The Agent Skills Directory

[DATA_EXPOSURE]: The skill instructs the agent to start a local HTTP server using python3 -m http.server during the diagram export process.
Evidence: Located in references/export.md, the procedure requires starting a server to provide a valid HTTP origin for browser-based ESM imports.
Impact: By default, this server binds to all network interfaces, serving the entire contents of the current working directory (which likely contains the codebase being analyzed) to the local network.
[DYNAMIC_EXECUTION]: The skill generates and executes JavaScript code at runtime within a Playwright browser context to perform SVG and PNG exports.
Evidence: The references/export.md file provides complex JavaScript blocks to be passed to the browser_run_code tool.
Behavior: This code dynamically imports external modules and handles data processing for binary file generation.
[EXTERNAL_DOWNLOADS]: The export procedure fetches external code from a remote CDN at runtime.
Evidence: The JavaScript executed in the browser imports the @excalidraw/utils library from https://esm.sh/@excalidraw/utils@0.1.2.
Note: While esm.sh is a well-known service, this introduces a runtime dependency on external infrastructure.
[COMMAND_EXECUTION]: The skill manages background shell processes to facilitate the export workflow.
Evidence: Instructions in references/export.md include running python3 -m http.server 8765 & and later terminating the process using kill $SERVER_PID.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface for indirect injection as it processes untrusted codebase data and uses it to drive automated tools.
Ingestion points: The skill analyzes codebase files using Glob, Grep, and Read tools to identify components and relationships (SKILL.md).
Boundary markers: No explicit delimiters or instructions are provided to the agent to treat codebase content as untrusted or to ignore embedded instructions.
Capability inventory: The skill has the ability to write files, execute shell commands (for server management), and run JavaScript in a browser context.
Sanitization: There is no evidence of sanitization logic for data extracted from the codebase before it is interpolated into diagram labels or browser code strings.

excalidraw