dev1-start

Starting a new feature development workflow involves performing a high-level codebase survey to identify the stack and dependencies, then scaffolding a dedicated workspace directory with initial context files for handoff.

Scaffolding a workspace for feature requests is best done by deriving a snake_case feature slug, generating an .llm/<slug>/ directory, and writing a tmp_context.md file to capture the original request, timestamp, and working directory.

Automating initial feature discovery requires reading project configuration files like CLAUDE.md to identify the project structure, stack, and dependencies, ensuring deterministic outputs for the next development phase.

Yes, this workflow summarizes high-level discovery findings and generates an exploration log, explicitly triggering the next design step via a command like /dev2-design to ensure a smooth handoff.

You do not strictly need existing artifacts to run feature discovery, but having configuration files like CLAUDE.md or .claude/CLAUDE.md present allows the workflow to accurately identify your stack and project layout.

Limitations of automating feature discovery include relying solely on high-level codebase surveys, which may miss deep architectural nuances, and generating only brief exploration logs suitable for initial design handoff rather than detailed implementation.