The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill interacts with and processes untrusted data from web pages, including page content, console messages, and network responses.
Ingestion points: Data is ingested via browser snapshots, screenshots, console logs, and network panel captures in the Setup and Test the criteria as a user sections.
Boundary markers: No specific delimiters are used to separate browser content from instructions.
Capability inventory: The skill can execute shell scripts (manual-env.sh, Playwright scripts in .scratchpad/), write to the filesystem (.scratchpad/), and perform network operations via browser tools.
Sanitization: No explicit sanitization of external content is mentioned.
[DYNAMIC_EXECUTION]: The skill generates and executes scripts to automate browser testing.
Evidence: Instructions in the Setup section direct the agent to "write one script in .scratchpad//scripts/ that drives the same checks" if using Playwright, and uses the evaluate script capability in Chrome DevTools.
[COMMAND_EXECUTION]: The skill triggers shell commands to manage the testing environment.
Evidence: It starts the application using a script at .scratchpad/<slug>/manual-env.sh and manages server processes.

browser-acceptance-testing