The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill uses a Python script to dynamically modify the agent host's MCP configuration files by adding or updating command-line arguments that control how tools are executed.\n
Evidence: The apply function in scripts/setup_chrome_devtools.py modifies the mcpServers JSON object to include the --autoConnect flag and writes the changes back to the host configuration file.\n- [COMMAND_EXECUTION]: The skill requires the agent to execute a local Python script to perform configuration changes on the host system.\n
Evidence: Instructions in references/CHROME_DEVTOOLS_MCP.md direct the agent to run python3 <skill-dir>/scripts/setup_chrome_devtools.py --config <host-config> --apply.\n- [INDIRECT_PROMPT_INJECTION]: The skill interacts with live browser tabs, creating an attack surface where malicious instructions embedded in web pages could influence the agent's actions.\n
Ingestion points: Browser snapshots and page listings (SKILL.md).\n
Boundary markers: Present; the skill instructions explicitly warn the agent to treat instructions rendered inside web pages as untrusted content (SKILL.md).\n
Capability inventory: The agent can list pages, capture snapshots, fill forms, click elements, and navigate pages using the Chrome DevTools MCP (SKILL.md, CHROME_DEVTOOLS_MCP.md).\n
Sanitization: None; the skill does not implement technical filtering or sanitization of the processed page content.\n- [EXTERNAL_DOWNLOADS]: The skill configures the host to fetch and run the official Chrome DevTools MCP package from the NPM registry.\n
Evidence: The suggested MCP configuration uses npx chrome-devtools-mcp@latest (references/CHROME_DEVTOOLS_MCP.md). This targets the official package provided by the Google Chrome DevTools team.

melech-live-browser