSAFE]: Instructions in 'SKILL.md' and 'templates/reviewer-prompt.md' direct the agent to retrieve passwords from the macOS Keychain using the 'security find-generic-password' command. This grants the agent programmatic access to system-stored credentials. \n- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process content from external web applications via browser automation. \n1. Ingestion points: Web application UI elements (text, labels, badges) and browser console messages are read from the target 'base-url' through the 'chrome-devtools' MCP. \n2. Boundary markers: The 'reviewer-prompt.md' template defines a structured report format but lacks specific delimiters or instructions to the sub-agent to ignore potential malicious content embedded within the target application's UI. \n3. Capability inventory: The skill has the capability to execute shell commands, write files to the local filesystem, and publish HTML Artifacts via the platform's tools. \n4. Sanitization: The 'scripts/build-review-page.py' script applies basic HTML entity escaping to text fields before including them in the final report, providing a layer of protection against XSS in the generated artifact.

browser-review-report