tdd-migrate

name	tdd-migrate
description	TDD workflow for migrations - orchestrate agents, zero main context growth
allowed-tools	["Task","TodoWrite","Write","Read","Bash"]

TDD Migrate

Orchestrate TDD migrations with agents doing all work. Main context stays clean.

When to Use

• "Port X from Python to TypeScript"
• "Create N adapters following existing pattern"
• "Migrate module to new architecture"
• "TDD implementation of multiple similar items"

Parameters

/tdd-migrate <source_path> <target_path> --pattern <reference> --items "item1,item2,item3"

• source_path: Path to analyze (existing code)
• target_path: Where to create new code
• pattern: Reference file/pattern to follow
• items: Comma-separated list of things to create

Workflow

Phase 0: YAML TODO List
    │
    ▼
Phase 1: TLDR Analysis ─────────────────┐
    │                                    │
    ▼                                    │ Parallel scouts
Phase 2: Write Failing Tests ───────────┤ per item
    │                                    │
    ▼                                    │
Phase 3: Implement (minimal) ───────────┤
    │                                    │
    ▼                                    │
Phase 4: Build + Pass Tests ────────────┘
    │
    ▼
Phase 5: QLTY Check ────────────────────┐
    │                                    │ Parallel
Phase 6: Review Agent Validates ────────┘
    │
    ▼
Phase 7: TLDR Diff (new vs reference)
    │
    ▼
Phase 8: Fix Issues (if any)
    │
    ▼
Complete

Key Principles

1. Main context = orchestration only

   • Never read files directly (use scout)
   • Never implement directly (use kraken/spark)
   • Never run tests directly (use validator)
   • Only pipe context and coordinate

2. Agents do ALL work

   Task	Agent
   Explore/analyze	scout
   Write tests + implement	kraken
   Quick fixes	spark
   Run tests/validate	validator
   Code review	critic

3. Parallel where independent

   • All items can be implemented in parallel if independent
   • Review + QLTY run in parallel
   • TLDR analysis runs in parallel with planning

4. Review after each major step

   • After implementation: critic reviews
   • After fixes: validator re-validates

Instructions

Step 0: Create YAML TODO

Write a YAML plan file to thoughts/shared/plans/<name>-tdd.yaml:

---

title: <Migration Name>
date: <today>
type: implementation-plan
approach: TDD (test → build → pass → review)

items:
  - name: item1
    file: <target_path>/item1.ts
    test: <target_path>/__tests__/item1.test.ts
    deps: []
  - name: item2
    # ...

reference: <pattern_file>

workflow:
  per_item:
    1: Write failing test
    2: Implement minimal
    3: Build
    4: Pass test
    5: QLTY check
    6: Review
  final:
    7: Integration test
    8: TLDR diff

Step 1: Launch Scout Agents (parallel)

Task (scout): Analyze <source_path> with TLDR
Task (scout): Analyze <pattern> to understand structure
Task (scout): Read migration handoff if exists

Step 2: Launch Kraken Agents (parallel per item)

For each item, launch ONE kraken that does full TDD:

Task (kraken): Implement <item> using TDD workflow
  1. Read pattern file
  2. Write failing test
  3. Implement
  4. Run: bun test <test_file>
  5. Run: qlty check <impl_file>

Step 3: Review + Validate (parallel)

Task (critic): Review all new files against pattern
Task (validator): Run full test suite
Task (validator): QLTY check all files

Step 4: Fix Issues

If critic/validator found issues:

Task (spark): Fix <specific issue>
Task (validator): Re-validate

Step 5: TLDR Diff

Task (validator): TLDR diff new files vs reference
  - tldr structure <new_file> --lang <lang>
  - tldr structure <reference> --lang <lang>
  - Compare patterns

Step 6: Update Continuity

Update ledger with completed work.

Example: Rigg Adapters

/tdd-migrate ~/your-project/src/sdk/providers \
  ~/your-project/src/sdk/providers \
  --pattern lmstudio.ts \
  --items "xai,cerebras,togetherai,deepinfra,perplexity"

Resulted in:

• 5 parallel kraken agents
• 39 tests passing
• All adapters working
• ~15 minutes total

Anti-Patterns (AVOID)

Bad	Good
Read files in main context	Launch scout agent
Write code in main context	Launch kraken/spark agent
Run tests in main context	Launch validator agent
Skip review	Always launch critic
Sequential items	Parallel krakens
Fix in main context	Launch spark

Agent Prompts

Scout (analysis)

Explore <path> to understand:
1. Structure/patterns
2. Interfaces/types
3. Dependencies
Return actionable summary for implementation.

Kraken (TDD)

Implement <item> using TDD:
1. Read <pattern> for structure
2. Write failing test to <test_path>
3. Implement minimal to <impl_path>
4. Run: <test_command>
5. Run: qlty check <impl_path>
Report: status, issues, files created.

Critic (review)

Review <files> against <pattern>:
1. Pattern compliance
2. Type safety
3. Missing registrations
4. Security issues
DO NOT edit. Report issues only.

Spark (fix)

Fix <specific issue>:
1. Read <file>
2. Make minimal edit
3. Verify fix

Validator (test)

Validate <files>:
1. Run <test_command>
2. Run qlty check
3. Report pass/fail/issues

Success Criteria

• All tests pass
• QLTY reports no issues
• Critic found no critical issues
• TLDR diff shows pattern compliance
• All items registered/exported properly