The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill interacts with live browser sessions and processes external web content, which inherently creates a surface for indirect prompt injection attacks.
Ingestion points: Data enters the agent's context through take_snapshot and evaluate_script calls in references/browser-playbook.md and references/deep-dom.md.
Boundary markers: While the skill defines rigorous evidence verification and routing rules, it does not explicitly define delimiters or boundary markers for the data when it is interpolated into subsequent prompts.
Capability inventory: The skill possesses high-privilege browser capabilities, including click, type_text, navigate_page, upload_file, and evaluate_script across all reference files.
Sanitization: There is no explicit mention of sanitizing, escaping, or filtering content extracted from the browser before it is processed by the agent.
[DYNAMIC_EXECUTION]: The skill frequently uses evaluate_script to run JavaScript in the browser context. This is used for DOM-level inspection, extracting structured data, and injecting temporary markers to bridge selectors to accessibility tree targets.
[EXTERNAL_DOWNLOADS]: The skill's setup instructions guide the user to download and run the chrome-devtools-mcp package from the npm registry. This is a well-known service and the tool is from a recognized developer source.

use-my-browser