Xiaobai Workflow Enforcer 🔒

Mandatory workflows for AI Agents. Not suggestions, not "when appropriate" — mandatory.

Inspired by Superpowers (161K stars) which proved that enforced workflows transform chaotic AI outputs into reliable engineering.

Core Philosophy

Mandatory Workflows

Workflow 1: Pre-Action Design Gate 🔒

Trigger: Before any multi-step task or code creation

Mandatory Steps:

1. STOP. Don't write code yet.
2. Ask clarifying questions (minimum 3)
3. Present design/spec in chunks
4. Get user sign-off on design
5. Save design document

❌ Wrong:
User: Build me a scraper
Agent: [Writes code]

✅ Right:
User: Build me a scraper
Agent: Before I code, let me understand:
       1. What site are we scraping?
       2. What data do you need?
       3. How often should it run?
       4. Any rate limits to consider?
       [After answers, presents design]
       Does this design match what you need?

Workflow 2: Implementation Planning 🔒

Trigger: After design approval, before implementation

Mandatory Steps:

1. Break into 2-5 minute tasks
2. Each task has: file path, exact code, verification step
3. Present plan for approval
4. Save plan to checkpoint file

Plan Format:

## Task 1: Create scraper module (3 min)
- File: src/scraper.py
- Code: [exact code or pseudocode]
- Verify: `python -c "import scraper"`

## Task 2: Add rate limiting (2 min)
- File: src/scraper.py
- Code: [exact changes]
- Verify: Run with test request, check delay

...

Workflow 3: Test-First Gate 🔒

Trigger: Before implementing any function

Mandatory Steps:

1. Write test first
2. Run test, confirm it FAILS (RED)
3. Write minimal code to pass
4. Run test, confirm it PASSES (GREEN)
5. Refactor if needed
6. Commit only after GREEN

❌ Wrong:
[Writes function]
[Tests it manually]
"It works"

✅ Right:
1. Write test_function()
2. Run: pytest test_module.py
3. See: FAILED (expected)
4. Write function()
5. Run: pytest test_module.py
6. See: PASSED
7. Commit

Workflow 4: Execution Gate 🔒

Trigger: During task execution

Mandatory Steps:

1. Read task from plan
2. Execute exactly as planned
3. Verify (run command, check output)
4. Update checkpoint
5. Only then move to next task

Checkpoint Update:
- Task 1: DONE (verified: scraper.py imports successfully)
- Task 2: IN_PROGRESS
- Tasks 3-5: PENDING

Workflow 5: Verification Gate 🔒

Trigger: Before claiming "done" or "complete"

Mandatory Steps:

1. Run verification command
2. Show output to user
3. Confirm evidence matches claim
4. Only then say "done"

❌ Wrong:
"Scraper is done!"

✅ Right:
"Scraper implementation complete.