# relay-baton v0.7.0 Release Notes Status: **Released** Date: 2026-05 Theme: **Review & Diagnose** — let you inspect *what actually changed* against the plan, record execution receipts, and track how a plan evolves — all deterministically, with **no model calls**. Also lays the append-only conversation-event groundwork for the future Agent Room (see `docs/AGENT_ROOM.md`). ## Summary v0.7 closes the loop opened by v0.5 plan-execute. After a plan runs you can now `review` the working-tree diff against the plan's Steps, mark steps `done` / `skipped` with append-only receipts, and see a section-level delta whenever a plan is re-authored. `status`/`budget`/`review` gained `--json` for scripting. Every new capability is deterministic and additive. ## Added - **`relay-baton review`** — deterministic correlation of the working-tree diff against `plan.md` Steps. No model call. For each step it extracts file-path tokens and matches them to git-changed files, reporting touched / untouched / uncorrelated steps and "unplanned" changed files, plus the first line of `test-results.md` and whether a handoff exists. Supports `--json`. - **Execution receipts** — `relay-baton receipt done ` / `receipt skip ` (with optional `--note`) append an immutable `[done]` / `[skipped]` line under an `## Execution receipts` heading in `plan.md`. Receipts are never rewritten; a correction is a new receipt. `receipt list` (with `--json`) reads them back. - **Plan diffing** — `relay-baton plan` now backs up the existing `plan.md` to `plan..md` before a re-plan and prints a section-level delta (`added` / `removed` / `changed` / `unchanged`) versus the previous plan. - **`--json` output** for `status`, `budget`, and `review` — machine-readable snapshots for scripting and the future Agent Room. - **Conversation event schema (DRAFT)** — append-only `.ai-session/conversation.jsonl` log (`ConversationLog`) and the `ConversationEvent` schema in `@relay-baton/shared`. `review` and `receipt` now persist a token-diet-friendly summary event (never a full diff/log; large content stays in its own artifact and is referenced). This is groundwork only — no chat/room command ships in v0.7. ## Changed - CLI version bumped to `0.7.0`. ## Docs - `docs/ROADMAP.md` and `docs/AGENT_ROOM.md` already describe v0.7's review/diagnose-as-events direction; this release implements the deterministic review/receipt/diff pieces. ## Tests - `PlanReview.test.ts` — path extraction, step parsing, and diff-vs-plan correlation (touched/untouched/uncorrelated, unplanned files). - `ConversationLog.test.ts` — append/read JSONL, empty-log handling, malformed line skipping. - `PlanDiffReceipts.test.ts` — section-level diff classification, plan backup / latest-backup discovery, and append-only receipts. - `review.test.ts` (cli) — human + `--json` output and conversation-event persistence. ## Not included (still deferred) - Actual chat/room command, TUI input box, bounded multi-turn `continue` → **v0.8+** (Agent Room). - 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 commands and `--json` flags are optional. `conversation.jsonl` is created lazily and is best-effort — logging never breaks an underlying command. ## Verification commands ```bash pnpm build pnpm test relay-baton plan "" --no-run relay-baton receipt done 1 --note "implemented" relay-baton receipt list relay-baton review relay-baton review --json relay-baton status --json relay-baton budget --json ``` None of these invoke a real Codex/Claude model.