Relay Baton
dgl1231/relay-baton
Transfere sessões de código entre agentes sem perder contexto
Ver no GitHubAgent que transfere sessões de codificação entre Codex CLI e Claude Code preservando todo o contexto, histórico de diffs e estado do repositório sem necessidade de repastagem manual. Compacta o estado da codificação em handoffs portáveis entre agentes de IA, eliminando perda de contexto em mudanças de ferramentas.
Instalação
npx convez-ai dgl1231/relay-baton1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Continuar tarefas de refatoração quando um agente atinge limite de quota
- Migrar contexto de codificação entre diferentes CLIs e interfaces
- Preservar estado completo do projeto ao trocar de ferramenta IA
Veredito

Solução técnica sólida para um problema real: perda de contexto ao alternar entre agentes de IA. Implementação bem pensada com suporte multi-plataforma, mas nicho específico para quem trabalha com múltiplos CLIs de codificação.
README
A tradução para o português deste README ainda está na fila.
relay-baton
Portable continuation infrastructure for coding agents.
Pass compressed coding state between Codex CLI, Claude Code, and whatever ships next — without re-pasting the chat log, the diff, or the repo.
English · 한국어 · 日本語 · 简体中文 · 繁體中文 · Español · Français · Deutsch · Português · Русский
# Codex hits a quota wall mid-task. relay-baton notices, builds a compact
# handoff from the repository's actual state, and resumes the work in Claude.
$ relay-baton run "refactor the upload pipeline" --diet caveman
[relay-baton] codex: ... rate limit exceeded ...
[relay-baton] fallback pattern detected: rate limit exceeded
[relay-baton] building handoff for claude...
[relay-baton] claude resumed from .ai-session/handoff.md
Download
Prebuilt, single-file executables are attached to every release — no Node install required to run them.
One-line CLI install (verifies SHA256SUMS, no admin required):
# macOS / Linux
curl -fsSL https://raw.githubusercontent.com/dgl1231/relay-baton/main/install/install.sh | sh
# Windows PowerShell
iwr https://raw.githubusercontent.com/dgl1231/relay-baton/main/install/install.ps1 -UseB | iex
| OS | Binary | Get it |
|---|---|---|
| macOS (Apple Silicon) | relay-baton-macos-arm64 | Latest release |
| Linux (x64) | relay-baton-linux-x64 | Latest release |
| Windows (x64) | relay-baton-windows-x64.exe | Latest release |
# macOS / Linux
chmod +x relay-baton-macos-arm64
./relay-baton-macos-arm64 --version
# Windows
.\relay-baton-windows-x64.exe --version
Desktop app (Tauri shell over the same CLI — see desktop/):
| OS | Installer | Get it |
|---|---|---|
| macOS (Apple Silicon) | relay-baton_…_aarch64.dmg | Latest release |
| Linux (x64) | relay-baton_…_amd64.AppImage | Latest release |
| Windows (x64) | relay-baton_…_x64_en-US.msi | Latest release |
Desktop installers are signed/notarized only when release signing secrets are configured. Unsigned fallback builds still ship, so expect Gatekeeper / SmartScreen prompts on unsigned releases.
SHA256SUMSis attached to new releases for CLI installer verification. Desktop update checks are opt-in and manual. Signed updater artifacts andlatest.jsonare published only when Tauri updater signing secrets are configured for a release.Older versions: pick any tag on the releases page. Prefer to build from source instead? See Install below.
Why this exists
AI coding work is fragmenting across tools. A real session looks like this:
- Codex CLI for one batch of edits.
- Claude Code for another.
- A laptop in the morning, a different machine in the evening.
- A context window that fills up, falls over, or quietly truncates.
The default way to move work between agents today is copy-paste the chat log — or worse, the whole repo dumped into a prompt. That approach has three problems:
- Tokens. Chat logs are mostly noise. You pay for that noise every turn.
- Continuity. The next agent doesn't get intent; it gets transcripts.
- Fragility. One missed file, one stale diff, and the agent restarts from a wrong premise.
relay-baton is a local harness that sits underneath the agents and carries the minimum sufficient state across the handoff: a compact summary, a repo map, and file references — not a transcript.
Spend as few tokens as possible while merging Codex CLI and Claude Code CLI into a single workflow.
When one agent hits a quota / context limit, the next one resumes from the exact repository state — without re-pasting the whole log, diff, or repo.
The idea
┌─────────┐ ┌──────────────┐ ┌──────────────┐ ┌────────┐
│ Codex │ → │ Fallback │ → │ Token Diet │ → │ Claude │
│ exec │ │ Detector │ │ Handoff │ │ Code │
└─────────┘ └──────────────┘ └──────────────┘ └────────┘
↓ ↓
.ai-session/handoff.md, compact-state.md,
repo-map.md, full-diff.patch, commands.log
A baton-pass for coding agents:
- Detect when the current agent stops being useful (quota, context, rate, errors).
- Capture what matters — repo state, changed files, decisions, next step.
- Compact it under a budget the next agent can actually consume.
- Hand off with quality gates so the receiver isn't fed broken context.
The handoff is a small file (.ai-session/handoff.md) plus references. Everything heavy — the full diff, the full log, the full repo map — stays on disk and is loaded on demand.
Workflow
$ relay-baton init # create .ai-session/ in the current repo
$ relay-baton doctor # check git, codex, claude, env, config
$ relay-baton run "fix flaky upload test" --diet balanced
[relay-baton] running codex exec --sandbox workspace-write "fix flaky upload test"
... codex output streams here ...
[relay-baton] fallback pattern detected: maximum context length
[relay-baton] building handoff for claude...
[relay-baton] HandoffQualityGate: ok
[relay-baton] TokenDietQualityGate: ok
[relay-baton] running claude --permission-mode acceptEdits -p "<continuation>"
... claude resumes, edits files, finishes ...
$ relay-baton status # what state is the session in?
$ relay-baton budget # how much of the diet budget did we use?
Manual handoff, no auto-fallback:
$ relay-baton handoff --to claude --no-run --diet caveman
Across multiple repos:
$ relay-baton project add /path/to/repo-a --diet caveman
$ relay-baton project add /path/to/repo-b --diet balanced
$ relay-baton project switch repo-b
$ relay-baton run "wire up the new metrics endpoint"
Plan-execute mode
A proactive alternative to reactive fallback: have a planner write a structured plan, gate it for completeness, then have an executor implement it.
# 1) Planner (default: Claude) writes .ai-session/plan.md, gated by PlanQualityGate.
$ relay-baton plan "add rate limiting to the upload endpoint" --with claude
# 2) Inspect / edit the plan if you want.
$ cat .ai-session/plan.md
# 3) Executor (default: Codex) implements the plan.
$ relay-baton execute --with codex
# Or do both in one shot once the plan passes its gate:
$ relay-baton plan "add rate limiting" --then-execute
# Scaffold an empty plan template without launching an agent:
$ relay-baton plan "explore options" --no-run
Context-compression mode
Deterministic mid-session compaction of state.md / commands.log so an agent
runs longer before hitting a context wall. No model call, fully reversible
(raw artifacts are rotated to commands.log.full.<ts>).
# See what would change without writing anything.
$ relay-baton compress-context --dry-run
# Compress when weight/budget exceeds the threshold (default 0.8).
$ relay-baton compress-context --threshold 0.8
# Force compression regardless of threshold.
$ relay-baton compress-context --force
Verify the harness (no model calls)
# Simulated end-to-end check: resolution, fallback detection, handoff no-run,
# token budget, API-key env block. Exits non-zero on failure.
$ relay-baton verify
# Extended environment diagnostics (OK/WARN/FAIL checklist).
$ relay-baton doctor --deep
Neither verify nor doctor --deep makes any network or model call, and
verify writes nothing into your repo (it uses a throwaway temp dir).
Quick start
# 1) Install & build
pnpm install
pnpm build
# 2) Log in to both CLIs in one go
pnpm relay-baton login
# 3) Sanity check
pnpm relay-baton doctor
# 4) Start a task on Codex; auto-handoff to Claude on quota/context limit
pnpm relay-baton run "Fix the mail attachment upload flow" --diet balanced
pnpm relay-batonis an alias fornode packages/cli/dist/index.js. For a global binary, link it withpnpm -F @relay-baton/cli link --global(optional).
One-line install via an agent
install/install.md is both a human install guide and an instruction surface that coding agents can execute as-is:
# Codex CLI
codex exec --sandbox workspace-write "Read https://github.com/<your-org>/relay-baton/blob/main/install/install.md and install relay-baton on this machine. Do not print or store API keys. Finish by running 'pnpm relay-baton doctor'."
# Claude Code (headless)
claude --permission-mode acceptEdits -p "Read https://github.com/<your-org>/relay-baton/blob/main/install/install.md and follow it step by step. Do not print or store API keys. End by running 'pnpm relay-baton doctor'."
Features
- Automatic fallback. Detects
quota exceeded,rate limit exceeded,maximum context lengthand similar in agent output and hands off to the fallback agent. False-positive guards skip grep results and prose about the patterns themselves. - Token diet. Five deterministic compaction profiles (
off · lite · balanced · caveman · ultra). Lock files, build output, minified bundles, and binary blobs are excluded; logs are tailed; repo maps replace source. - Plan-execute mode. A proactive alternative to reactive fallback: a planner agent (default Claude) writes a structured
.ai-session/plan.md, gated for completeness, then an executor agent (default Codex) implements it.relay-baton plan "<task>" --then-execute. - Context compression. Proactive, deterministic mid-session compression of
state.md/commands.logso an agent runs longer before hitting a context wall.relay-baton compress-context, plus an auto-pass insiderun. - Quality gates. A handoff is verified before the fallback agent is launched — required sections present, budget respected, no inlined manifests.
- Auth-safe by default.
OPENAI_API_KEY/ANTHROPIC_API_KEYare stripped from child processes. Opt-in only via--allow-api-key-env. No keys are ever stored, printed, or logged. - Project registry. Register multiple repos once, switch the active project, run commands against any of them with
--projector--path. - Ink TUI. Project / session dashboard with diet cycling, budget refresh, and no-run handoff generation. Never launches agents.
- No API of its own. relay-baton never calls the OpenAI / Anthropic API directly. It only spawns the local
codex/claudeCLIs as subprocesses.
Architecture
packages/
shared/ # types, constants
core/ # UI-independent business logic
agents/ # AgentAdapter, CodexAdapter, ClaudeCodeAdapter, FallbackDetector
git/ # GitService, RepoMapGenerator
session/ # SessionManager (.ai-session/ lifecycle)
token-diet/ # BudgetManager, DiffCompactor, LogCompactor, HandoffCompactor, ...
handoff/ # HandoffGenerator, HandoffQualityGate, TokenDietQualityGate, PromptBuilder
workflow/ # BatonWorkflow — the orchestration
projects/ # ProjectRegistry, ProjectManager, ProjectResolver
cli/ # commander entrypoints — thin shells over core
tui/ # Ink dashboard — read-only over core
Design rules:
coreknows nothing about UI.cliandtuihold no business logic — they wire user input tocore.- Each agent is an
AgentAdapter. Addingopencode,gemini,aidermeans adding an adapter, not patchingcore. - Everything written to disk lives under
.ai-session/. relay-baton never edits your source files.
Usage
Single-repository flow:
relay-baton init
relay-baton doctor
relay-baton run "Fix the failing upload flow" --diet balanced
relay-baton status
relay-baton budget
Generate a handoff without launching the next agent:
relay-baton handoff --to claude --no-run --diet caveman
Register and use a project:
relay-baton project add /path/to/relay-baton --name relay-baton --diet caveman --primary codex --fallback claude
relay-baton project list
relay-baton project switch relay-baton
relay-baton status --project relay-baton
relay-baton handoff --to claude --no-run --project relay-baton --diet caveman
Direct path without registering:
relay-baton status --path /path/to/repo
relay-baton budget --path /path/to/repo
Open the TUI dashboard:
relay-baton tui --project relay-baton
Commands
| Command | Description |
|---|---|
relay-baton init | Create .ai-session/ in the current repo |
relay-baton doctor | Check git / codex / claude / env / config (--deep for extended diagnostics) |
relay-baton verify | Simulated end-to-end check — no real model calls (--diet, --verbose, --keep-temp) |
relay-baton login [agent] | Run Codex / Claude auth flows (codex / claude / all) |
relay-baton run "<task>" | Run primary agent, detect fallback, hand off |
relay-baton handoff --to claude | Manual handoff (--diet, --no-run, --force) |
relay-baton handoff history | List the current + backed-up handoff documents (metadata only) |
relay-baton plan "<task>" | Plan-execute mode: planner writes plan.md (--with, --no-run, --then-execute) |
relay-baton execute | Plan-execute mode: executor implements plan.md (--with, --from) |
relay-baton review | Deterministically review the diff against plan.md Steps — no model call (--json) |
relay-baton receipt done/skip/list | Append-only execution receipts for plan steps (--note, --json) |
relay-baton compact / squeeze | Rebuild compact-state, repo-map, full-diff |
relay-baton compress-context | Compress running state.md / commands.log when over budget (--dry-run, --threshold) |
relay-baton budget | Show context-budget usage |
relay-baton compress <file> | Deterministically compress a markdown file |
relay-baton status | Print session status |
relay-baton project add <path> | Register a project directory |
relay-baton project list / current / doctor | Inspect registered projects |
relay-baton project switch <name-or-id> | Change the active project |
relay-baton project remove <name-or-id> | Unregister a project |
relay-baton tui | Launch the Ink dashboard (display-only; never spawns agents) |
relay-baton chat / room | Agent Room: turn-based, confirmation-first multi-agent REPL (v0.9: /continue --max-steps N, /replan, /replay) |
relay-baton replay | Replay the recorded conversation timeline — read-only, no model call (--json, --session, --kind, --limit) |
Project-aware commands accept --project <name-or-id> and --path <repoPath>.
Resolution priority: --path > --project > active project > current working directory. compress <file> stays cwd-based — it's a file-level utility.
Project registry
Projects are stored at ~/.relay-baton/projects.json. Set RELAY_BATON_PROJECTS_FILE to use a different path (CI, sandboxing, tests).
relay-baton project add /path/to/relay-baton --name relay-baton --diet caveman --primary codex --fallback claude
relay-baton project list
relay-baton project switch relay-baton
project add requires the path to exist and be a directory. It may be a plain, non-git project folder; git-backed commands such as run and handoff still require a git repository when executed. Duplicate paths are not added twice. A corrupt projects.json is backed up to projects.json.corrupt-<timestamp>.bak and reset to an empty registry — commands keep working. Read-only commands (status, budget, doctor) do not write the registry; only run and handoff update lastUsedAt.
Handoff flow
relay-baton run "..."spawns the primary CLI as a subprocess.- stdout / stderr stream live and append to
.ai-session/commands.log. - Output is scanned for fallback phrases. Grep-style result lines and prose explaining the patterns are skipped to avoid false positives.
- On fallback, relay-baton collects git state, builds a repo map, compacts the state, and writes the handoff document.
- HandoffQualityGate + TokenDietQualityGate run before the fallback agent launches.
- The fallback agent resumes from
.ai-session/handoff.mdand the referenced files.
Token diet
Principles:
- Don't paste the whole thing.
- Prefer log tail + known errors over raw logs.
- Prefer repo map over full source.
- Prefer focused diff (lock / build /
*.min.jsexcluded) over full diff. - Delegate large content to file references.
Profiles
| Profile | Intent | Notes |
|---|---|---|
off | Minimal truncation | Mostly pass-through |
lite | Light cleanup | Strip blank lines, duplicate instructions |
balanced (default) | Day-to-day | Compact state + diff summary + log tail + repo map |
caveman | Aggressive minimal | Short direct bullets; no full diff/log inline |
ultra | Extreme | Almost everything via reference |
cavemanis not a silly tone — it means aggressive minimal-context. Technical accuracy is preserved.
.ai-session/ files
| File | Role |
|---|---|
task.md | The user's task |
state.md | Current state (Goal / Done / In Progress / Remaining / Decisions / Risks / Next Step) |
compact-state.md | Deterministic compaction of state.md |
handoff.md | Handoff document for the next agent |
decisions.md | Decision log |
changed-files.md | List of changed files |
repo-map.md | Directory tree + key files |
commands.log | Codex / Claude execution log (append-only) |
errors.md | Error notes |
test-results.md | Test / build results |
full-diff.patch | git diff HEAD snapshot |
context-budget.json | Token-diet usage snapshot |
session.json | Session metadata |
Configuration — relay-baton.config.json
{
"primaryAgent": "codex",
"fallbackAgent": "claude",
"agents": {
"codex": { "command": "codex", "args": ["exec", "--sandbox", "workspace-write"] },
"claude": { "command": "claude", "args": ["--permission-mode", "acceptEdits", "-p"] }
},
"fallbackPatterns": [
"usage limit reached",
"rate limit exceeded",
"context length exceeded",
"context limit exceeded",
"token limit exceeded",
"quota exceeded",
"quota limit",
"insufficient quota",
"maximum context length",
"too many requests"
],
"authPolicy": {
"mode": "cli-session",
"allowApiKeyEnv": false,
"warnIfApiKeyEnvDetected": true,
"blockedEnvVars": ["OPENAI_API_KEY", "ANTHROPIC_API_KEY"]
},
"tokenDiet": {
"enabled": true,
"profile": "balanced",
"outputCompression": true
}
}
If no config file exists, defaults are used.
TUI
relay-baton tui opens an Ink dashboard for the active project, or relay-baton tui --project <name-or-id> for a specific one.
- Sidebar: active project, registered projects, selected diet, primary / fallback agents, key hints.
- Main panels: task, session status, active agent, fallback reason, repo path, changed files, compact-state preview, token budget, log tail, recent errors.
- Keys:
qquit ·rrefresh ·pnext project ·dcycle diet (in memory) ·breload budget ·hgenerate handoff with no agent run.
The TUI never launches Codex or Claude. It reads session files and can generate a no-run handoff.
Quality gates
Handoff Quality Gate — runs before fallback launch:
- Required files exist and are non-empty.
handoff.mdcontains the sectionsGoal,Previous Agent,Next Agent,Changed Files,Known Errors,Next Steps.- Fails the launch if anything is missing.
--forceoverrides.
Token Diet Quality Gate:
handoff.md≤maxHandoffCharsfor the active profile.Token Diet Summarysection present.- Truncate marker present when content was actually truncated.
commands.log/AGENTS.md/CLAUDE.mdbodies must not be inlined.- For
caveman/ultra, the Important Diff block stays withinmaxDiffChars.
Auth & safety
- Never stores, prints, or logs API keys.
OPENAI_API_KEY/ANTHROPIC_API_KEYstripped from child processes by default.- Only
--allow-api-key-envorauthPolicy.allowApiKeyEnv: truelets them through. doctorreports whether each variable is set — but never prints its value.- Refuses to run
run/handoffoutside a git repo. - Backs up an existing
handoff.mdwith a timestamp suffix. commands.logis append-only.- relay-baton never modifies your source files directly — only
.ai-session/. compressdoes not overwrite the input file unless--writeis passed.
vs the alternatives
| Approach | What carries over | Token cost | Continuity | Failure mode |
|---|---|---|---|---|
| Raw chat export | Entire transcript | High (most of it is noise) | Brittle — agent re-reads its own thinking | Context window overflow |
| Copy-paste prompting | What the human remembered | Variable | Fragile — depends on memory | Silent drift from real state |
| Whole-repo dump | Everything | Very high | Strong but expensive | Truncated mid-file by the model |
| relay-baton | Compact summary + repo map + file references | Low, bounded by profile | Strong — driven by actual repo state | Fails loudly via quality gates |
Philosophy
relay-baton is small sharp tooling for AI-native development workflows.
- Local-first. Everything lives on your disk. No cloud, no daemon, no telemetry, no account.
- Composability. A
.ai-session/directory is just files. Read them, grep them, diff them, ship them in a PR. - Lightweight state transfer. A handoff is a markdown file, not a database.
- Deterministic over clever. No LLM-driven summarization in the harness itself — if the model gets summarization wrong, the handoff lies. relay-baton uses character budgets, structural rules, and explicit references.
- Repo state is the source of truth. The conversation is interpretation; the repo is fact.
- Token efficiency is the feature, not a knob hidden in a menu. Every design decision is measured against "does this bloat the handoff?"
Design principles
- It's work handoff, not chat relay.
- The current repository state beats any conversation history.
- The handoff must be readable by a human.
- Every UI is just a thin shell over
core. - Token diet is not a side feature — it's the core feature.
Demo
GIF / asciinema recording placeholder — coming with v0.4.
A 30-second clip will show a Codex run hitting a quota wall, relay-baton emitting a compact handoff, and Claude Code resuming the task from the exact same repo state.
Requirements
| Item | Version / Note |
|---|---|
| Node.js | ≥ 20 |
| pnpm | ≥ 9 |
| git | required |
codex | primary agent — ChatGPT Plus or higher subscription required |
claude | fallback agent — Claude Pro or higher subscription required |
Subscription notice
relay-baton never calls the OpenAI / Anthropic API directly. It uses the subscription auth of the local
codex/claudeCLIs.
- Codex CLI — sign in with ChatGPT Plus / Pro / Team / Enterprise.
- Claude Code CLI — sign in with Anthropic Claude Pro / Max / Team / Enterprise.
- Free-tier accounts will not work reliably for either CLI.
- API-key auth is technically possible but blocked by default behind
--allow-api-key-env, to prevent accidental usage-based billing.
Default agent invocations:
- Codex:
codex exec --sandbox workspace-write "<task>" - Claude Code:
claude --permission-mode acceptEdits -p "<prompt>"
--full-auto, --ask-for-approval, and bypassPermissions are intentionally not used (deprecated / unsupported / unsafe). Power users can override agents.<id>.args in relay-baton.config.json.
Logging in
relay-baton login launches each CLI's native auth flow.
relay-baton login # both
relay-baton login codex # codex only
relay-baton login claude # claude only
Under the hood:
- Codex → runs
codex login. Complete browser auth and you return automatically. - Claude → opens an interactive
claudesession. Type/login, complete browser auth, then/exit.
claude --version succeeding does not mean you're logged in. If you see "Not logged in", re-run the command above.
relay-baton does not store, print, or log credentials during the auth process. All credentials are managed by each CLI itself.
Limitations (MVP)
- No LLM / tokenizer calls of its own — runs on character budgets.
- No semantic summarization — deterministic compaction only.
- Agent adapters: Codex / Claude only.
- TUI does not run agents; v0.2+ allows no-run handoff generation only.
Future direction
relay-baton starts as a two-agent fallback harness. The same primitive scales further:
- Multi-agent relay chains. Codex → Claude → OpenCode → back to Codex, each leg carrying a compact handoff.
- Branching session trees. Fork a task into parallel agent attempts; reconcile via diffs.
- Remote relay state. Push
.ai-session/to a shared remote so the next machine picks up where the last one stopped. - Orchestrated workflows.
review,diagnose,continuemodes — bounded autopilot loops with explicit checkpoints. - More adapters. OpenCode, Gemini CLI, Aider, anything else with a sane local subprocess interface.
The harness stays the same shape: detect, capture, compact, hand off.
Roadmap
The full roadmap lives in docs/ROADMAP.md:
- v0.6 — Trust & Verify:
verify,doctor --deep, TUI mode panel. - v0.7 — Review & Diagnose:
review, plan execution receipts, plan diffing,--jsonoutputs, conversation event schema. - v0.8 — Adapter Expansion + Agent Room (first cut): OpenCode / Gemini / Aider adapter scaffolds, project-level fallback overrides, OS CI matrix, and
relay-baton chat/room(turn-based, confirmation-first REPL). - v0.9 — Automation & Runtime (current): bounded multi-turn loop (
LoopController),/continue --max-steps N,/replan, sessionreplay, adaptive per-agent compression thresholds. - v1.0 — Stable Local Release: frozen schema, full command reference, i18n parity.
Future: Agent Room / Conversation Mode
A planned direction (first cut in v0.8) to turn relay-baton into an interactive
multi-agent workspace: talk to Claude and Codex by turns in one session,
with per-agent messages clearly distinguished — Claude as planner/reviewer,
Codex as executor. It keeps every existing safety rail (subprocess-only, no API
keys, no auto commit/push/PR, confirmation-first with a prompt preview before
any real run). Design only for now — not implemented yet. Full design:
docs/AGENT_ROOM.md.
Documentation
docs/COMMANDS.md— full command reference (한국어)docs/ARTIFACTS.md—.ai-session/artifact stability contract (v1.0)docs/AGENT_ROOM.md— Agent Room designdocs/ROADMAP.md— roadmapdocs/RELEASE.md— release & distribution runbook (how the per-OS binaries are built/shipped)docs/HANDOFF.md— session handoff / current state (start here in a new session)
Release notes
Detailed notes live in release-notes/ (index). They are user-facing changelogs, not agent handoff material — that role belongs to .ai-session/handoff.md.
Latest: v1.6.0-alpha.0 — English · 한국어
| Version | English | 한국어 | One-line summary |
|---|---|---|---|
| v1.6.0-alpha.0 | Read → | 읽기 → | Session archives, recovery, and desktop browsing: read-only session list/inspect/resume with manifest + SHA-256 integrity and stale/incomplete diagnosis, plus a desktop session-archive panel and Agent Room /sessions /inspect /resume. |
| v1.5.0-alpha.1 | Read → | 읽기 → | Distribution cleanup: opt-in/manual desktop updater wiring, signed updater artifact/latest.json release support, v1.4 deferred-item closure, and v1.6-to-v2.0 roadmap. |
| v1.5.0-alpha.0 | Read → | 읽기 → | Git tracking first cut: read-only relay-baton git status --json, non-git project fallback, desktop Git panel, and /git in Agent Room. |
| v1.4.0-alpha.1 | Read → | 읽기 → | Distribution polish: one-line CLI installers with SHA-256 verification, release SHA256SUMS/SBOM metadata, package-manager templates, optional signing hooks, and a desktop Codex/Claude preview switcher. |
| v1.3.0-alpha.1 | Read → | 읽기 → | Desktop Agent Room UX revamp: tabbed dashboard/room, full-height chat, handoff modal, slash-command palette with descriptions, decluttered buttons, and a full ? usage guide (en/ko/ja/zh). |
| v1.3.0-alpha.0 | Read → | 읽기 → | Desktop conversation + project-scoped sessions: Agent Room composer, persisted conversation events via conversation append, visible project/session context, and confirmation-first slash actions. |
| v1.2.0-alpha.3 | Read → | 읽기 → | Desktop project management + i18n: add projects with the folder picker, switch/remove projects in the GUI, register non-git project folders, and switch UI chrome between English/Korean/Japanese/Simplified Chinese. |
| v1.2.0-alpha.2 | Read → | 읽기 → | Desktop sidecar fix: the GUI now actually reaches the bundled CLI (withGlobalTauri + window.__TAURI__), so status/budget/handoff/timeline panels populate. |
| v1.2.0-alpha.1 | Read → | 읽기 → | Desktop prerelease completed: fixes the Windows .msi build (non-numeric prerelease) so all 6 assets ship; adds Phase D — signing docs, window-state persistence, light/dark toggle, TUI-mirrored keyboard shortcuts. |
| v1.2.0-alpha.0 | Read → | 읽기 → | Desktop GUI prerelease (Phase A+B): build-desktop release job (.dmg/.msi/.AppImage), deterministic icons, sidecar staging, read-only dashboard, and CLI JSON surface (project list/current --json, handoff show). Unsigned installers. |
| v1.1.3 | Read → | 읽기 → | Distributable (shipped): per-OS standalone executables (Node SEA) attached to the GitHub Release, automated release.yml pipeline, README downloads, Tauri desktop scaffold. (v1.1.0–1.1.2 were release-pipeline hotfixes.) |
| v1.0.0 | Read → | 읽기 → | Stable Local Release: frozen config/session contracts, .ai-session/ artifact validation (doctor --deep), full command reference (docs/COMMANDS.md EN+KO), finalized Agent Room set with read-only /diagnose. |
| v0.9.0 | Read → | 읽기 → | Automation & Runtime (bounded): LoopController, /continue --max-steps N, /replan, relay-baton replay / room /replay, adaptive per-agent compression thresholds. |
| v0.8.0 | Read → | 읽기 → | Adapter Expansion + Agent Room (first cut): OpenCode/Gemini/Aider scaffolds, project-level fallback overrides, OS CI matrix, relay-baton chat/room (turn-based, confirmation-first REPL). |
| v0.7.0 | Read → | 읽기 → | Review & Diagnose: relay-baton review (deterministic diff-vs-plan), execution receipts, plan diffing, --json for status/budget/review, conversation event schema (draft). |
| v0.6.0 | Read → | 읽기 → | Trust & Verify: relay-baton verify (simulated E2E, no model calls), doctor --deep, TUI mode panel, docs/ROADMAP.md. |
| v0.5.0 | Read → | 읽기 → | Plan-execute mode (plan / execute, planner→executor, PlanQualityGate) + context-compression mode (compress-context). |
| v0.4.0 | Read → | 읽기 → | GitHub Actions CI, critical-path tests, CLI smoke test, session observability, handoff history. |
| v0.3.0 | Read → | 읽기 → | Side-effect-free ProjectResolver, registry recovery with backup on corrupt projects.json, RELAY_BATON_PROJECTS_FILE override, lastError cleanup on fallback. |
| v0.2.0 | Read → | 읽기 → | Multi-project registry, --project / --path, project CLI commands, improved TUI dashboard, .gitattributes. |
| v0.1.0 | Read → | 읽기 → | Initial MVP for Codex-to-Claude handoff with token diet, fallback detection, quality gates, and auth-safe subprocess execution. |
Smoke test
After any change:
pnpm build
pnpm test
pnpm relay-baton doctor
pnpm relay-baton project add /path/to/relay-baton --name relay-baton --diet caveman
pnpm relay-baton status --project relay-baton
pnpm relay-baton handoff --to claude --no-run --project relay-baton --diet caveman
pnpm relay-baton budget
Contributing
Issues and PRs welcome. For larger changes, open a short RFC issue first.
git clone https://github.com/<your-org>/relay-baton
cd relay-baton
pnpm install
pnpm build
pnpm test
License
MIT. See LICENSE.
Translations · 한국어 · 日本語 · 简体中文 · 繁體中文 · Español · Français · Deutsch · Português · Русский