The Agent Skills Directory

[COMMAND_EXECUTION]: The skill automatically detects and runs project-specific build and test commands (e.g., npm run build, flutter build, go test) in Phase 5 to verify the quality of the generated UI. These commands are executed locally within the project context.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) because it autonomously ingests and processes untrusted data from the codebase (Phase 0 and Phase 2) to generate code modifications and UI fixes.
Ingestion points: Entire project codebase, CLAUDE.md, .design-context.md, and docs/design-system.md.
Boundary markers: No specific delimiters or safety warnings are included in the prompts used by the sub-agents when processing project files.
Capability inventory: Ability to modify project files, execute build/test shell scripts, and perform git push operations.
Sanitization: The skill does not implement explicit sanitization or validation of the ingested codebase content before it is used to drive the AI's output.
[SAFE]: The skill's practice of reading and writing telemetry data to ~/.claude/projects/ is a local configuration management task intended for skill improvement and does not involve external data exfiltration.

design-pipeline