The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions in references/motion.md and references/upstream.md direct the agent to fetch external skill definitions (SKILL.md) and supporting files from the emilkowalski/skills and jakubkrehel/skills repositories on GitHub at runtime. These sources are not recognized as trusted organizations or well-known services.
[INDIRECT_PROMPT_INJECTION]: The mechanism for fetching and incorporating external instructions from non-trusted repositories creates a potential surface for indirect prompt injection attacks.
Ingestion points: Content is dynamically retrieved from external GitHub repositories as specified in references/motion.md and references/upstream.md.
Boundary markers: The instructions include explicit warnings to ignore "unrelated upstream workflow automation" and "upstream greetings," and to preserve the user's preferences over fetched guidance.
Capability inventory: The skill has capabilities to write files (05-direction.md), interact with the browser (chrome-devtools-cli), manage local servers (hitch), and run project scripts via package managers.
Sanitization: The skill explicitly forbids the automatic execution of downloaded scripts or the starting of subagents suggested by the fetched content, though the agent still reads and follows the imported instructions.
[COMMAND_EXECUTION]: The skill relies on local environment tools including chrome-devtools-cli, hitch, and project-specific package managers (e.g., npm, pip) and scripts to render design concepts and perform browser-based inspections.

webcraft-direction