The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local scripts to handle its persistent state and project initialization.
It runs scripts/user_context_preflight.py to parse saved user preferences and assets.
It runs scripts/init_user_context.py to create the initial state directory structure in the home folder.
It references a Node.js script scripts/bootstrap-prototype.mjs (not included in the source files) to scaffold new projects.
[EXTERNAL_DOWNLOADS]: The skill facilitates the download of external software dependencies.
The onboarding and prototyping workflows instruct the agent to execute npm install, which fetches packages from the public npm registry.
[PROMPT_INJECTION]: The skill implements a data-ingestion pattern that exposes it to indirect prompt injection.
Ingestion points: The agent reads the content of ~/.open-cowork/state/plugins/product-design/user-context.md to ground its design tasks.
Boundary markers: The parsed content is injected into the agent's context without specific delimiters or "ignore instructions" guardrails.
Capability inventory: The skill has access to shell execution, file system writes, and browser automation.
Sanitization: The parsing logic in the Python preflight script does not sanitize the markdown content for embedded prompt injection commands.

product-design-user-context