The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run to call the webctl utility for browser automation tasks, including navigating to URLs, taking screenshots, and capturing accessibility trees. It also executes which to verify the presence of the tool.
[EXTERNAL_DOWNLOADS]: The skill performs network requests using urllib.request to communicate with the Anthropic API (api.anthropic.com) or a Cloudflare AI Gateway. These requests transmit source code and screenshots for analysis by LLM models.
[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data from the codebase being analyzed and the accessibility tree of the target web application.
Ingestion points: Reads source code files in analyze.py, captures accessibility trees in describe.py, and extracts link labels in discover.py.
Boundary markers: Prompt templates use Markdown headers and code blocks to separate application data from instructions.
Capability inventory: Performs network requests to LLM APIs and executes the webctl browser tool via subprocess.
Sanitization: External content is encapsulated within JSON payloads when sent to the analysis APIs.

mapping-webapp