Quick start¶
This guide walks through creating a spec from scratch and running the first implementation iteration.
Step 1: Start a new spec¶
Claude Code:
Codex:
The command will ask you 2-3 rounds of questions about scope, user roles, key behaviors, edge cases, and non-functional requirements. Then the spec-planner agent (reasoning tier) writes formal requirements and design:
- Requirements — It asks clarifying questions, identifies user roles, and writes user stories with EARS acceptance criteria
- Design — It defines the architecture, data models, and key sequences based on your requirements
After both phases, it generates a tasks.md file and syncs the tasks to Claude Code's todo list.
Your spec files are created at:
Step 2: Check the spec status¶
Claude Code:
Codex:
This shows task completion, dependency status, and which tasks are pending, in progress, or verified.
Step 3: Run the first implementation iteration¶
Claude Code:
Codex:
The agent picks the highest-priority pending task, implements it, verifies it, updates tasks.md, and commits. Run it again for the next task, or use /spec-loop (Claude Code) / $spec-loop (Codex) to run all tasks automatically.
Tip
If you have only one spec in .claude/specs/, the --spec-name argument is auto-detected. With multiple specs, pass --spec-name <name> explicitly.
Step 4: Run all remaining tasks automatically¶
To let the plugin work through all remaining tasks without manual intervention:
The loop checks for the <promise>COMPLETE</promise> marker in Claude's output and stops when all tasks are verified. See Execution for details on the three execution modes.
Step 5: Post-implementation¶
Once all tasks are complete, run the post-implementation pipeline:
spec-accept.sh --spec-name user-authentication
spec-docs.sh --spec-name user-authentication
spec-release.sh --spec-name user-authentication --version-bump minor
See Post-implementation for the full pipeline.
What's next¶
- Read Workflow overview to understand the full lifecycle
- See Commands for all available commands
- Read Concepts for a deeper explanation of EARS notation and task lifecycle