Feature Run

Mode: Work on all features or a specific one

⚡ UNATTENDED MODE (when no feature_id provided)

• NO questions allowed
• NO stopping for errors
• MUST complete all features

Mode Detection

If feature_id provided (e.g., feature-run auth.login):

• Work on that specific feature only
• Complete it and stop

If no feature_id (e.g., feature-run):

• Auto-complete all pending features
• Loop until all done
• UNATTENDED MODE ACTIVE - see rules below

Single Feature Mode

When feature_id is provided:

# STEP 1: Get the specified feature
agent-foreman next <feature_id>

# STEP 2: Implement (satisfy ALL acceptance criteria)
# ... write code ...

# STEP 3: Verify implementation
agent-foreman check <feature_id>

# STEP 4: Mark complete + commit
agent-foreman done <feature_id>

All Features Mode

When no feature_id:

# STEP 1: Check remaining features
agent-foreman status

# STEP 2: Get next priority feature
agent-foreman next

# STEP 3: Implement (satisfy ALL acceptance criteria)
# ... write code ...

# STEP 4: Verify implementation
agent-foreman check <feature_id>

# STEP 5: Mark complete + commit
agent-foreman done <feature_id>

# STEP 6: Handle result
# - Verification passed? → Continue to STEP 1
# - Verification failed? → Run 'agent-foreman fail <feature_id> -r "reason"', continue to STEP 1
# - All features processed? → DONE (show summary)

Rules

Priority Order

1. needs_review → may be broken (highest)
2. failing → not implemented
3. Lower priority number

Check TDD Mode First

Look for "!!! TDD ENFORCEMENT ACTIVE !!!" in agent-foreman next output.

TDD Workflow (when strict mode active)

# STEP 1: Get feature + TDD guidance
agent-foreman next <feature_id>

# STEP 2: RED - Write failing tests FIRST
# Create test file at suggested path
# Run tests: <your-test-command>
# Verify tests FAIL (confirms tests are valid)

# STEP 3: GREEN - Implement minimum code
# Write minimum code to pass tests
# Run tests: <your-test-command>
# Verify tests PASS

# STEP 4: REFACTOR - Clean up
# Clean up code while keeping tests passing

# STEP 5: Verify + Complete
agent-foreman check <feature_id>
agent-foreman done <feature_id>

CRITICAL: DO NOT write implementation code before tests exist in strict TDD mode!

NO QUESTIONS ALLOWED (CRITICAL)

When in ALL-FEATURES mode (no feature_id provided):

⛔ NEVER ASK THE USER ANY QUESTIONS

Forbidden phrases - NEVER output:

• "Should I...?"
• "Do you want me to...?"
• "Which approach would you prefer?"
• "I need clarification on..."
• "Before I proceed, could you..."

Required behavior:

• ALWAYS make autonomous decisions
• ALWAYS continue to the next feature
• ONLY stop when status shows 0 pending features

Loop Enforcement (MUST FOLLOW)

When in ALL-FEATURES mode:

1. CHECK: Run agent-foreman status - note pending count
2. LOOP: While pending count > 0: a. Run agent-foreman next - get feature details b. Implement feature - satisfy ALL criteria c. Run agent-foreman check <id> - verify d. Run agent-foreman done <id> - complete e. Handle result:
   • Pass? → Continue to step 2
   • Fail? → Run agent-foreman fail <id> -r "reason", continue to step 2
3. EXIT: Only when pending count = 0

Never Stop For:

• Questions
• Errors
• Failures
• Ambiguity
• Missing files
• Test failures

Only Stop When:

• agent-foreman status shows 0 features with failing or needs_review status

On Verification Failure

When agent-foreman done reports verification failure:

1. DO NOT STOP - This is the most critical rule
2. DO NOT ASK - Never ask user what to do
3. Mark the failed feature using the fail command:

   agent-foreman fail <feature_id> --reason "Brief description of failure"
   

4. IMMEDIATELY run agent-foreman next for the next feature
5. Continue the loop - DO NOT pause, reflect, or ask for guidance

This applies to ALL errors, not just verification failures.

Exit When

CRITICAL: NEVER stop due to verification failure - always use agent-foreman fail and continue!

Loop Completion

When all features have been processed:

1. Run agent-foreman status to show final summary
2. Report counts:
   • X features passing
   • Y features failed (need investigation)
   • Z features needs_review (dependency changes)
   • W features still failing (not attempted)
3. List features that failed verification with their failure reasons