The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze data from arbitrary external sources, including web pages, API responses, and network traffic (HAR files). This creates an attack surface where malicious instructions embedded in the target content could attempt to influence the agent's behavior during the reconnaissance process. While the skill includes 'gates' to ensure the integrity of the technical report, it does not explicitly detail sanitization protocols for preventing the LLM from following instructions discovered on target sites.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool to run commands, including the use of npx to execute the agent-browser utility. It also references the surfacer compiler for generating interface descriptors from mapped data.
[DYNAMIC_EXECUTION]: To extract data from modern web applications (SPAs), the skill uses agent-browser eval to execute JavaScript within the context of the target web page. This is used to inspect framework globals (e.g., window.__NEXT_DATA__) and the React component tree. While restricted to the browser context, this represents dynamic code execution on data derived from external sources.

surface-recon