panda 58dbe8e2e1 feat: ADR-0015 TDD loop architecture — plan→test→implement→review→judge ...

5 stories implemented via loop-tdd-v1:
- LOOP-1: Test writer Inngest function (agent/loop.test)
- LOOP-2: Restructured event chain (plan→test→implement→review→judge)
- LOOP-3: Reviewer rewritten as 4-question evaluator (no longer writes tests)
- LOOP-4: Judge consumes reviewer notes + LLM verdict
- LOOP-5: Integration verification (all wiring confirmed)

Key changes:
- New test-writer.ts: writes acceptance tests BEFORE implementation (TDD)
- review.ts: evaluates 4 questions (new tests? substantive? truthful? intent?)
  outputs structured JSON notes, no longer writes test files
- judge.ts: mechanical gates + reviewer notes + llmEvaluate verdict
  FAIL if reviewer flags issues even when tests pass
- plan.ts: emits agent/loop.test instead of agent/loop.implement
- 185 tests across 7 test files, all passing

See: ADR-0015 (docs/decisions/0015-loop-architecture-tdd-roles.md)

2026-02-14 22:35:02 -08:00

..

__tests__

test: [loop-mln0sl5l-y5k32r] [CLI-1] attempt-2 \u2014 reviewer tests

2026-02-14 16:54:29 -08:00

src

feat: ADR-0015 TDD loop architecture — plan→test→implement→review→judge

2026-02-14 22:35:02 -08:00

package.json

feat: generalize loop with PRD-level checks config + ADR-0010 PRD

2026-02-14 16:57:19 -08:00

tsconfig.json

feat: ADR-0015 TDD loop architecture — plan→test→implement→review→judge

2026-02-14 22:35:02 -08:00