The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a wide range of CLI tools to manage Frappe environments, including frappectl, bench, git, uv, npx, cloudflared, and ngrok. This includes high-privilege operations like setting up network tunnels and managing local Frappe benches.
[EXTERNAL_DOWNLOADS]: The skill automatically manages its own dependencies and components. It uses the uv tool to install frappectl and npx to fetch and update the frappe-builder skill package from the internet.
[REMOTE_CODE_EXECUTION]: Automated update processes, specifically bench update and npx skills update, involve fetching, patching, and executing remote code. While these target vendor-controlled sources, they represent a powerful capability for remote execution.
[INDIRECT_PROMPT_INJECTION]: The skill processes UI definitions (blocks.json) and Python data scripts (data_script.py) which are ultimately rendered on the live site using Jinja templates and JavaScript.
Ingestion points: UI block trees and scripts are pulled from remote Frappe sites into local workdirs (scripts/builder.py).
Boundary markers: The skill lacks strict delimiters for processing these files, relying instead on manual review and linter warnings.
Capability inventory: The skill has extensive command execution capabilities through the subprocess module to interact with site management infrastructure.
Sanitization: scripts/lint.py provides some sanitization by checking for dangerous Jinja patterns (e.g., .__ access), but this does not eliminate the risk of runtime injection from malicious UI definitions.

frappe-builder