The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/_skill_cli.py contains a utility function run_command that wraps subprocess.run to execute system commands. While this function does not use shell=True and is a standard helper for agent environments, it provides a capability for the skill to interact with the underlying operating system.
[DYNAMIC_EXECUTION]: The skill relies on the Penpot MCP execute_code tool to dynamically generate and execute JavaScript code within the Penpot plugin environment. This is the primary mechanism for reading and modifying design assets, but it involves runtime code generation based on agent instructions and templates.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources, specifically Penpot design files (pages, structures, tokens, and components) and user-supplied design briefs. This creates a potential surface for indirect prompt injection if the processed data contains instructions intended to influence the agent's behavior during its analysis or critique phases.
[EXTERNAL_DOWNLOADS]: The skill references and incorporates design guidelines and third-party resources from trusted organizations, including Anthropic's public repository and other well-known industry sources such as Paul Bakaus's Impeccable guidelines. These references are used to steer design direction and calibrate output.

taste-penpot