TDD Command

Test-Driven Development workflow: write tests first, then implement to make them pass.

Step 1: Determine Topic

If task-focus is empty, use the most recent PLAN file or ask the user.

Output: A topic string.

Step 2: Resolve Topic Directory

All planning files for a topic live in .agent_planning/<topic-slug>/.

Process:

1. Generate a slug from the topic (lowercase, hyphenated, short)

2. Check for existing topic directories

3. Exact match exists → Use it Similar directories found → Ask user to choose No match → Create new directory and chain to /do:plan $TOPIC

Output: Topic directory path (e.g., .agent_planning/auth/)

Step 3: Find or Create a Plan

Check topic directory for PLAN-*.md (newest by timestamp).

• Plan exists → Proceed to TestLoop
• No plan → Run /do:plan $TOPIC, then proceed

TestLoop

Tests MUST be (TestCriteria):

• useful (no useless tests)
• complete (test all edge cases)
• flexible (allow refactoring without changing tests)
• Fully automated using existing or standard framework

Step 1: Design and write tests Use the do:functional-tester agent to design and write high-level functional tests that validate real user workflows and follow all TestCriteria.

Step 1b: Display results - Show functional-tester's summary to user before proceeding.

Step 2: Evaluate tests Use the do:project-evaluator agent to evaluate the tests just written. Evaluate in context of the plan to ensure they follow TestCriteria.

Step 2b: Display results - Show project-evaluator's summary and loop decision to user.

Step 2c: Handle PAUSE (if applicable) If project-evaluator returns PAUSE with ambiguities about test design:

1. Use do:researcher to explore the testing question
2. Use project-evaluator (research evaluation mode) to assess if research is sufficient
3. If sufficient, project-evaluator makes the decision
4. Continue TestLoop with resolved ambiguity

Exit Condition: When TestCriteria are met with NO EXCEPTIONS, exit and proceed to ImplementLoop.

Continue Condition: If tests don't meet TestCriteria, restart the loop with specific feedback.

ImplementLoop

Step 1: Implement Use the do:iterative-implementer agent to implement the functionality that makes tests pass. The agent will automatically detect TDD mode and iterate until all tests pass.

Step 1b: Display results - Show iterative-implementer's summary (tests passing/failing, files, commits) to user.

Step 2: Evaluate implementation Use the do:work-evaluator agent to evaluate the current implementation.

Step 2b: Display results - Show work-evaluator's summary and loop decision to user.

Step 2c: Handle PAUSE (if applicable) If work-evaluator returns PAUSE with ambiguities about implementation:

1. Use do:researcher to explore the specific technical question
2. Use work-evaluator (research evaluation mode) to assess if research is sufficient
3. If sufficient, work-evaluator makes the decision
4. Continue ImplementLoop with resolved ambiguity

Exit Condition: No outstanding issues with well-defined solutions.

Continue Condition: If known issues remain and solution is well defined, restart the loop.

Final Step

After BOTH loops complete, run /do:status to show current state.

Display summary:

═══════════════════════════════════════
TDD Complete
  Topic: $TOPIC
  TestLoop: n iterations | ImplementLoop: m iterations
  Tests: all passing | Files: [count] | Commits: [count]
  Research: [n decisions made OR "none needed"]
Next: Review STATUS or continue development
═══════════════════════════════════════

Important Notes

• PAUSE triggers automatic research - user only involved if research cannot resolve after 3 iterations
• Both loops must complete before final evaluation
• Tests define the contract - implementation must pass through real functionality