✨ New feature — Test-Driven Repair
Closes the loop the existing TDD plugins didn't: we generate tests; now we also fix the code to satisfy them. Inspired by agent-harness-generator ADR-175; implemented via headless `claude -p` for tighter integration with our stack.
What you can do now
```bash
Failing CI test → verified fix PR for pennies
npx ruflo@latest hooks testgen tdd-repair
--repo .
--test tests/failing.test.ts
--test-command "npx vitest run tests/failing.test.ts"
--confirm
```
Or call `mcp__claude-flow__testgen_tdd_repair` from any agent.
How it works
- Pre-flight: run the test command. Refuse if it already passes (catches `--test-command` typos).
- Spawn `claude -p` with capability-restricted toolset (`--allowedTools Read,Edit,Bash`) and hard-capped budget (`--max-budget-usd`).
- Focused prompt: read the failing test, fix the source, do NOT modify the test, verify by running it.
- Re-run the test. The exit code IS the fitness function — no LLM-as-judge.
- If green: emit success + per-attempt usage (`cost_usd`). If red after `--max-attempts`: emit failure receipts; workspace left as-is for review.
Safety posture
- `--confirm` REQUIRED (dry-run otherwise)
- Hard `--max-budget-usd` cap (default $5)
- `--allowedTools Read,Edit,Bash` (no MCP, no network, no arbitrary writes)
- Prompt forbids modifying the test or adding dependencies
- 15 min hard timeout
Smoke proof
`docs/benchmarks/tdd-repair/SMOKE-2026-06-22.md` — built fixture with `add(a,b) → a-b` bug; pre-repair 2/2 fail, post-repair 2/2 pass.
Cost ladder
| Tier | Model | Per-attempt typical |
|---|---|---|
| 1 | Haiku (default) | $0.02 – $0.20 |
| 2 | Sonnet | $0.30 – $2.00 |
| 3 | Opus | $1.50 – $8.00 |
Conformant mode (`--no-test-oracle`)
Scoped for a follow-up ADR. Today `--no-test-oracle` returns a clear config error.
Distribution
| Package | latest | alpha | v3alpha |
|---|---|---|---|
| `@claude-flow/cli` | 3.14.0 | 3.14.0 | 3.14.0 |
| `claude-flow` | 3.14.0 | 3.14.0 | 3.14.0 |
| `ruflo` | 3.14.0 | 3.14.0 | 3.14.0 |
Cross-references
- 🔗 Inspired by: agent-harness-generator/packages/darwin-mode ADR-175
- 🔗 Smoke receipts: docs/benchmarks/tdd-repair/SMOKE-2026-06-22.md
- 🔗 Companion: testgen `test-gaps` skill (gap → fix loop scoped for next iteration)
🤖 Generated with RuFlo