# relay-baton v0.6.0 Release Notes

Status: **Released**

Date: 2026-05

Theme: **Trust & Verify** — make the existing pipeline provable and observable
before adding new surface area. No new workflow mode; every change is additive
and backward compatible.

## Summary

v0.6 hardens what v0.4/v0.5 built. You can now *prove* the harness is wired
correctly without spending a single model token: `verify` runs a simulated
end-to-end pass, and `doctor --deep` gives a full OK/WARN/FAIL diagnostics
checklist. The TUI now shows the current workflow mode and plan/compression
state at a glance.

## Added

- **`relay-baton verify`** — a simulated end-to-end check that runs **no real
  model calls** by default. It validates: repoRoot resolution, `doctor` core
  checks, fallback-phrase detection (real phrases detected, grep/result-like
  lines ignored), the handoff no-run workflow (against a throwaway temp repo,
  never touching your real `.ai-session/`), token-budget compliance, and the
  API-key env block. Exits non-zero on failure.
  - Options: `--project`, `--path`, `--diet`, `--real-agents`, `--keep-temp`,
    `--verbose`.
  - `--real-agents` is **experimental scaffolding only** and is never executed.

- **`relay-baton doctor --deep`** — extended diagnostics on top of the core
  checks: Node/pnpm versions, adapter-args sanity (flags forbidden args like
  `--full-auto` / `--ask-for-approval` / `bypassPermissions` as failures),
  `.gitattributes`/CRLF hints on Windows, project-registry health,
  `.ai-session/` health, and the latest handoff budget / `plan.md` status /
  compression (rotated-log) status. **API key values are never printed** — only
  presence.

- **TUI mode panel** — the dashboard now shows current mode
  (idle/run/handoff/plan/execute/compress-context), latest plan status, latest
  compression status, and latest handoff info. Display only — the TUI still
  never launches an agent.

## Changed

- `doctor` output is reorganized into Core / Deep / Summary sections with an
  aggregate pass/warn/fail count; it exits non-zero only on hard failures.
- CLI version bumped to `0.6.0`.

## Docs

- New `docs/ROADMAP.md` covering v0.6 (Trust & Verify), v0.7 (Review &
  Diagnose), v0.8 (Adapter Expansion), v0.9 (Automation & Runtime), and v1.0
  (Stable Local Release).
- README gained a plan-execute and context-compression workflow walkthrough
  plus a consolidated verification-commands section.

## Tests

- `diagnostics.test.ts` — core/deep checks, forbidden-arg detection, and the
  guarantee that API key values are never printed.
- `verify.test.ts` — simulated pipeline passes with exit 0, does not write into
  the cwd's `.ai-session/`, and treats `--real-agents` as non-executing.
- `docs.test.ts` — ROADMAP and v0.6 release notes (EN + KO) exist and the
  README documents plan/execute/compress-context/verify.

## Not included (still deferred)

- Plan execution receipts, plan diffing, and the multi-turn loop → **v0.7+**.
- Adaptive per-model thresholds and cross-session compression → **v0.9**.
- OpenCode / Gemini / Aider adapters → **v0.8**, on demand.
- Semantic summarization, exact tokenizers, auto commit/push/PR — permanently
  out of scope.

## Upgrade notes

- Fully additive. No config migration required. New CLI flags are optional.
- `verify` and `doctor --deep` make no network or model calls and write nothing
  to your repo (verify uses an OS temp dir it cleans up).

## Verification commands

```bash
pnpm build
pnpm test
relay-baton doctor
relay-baton doctor --deep
relay-baton verify
relay-baton verify --project relay-baton
relay-baton handoff --to claude --no-run --diet caveman
relay-baton budget
relay-baton project current
```

None of these invoke a real Codex/Claude model.