The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses a variety of shell commands through the compose-preview CLI tool to manage UI designs. Key commands include starting the local builder (compose-preview ui-builder), handling authentication (compose-preview auth request), and exporting design files to Kotlin (compose-preview design export).
[DYNAMIC_EXECUTION]: In the references/m3-catalog.md file, a Python script is provided to be executed via a shell heredoc (`python3
"$SAVED_RESULT" m3-catalog <<'EOF'`). This script dynamically processes JSON tool outputs to regenerate documentation tables at runtime.
[EXTERNAL_DOWNLOADS]: The skill mentions that the compose-preview tool fetches its own server distribution upon first use. Additionally, it interacts with external domains such as preview.coo.ee and ui.coo.ee to host designs and provide MCP endpoints.
[INDIRECT_PROMPT_INJECTION]: The skill's primary workflow involves reading external data that the agent does not control, such as design document content (.uid files) and designer comments retrieved via ui_builder_list_comments.
Ingestion points: Design documents fetched via ui_builder_get_design and designer feedback threads from ui_builder_list_comments in SKILL.md.
Boundary markers: The instructions recommend reviewing comments before performing edits but do not specify delimiters or warnings to ignore instructions embedded in those comments.
Capability inventory: The skill possesses extensive mutation and execution capabilities, including design modification (ui_builder_apply), code generation (ui_builder_export), and shell command execution.
Sanitization: No explicit sanitization or filtering of the external design data or comments is mentioned before they are processed by the agent.

compose-ui-builder