From any MCP client, launch Claude Code, Codex CLI, Grok CLI, Cursor Agent CLI, or Antigravity CLI through one harness API inside a persistent interactive TUI.

Aiterm

(日本語: README.ja.md)

Let your AI orchestrate other AIs. One agent_launch call selects the execution harness separately from its model and hands you a persistent session to drive. Cursor can run GPT, Claude, or Grok while Cursor still owns the session, hooks, and transcript.

What it is: one persistent MCP terminal your AI drives — and can launch other coding agents into. ssh, docker exec, a REPL, or another agent's TUI all nest inside that one terminal as just text you send in. The mechanism is deliberately plain — your MCP client drives the other agent's terminal turn by turn: no hidden protocol, no separate aiterm-owned shared-memory layer, no autonomous negotiation. Launched agents still read the normal project and harness memory/configuration that a direct CLI launch would use.

No human at a terminal required. aiterm is driven programmatically over MCP, so an AI can launch and drive another agent with no one sitting in the terminal — from an orchestration loop, a CI step, or a cron job.

MCP = Model Context Protocol — the open standard that lets tools like Claude Code plug capabilities into an AI.

Built and maintained by Quo at kitepon.dev.

Install in your MCP client

検出したClaude Code・Codex・Grok・Cursor・Antigravityのユーザー設定へ登録する標準入口:

npm install -g aiterm-mcp@latest
aiterm-setup --json

aiterm-setupは端末の依存準備、MCP経由の端末実行、登録と読戻しまでを一回で行う。 WindowsはwingetでPowerShell 7・Git for Windows・psmux、macOSはHomebrewでtmux、 Ubuntu／Debianはsudoとaptでtmuxを準備する。必要な公式package managerと実行権限は事前に必要。 他のLinuxでも既存tmuxを利用できるが、自動導入はunsupportedで停止する。 既存設定の他サーバーを保持し、JSON設定は変更前の.aiterm-backupを残す。 結果のstatusはready／unsupported／failed／restart_required。未検出のAIはnot_detectedとし、全AI未検出は成功にしない。 登録先はglobal packageのNodeとMCP入口の絶対パスで、npm一時cacheやsource checkoutは登録しない。 global packageは、npmの現在のglobal rootか、実行中のNodeの既定のglobal rootにあるものを指す。npmのprefixを利用者ごとの場所へ向けた環境でも、共通の場所へ導入したAitermを登録できる。 CodexとGrokは、登録が同じなら公式CLIで作り直さず、利用者が足した項目（待ち時間など）を保つ。 HomebrewのNodeは更新後も有効なoptのパスをMCP登録とCodexのhookに使う。旧版の登録でNode更新後に起動できなくなった場合も、更新後のaiterm-setup --jsonで修復できる。 更新後も同じ入口を実行し、MCP clientを再起動する。npm install自体はユーザー設定を変更しない。 公開JSONはschema: "aiterm.setup-result.v1"、全体のstatus、端末のbackend、 AI別のintegrationsと選択機能のcodex_steerを持つ。失敗時はreason_codeを付け、終了コードはreadyなら0、再起動待ちは3、それ以外は2となる。

MCP登録を利用者や他の製品が管理する環境では、aiterm-setup --hooks-onlyでClaude Code・Cursorの親配送hookだけを登録できる。 依存準備、端末の実動作確認、MCP登録、Codex Steerには触れない。登録済みなら設定を書き換えない。 結果はschema: "aiterm.parent-hooks-result.v1"、全体のstatus（ready／unsupported／failed）、AI別のhooks（configured／unchanged／not_detected／failed）を持つ。終了コードはreadyなら0、それ以外は2となる。

CodexへSteerを有効にする（macOS・Windows・Linux）

対話実行のaiterm-setupで「Aiterm単品」と「Steer付き」を選べます。無人導入では明示します。

aiterm-setup --json --codex-steer enable

公式キューと公式hookを使い、実行中の親には同じターンの次の推論へ回答を渡し、終了後は同じ会話を自動再開します。 Codexの起動プログラムと通常のstdio通信は変更しません。hookの実行ファイルが失われてもCodexの起動・応答は継続します。 終了後の再開は公式キューの監視周期に従い、約10秒かかる場合があります。 Codexのアプリは、見ていない会話を60秒で寝かせます。寝ている会話のキューは、その会話が開かれるまで動きません。 Aitermは、入れた約15秒後にまだ文が残っていて、番が普通に終わって寝ている会話なら、アプリにその会話を開かせて起こします（codex://threads/<会話>）。 その時、アプリの画面はその会話へ切り替わります。番の途中の会話、途中で止められた会話、CLIの席は起こしません。止める時は環境変数AITERM_STEER_CODEX_WAKE=0を立てます。 起こすのはmacOSとWindowsのアプリです。Linuxのアプリが寝かせた会話は、まだ起こしません（実物で確かめられていません。会話が開かれた時に届きます）。

aiterm-setupはCODEX_HOME/hooks.jsonへ専用のPostToolUseとStopを追加し、公式APIでその2件だけを承認・読戻しします。 他のhookや承認は保持します。選択と配送の所有記録は~/.config/aiterm-mcp/codex-parent-hooks/へ保存します。 同じ設定で再実行しても既存hookの順序を変えず、新たな再起動要求を発生させません。 WindowsのhookはPowerShell 7で実行します。更新後のsetupで、既存のAiterm hookコマンドも更新します。 既存の中継は新しいhookの確認後に解除し、保存していたCODEX_CLI_PATHを復元します。macOSの専用LaunchAgentも解除します。 移行前から動いているCodexがあればrestart_required（終了コード3）を返します。完全終了・再起動後に aiterm-setup --codex-steer statusでreadyを確認してください。旧設定は移行を実行するまで維持します。

hookはAiterm自身の配送記録と本文が一致する回答だけを取り出し、利用者がキューに入れた入力は保持します。 取り出し中断や出力失敗はparent_deliveriesにunknownとCODEX_HOOK_DELIVERY_UNCONFIRMEDで現れ、本文を保存します。 自動再送はしません。長い回答はCodexの公式hook処理で抜粋と全文ファイルへの参照になる場合があります。

解除・hook未対応の旧版への巻き戻し前はaiterm-setup --codex-steer disableを実行してCodexを再起動してください。 公式Codex Desktop（macOS・Windows・Linux）の同梱CLIを先に使い、Desktopが無い端末では通常のCodex CLI（公式キュー・hookに対応する0.154以上）を使います。 When a Desktop update moves the bundled Codex CLI, Aiterm finds it again at use time and updates its configuration. If it cannot, it returns CODEX_DESKTOP_BINARY_MOVED; start Desktop and rerun setup. Aiterm単品の公式キュー配送は従来どおり利用できます。

No clone or build is required. Each client launches the published package with:

npx -y aiterm-mcp

Requires Node.js ≥ 18 and a supported multiplexer backend: tmux on POSIX or psmux 3.3.8+ on native Windows. Driving Codex also requires the Codex CLI to be installed and authenticated.

Claude Code

Add it for your user account:

claude mcp add --scope user --transport stdio aiterm -- npx -y aiterm-mcp

Or commit this as a project-scoped .mcp.json:

{
  "mcpServers": {
    "aiterm": {
      "command": "npx",
      "args": ["-y", "aiterm-mcp"]
    }
  }
}

Claude Desktop

Add this server to claude_desktop_config.json:

{
  "mcpServers": {
    "aiterm": {
      "command": "npx",
      "args": ["-y", "aiterm-mcp"]
    }
  }
}

Cursor

Save this as .cursor/mcp.json for the project, or ~/.cursor/mcp.json globally:

{
  "mcpServers": {
    "aiterm": {
      "command": "npx",
      "args": ["-y", "aiterm-mcp"]
    }
  }
}

Ownership boundary: this repository owns installation, configuration, persistent PTYs, agent sessions, state/schema/migrations, diagnostics, recovery, updates, and releases. It can be cloned and operated on its own using this README and the product docs. dotagents optionally integrates Aiterm into the wider factory—host wiring, cross-product compatibility, and aggregate acceptance—but does not control Aiterm and is not a runtime dependency.

Measured, not claimed: in the recorded 203-test benchmark, a pty_read puts ~7.1× fewer tokens in your context than the raw log — and the pass/fail verdict survives the fold. → When to reach for it vs. the built-in shell

19 tools: seven PTY tools — pty_open / pty_send / pty_read / pty_key / pty_close / pty_list / pty_observe — to open, drive, read, and observe one persistent terminal; one canonical agent launcher, agent_launch, which selects claude-code, codex-cli, grok-cli, or cursor-cli as the execution harness; three deprecated launcher aliases kept for migration; agent_models; agent_configure; agent_goal; agent_auth; agent_approval; claude_turn; claude_approval; and diagnostics. The backend is tmux on POSIX and psmux on native Windows, so sessions survive even if the MCP server or the AI client restarts.

v0.28.0 separates the execution harness from the model. The harness owns the agent loop, authentication, hooks, session, and transcript; model is what that harness runs. Cursor Agent CLI can therefore select GPT, Claude, or Grok without changing the completion contract from Cursor hooks to another harness's. Composer is one of Cursor's models, not a harness and not a Grok model: use harness: "cursor-cli", model: "composer-2.5-fast" (or composer-2.5). The old launcher tools are thin compatibility aliases over the same implementation.

v0.25.2 stabilizes repeated in-place configuration changes, including Grok 4.6. If Grok Build 1.0.3 redraws before its /model success notice can be observed, aiterm confirms the requested model/effort from the persistent footer when that state was absent before the command. Callers do not retry, restart, or round a failure into success; explicit grok-4.6 launch and configuration still pass the live catalog check.

v0.25.0 gives Grok and Composer the same shared launcher controls. Their launchers now pass reasoning_effort, enforce write_scope: "read-only" with --sandbox read-only, and support in-place model/effort changes through agent_configure. Before creating a PTY, aiterm checks an explicit Grok/Composer model—and Composer's default model—against the live grok models catalog. An unavailable model fails visibly instead of letting the harness CLI fall back to another model. Composer has since left the Grok CLI and is now one of Cursor's models.

v0.24.3 forwards explicitly selected launcher environment variables from the current MCP process. Pass variable names in env_vars; aiterm reads their current values at launch and injects only the present ones into that agent. This works even when the persistent multiplexer server predates the MCP process, so a stale backend-server environment cannot erase per-seat identity or workflow variables. It also recognizes Codex v0.147's optional fast token in long-lived model/effort footers, keeping agent_configure available on an idle medium fast · session without redraw, retry, or restart.

v0.24.2 keeps in-place configuration working in long-lived Codex sessions. Once the startup header has scrolled out of the captured pane, aiterm recognizes Codex by its persistent model/effort footer together with the input prompt. An idle session is therefore configured directly; callers do not need to redraw the TUI, retry, or restart the agent.

v0.24.0 adds in-place agent configuration. agent_configure uses each harness's native controls to change the model and/or reasoning effort of a running Codex or Claude session while preserving its PTY, harness session, and conversation context.

v0.23.0 adds a local, cross-harness portable fork. Pass throughline_source_session with a mission in prompt to any launcher, and aiterm asks the locally installed Throughline for that session's read-only handoff context before creating the PTY. The exact returned memory is prepended to the mission without moving or copying the source session's database ownership. If Throughline is missing or returns an invalid/empty result, launch fails visibly with no clean fallback. Omitting the field preserves the ordinary clean launch.

v0.22.0 makes launched agents full project collaborators. All four launchers now use the same normal HOME, working tree, harness home, project/user/local configuration, MCP servers, plugins, skills, permissions, trust, memory, and session history as a direct CLI launch. Aiterm isolates only its own per-launch completion correlation state. Every child is told that it is a sub-agent and receives its parent session, delegation depth, lineage, and delegation_allowed=true; a child may delegate further, while the lineage makes reflexive self-copy loops visible and avoidable. The historical managed_completion receipt field remains for API compatibility and means “completion correlation enabled,” not environment isolation.

v0.21.3 removes Codex Stop hooks from the completion path. Codex completion and final-message attribution now come from the root rollout transcript's durable task_complete.turn_id, observed after the dispatch byte boundary. A broken or stale hook executable can no longer strand aiterm-wait. v0.21.0 added explicit write_scope declarations for external-agent launchers; v0.21.3 also fixes their structured launch receipts so a supplied scope and its enforcement status are retained. v0.20.3 prevents concurrent correlated Claude/Fable sessions from turning one broken login into many competing login flows. Every new Claude launch verifies the harness-owned shared credential store before creating a PTY, while healthy credentials remain reusable across concurrent and repeated sessions. The v0.20 line also distinguishes a non-blocking aiterm-wait --timeout 0 observation (running, exit 5) from a real timed-out wait. The v0.19 line added the correlated Claude approval relay, preserved multiline shell delivery, and extended factory diagnostics on native Windows. As of v0.16/0.17 a parent agent never blocks on aiterm: agent sessionへの送信は非ブロックdispatchであり、Codex／Claude Code親には回答本文を自動配送する。 それ以外の親はreceiptのprocess起動情報でaiterm-waitを実行する。 終了コードは0=done、3=timeout、4=closed、待機しない照会の5=runningを表す。 Factory diagnostics and the local runtime-error store collect only when canonical dotagents config explicitly sets collection.enabled: true; collection is off by default and performs no network I/O. It ships via tag-triggered CI with npm provenance (OIDC Trusted Publishing); the GitHub Release re-registers the Official MCP Registry entry.

Status: actively maintained · current public release v0.63.7 · runs on Linux · WSL2 · macOS · native Windows (tmux on POSIX, the tmux-CLI-compatible psmux on native Windows — no WSL required) · MIT · see the CHANGELOG.

Update and rollback

The npm package is the standalone distribution; dotagents is not involved. For a global install, update with aiterm-update. It reinstalls the requested version into the same npm prefix, then reruns the new version's aiterm-setup --json to re-register and re-verify.

aiterm-update                                   # this machine to latest
aiterm-update --host rabbit --host win-test     # this machine and SSH hosts to the same version
aiterm-update --version 0.39.0 --check          # report current and target versions without changing anything

--host takes an ~/.ssh/config alias or host name; hosts are not stored. The version is resolved once on the calling machine so every host lands on the same version. Hosts older than aiterm-update get it through npm first. When the npm prefix is not writable, the result is permission_required with the command to run as an administrator. aiterm-mcp servers that were already running keep the old code until their MCP client restarts (running_servers); tmux/psmux sessions survive. Versions without aiterm-update use npm install -g aiterm-mcp@latest and aiterm-setup --json. To roll back, install a known-good immutable version, for example npm install -g "aiterm-mcp@<known-good-version>", then restart the MCP client. setupを持つ版では再起動前にaiterm-setup --jsonを再実行する。For an npx configuration, use aiterm-mcp@latest to update or replace it with aiterm-mcp@<version> to pin or roll back. Check the CHANGELOG for state/schema compatibility before downgrading. Maintainer release and artifact rollback are specified in the product-owned release procedure.

Related MCP server: claude-tmux

Why now

A lot of 2026's agent tooling is converging on orchestration: a lead model delegating a mechanical refactor to Codex, running Composer on a bulk edit while it reviews the diff, fanning one task across several agents to spare its own context window. All of those agents already live in a terminal. aiterm makes that terminal a first-class, MCP-native tool — so the model doing the orchestrating can spawn and steer the others without a human wiring up panes.

Built with Codex and GPT-5.6 for OpenAI Build Week 2026

aiterm predates Build Week, so the event work is kept visible in dated commits. During the submission window (July 14–16, 2026), I extended it with safe serialized delivery for long PTY input, correlated operation IDs and bounded result recovery, machine-readable launch and idempotent close receipts, and a hardened readiness gate that prevents prompts from disappearing during TUI startup redraws. The public comparison from the pre-event release is v0.12.2...main.

I used Codex with GPT-5.6 as an engineering collaborator: it inspected the implementation, challenged the API and recovery contracts, generated focused regression cases, and helped verify race, security, timeout, and malformed-event paths. I reviewed the diffs and test evidence and retained the final product and architecture decisions. At that Build Week checkpoint, the regression suite contained 262 tests covering normal operation as well as failure and recovery behavior; current release receipts live in the CHANGELOG and release ADRs.

ClaudeがAPIエラーや安全判定の拒否で終了した時は、Stop hookが発火しなくても次のpty_sendを新しいturnとして扱う。現在のturn開始後のエラー記録だけを確認し、過去のエラーで実行中のturnを解除しない。上流の拒否はエラーのまま返す。

Codexがサービスの誤りや通信の失敗でturnを打ち切った時は、完了待ちがdoneではなくerrorを返す（aiterm-waitはexit 7、親への配送はoutcome=error）。errorは応答の本文とURLを落とした1行の文で、Codexが記録した種類（internal_server_error・http_connection_failedなど）はerror_kindに載る。利用上限は今までどおりrate_limited。誤りで終わった後の席は、次のpty_sendを新しいturnとして受ける。 Cursorが誤りでturnを終えた時（記録のturn_endedがstatus: "error"）も同じで、完了待ちはerrorを返し、文の頭の角括弧の名前（unavailableなど）をerror_kindに載せる。利用上限の文はrate_limited、席で止めたturn（User aborted request）は今までどおりdone。

Two ways to use it

Failed Codex and Claude turns may add failure to the existing wait result. Its category is capacity, rate_limited, transient or other; execution reports observed work as started or leaves it unknown. A provider rejection never proves that earlier tools did not run. terminal refers only to the observed turn; future_effects:"unknown" preserves uncertainty about queued input, retries and other producers. No input is automatically replayed by Aiterm.

Claude dispatch and prompt-launch receipts include input_id. Keep the receipt's wait_process, or pass that ID to aiterm-wait --input <id> --cursor <original cursor> with the original session and launch. This prevents a later anonymous turn at the same event cursor from becoming the old request's result. Its saved transcript boundary also catches an API error written before the waiter starts. The failure boundary identifies the Aiterm input observation, not native input acceptance. Missing or legacy evidence is not upgraded. Parent receipts and notices preserve the classification. A caller can continue unfinished work from the retained conversation or saved handoff after applying its normal stop and configuration conditions; replaying the original message is a separate decision.

Grok can finish its current turn while pty_send is submitting input. Aiterm correlates the newly appended full user message with the native turn number and returns agent_dispatch for a proven new turn, without sending again. STEER_NOT_QUEUED and STEER_STILL_QUEUED preserve isError:true and their existing text prefix; structuredContent adds aiterm.agent-send-error.v1, a string code, outcome:"outcome_unknown", effect:"possible", and the original session/launch. A Grok observation handle is returned only when its pre-input record was saved; unavailable handles are null. Do not resend unknown input. Its waiter remains pending until the input's native acceptance is proven and cannot use the preceding turn's completion. Cursor errors retain null recovery handles until native acceptance can be proven.

1. Drive SSH, containers, and REPLs in one persistent terminal — the primitive

This is the base, and it works with just the platform backend — tmux on POSIX or psmux on native Windows. pty_open grabs one local terminal; ssh host, docker exec -it x bash, or a REPL are just text you pty_send into it — once. Every command after that rides the same already-authenticated session. Session kind is never a tool-level distinction.

pty_open()                         → grab one local terminal
pty_send(id, "ssh 192.168.1.2")    → authenticate once, inside that terminal
pty_send(id, "uname -a")           → every later command rides the SAME session
pty_read(id, { wait: true })       → read the token-reduced output, completion detected

Origin. I built aiterm for exactly this. Driving my homelab from Claude Code one command at a time meant every SSH command became its own connect → authenticate → disconnect: re-typing the passphrase and one-time code each time, short-lived sessions piling up, and eventually my own defenses (fail2ban, MaxStartups/MaxSessions, account lockout) locking me out — the security meant to stop attackers ended up stopping me. Holding one authenticated session fixes all three at once. That pain is why the persistent terminal exists; launching whole other agents inside it is what it grew into.

2. Launch other coding agents into that terminal — the orchestration flagship

The same primitive hosts another agent's TUI. agent_launch starts a selected execution harness inside a fresh persistent terminal and returns a session_id. harness names the component that owns the agent loop, authentication, hooks, session, and transcript; model remains an independent choice. The launched process sees the same project and user environment as a direct CLI invocation: normal configuration, MCPs, plugins, skills, permissions, trust decisions, memory, and history are not copied, filtered, or replaced. Aiterm adds only completion correlation and a non-user sub-agent context containing role=subagent, the parent session, delegation depth, lineage, and delegation_allowed=true.

起動結果には正規harnessを含むaiterm.agent-launch-result.v1が付き、旧providerは互換fieldとして残る。同じharnessはagent dispatch、aiterm-wait、agent_configure、pty_listにも載る。Codexは通常rollout、Grokは通常session event、Claudeはlaunch固有Stop hook、Cursorは通常agent transcriptのturn_endedを完了正本に使う。agentへの送信はpty_sendだけで行い、Aitermが送る時点で子の状態を見て振り分ける。Claudeは画面の実行中表示ではなくStopまで残るturnの印で判定する。実行中のturnへは差し込み（mode=agent_steer、新しいevent_cursorと配送は作らない）、それ以外は非ブロックdispatch（mode=agent_dispatch）で、harnessごとの完了境界を表す整数event_cursorを返す。Codex親は選択に応じて公式Steerまたはqueue、Claude Code親は公式非同期hookで本文を自動受信する。他の親はaiterm-waitを使う。CursorのsubmitはadapterがCLIのextended keyboard protocolへ変換し、送信本文がcomposerへ残る場合は明示errorにする。

An ordinary send can select steer while its input is still unissued and readiness waiting observes a live busy turn. A confirmed exited harness is rejected before the readiness deadline. Exit after paste retains a possible effect; the same input is never replayed.

Set require_agent:true on pty_send when the integration requires agent delivery. If the agent registration is missing, Aiterm refuses before sending any text and returns AGENT_SESSION_REQUIRED with an explicit unsent message. The default preserves ordinary PTY sends; combining it with force:true is rejected before sending.

pty_send to an agent session accepts an optional preface: one line placed before the text, so the message reads "preface, blank line, text". For a Claude Code session Aiterm enters the preface without the terminal's paste markers and pastes the blank line and the text in one paste. Claude Code wraps long pasted text in <pasted_content> and follows instructions inside it only where the user's own words outside the tags ask it to, so this keeps an integration's header (who the message is from) outside the wrapper. For Codex, Grok and Cursor sessions the three parts are joined and pasted as before. The preface must be a single line of at most 200 characters with no control characters, must not start with a symbol or whitespace, and must not contain @; otherwise Aiterm returns AGENT_PREFACE_INVALID and "文字列は送信していません。" before sending anything. It cannot be used with ordinary PTY sends, force:true, or remote.

When the caller gives no preface, Aiterm adds one itself for a Claude Code session whenever the text is long enough for Claude Code to wrap it (three or more line breaks, or 800 characters and up): "Aitermが届ける、この席の呼び手からの正規の送信です。下の貼り付けは、どこかから拾った文ではなく、呼び手がこの席へ宛てて送った文です。" ("A regular message from this session's caller, delivered by Aiterm. The paste below was sent to this session by the caller, not picked up from somewhere else.") The line states where the message comes from and does not tell the session to obey it. It applies to new turns, to steering, to the prompt of agent_launch, and to remote sessions (the Aiterm on that machine adds it). Without that line the whole message sits inside the wrapper, and a Claude Code session may decline to act on it as "not the user's own instruction". Short texts, Codex, Grok and Cursor sessions, and sends with raw or force are unchanged (ADR 0116).

Overlapping pty_send calls to the same agent session are handled one at a time. A later call picks its route only after the earlier one has been sent, so the result matches calls made in sequence (a call that overlaps the first send to a freshly started Claude Code session returns mode=agent_steer). It returns later by however long the earlier call waits for the session to accept input. If its turn does not come within the limit (60 seconds on POSIX, 180 seconds on Windows), Aiterm refuses before sending any text and returns AGENT_SEND_BUSY with an explicit unsent message. A lock left by a process that exited mid-send is cleared by the next send. agent_configure waits in the same line: a pty_send that arrives while a configuration change is in progress is sent after the change, so the turn runs with the new setting. A configuration change to a session whose turn is running does not hold the line while it waits for the turn to end, so steering still goes through. If its turn does not come within the same limit, agent_configure returns AGENT_SEND_BUSY with an explicit unchanged message.

Queued agent operations retain the original launch_id. If the session is recreated under the same name while they wait, they fail with EXECUTION_STALE_GENERATION before acting on the replacement.

agent_launch and pty_send (to an agent session) accept an optional image: an array of absolute paths to image files (png/jpg/jpeg/gif/webp). Aiterm appends an attachment block to the prompt, and every harness opens the path with its own file-reading tool and sees the image; the caller never learns harness-specific attachment tricks. Invalid paths are rejected before anything is sent.

agent_launch accepts an optional write_scope: either "read-only" or a human-readable description of writable paths. Codex/Grok use --sandbox read-only; Cursor uses its official read-only --mode ask. A path description remains declaration-only because these CLI launch surfaces provide no equivalent path allowlist flag.

Grokの無人起動は公式--trustで指定された作業フォルダを信頼登録し、確認画面を完了してから初回promptを送る。この登録はGrok CLIの信頼ストアへ保存され、フォルダ内のhook・MCP・LSPにも適用される。read-only sandboxの制限は維持する。画面に残る完了済みhookの結果は実行中と判定しない。

Grokで終了済みターンのweekly-limitパネルが残っている場合、次の通常pty_sendがShift+Xで一度閉じ、入力受付を確認して今回の本文を送る。同じsessionと会話を保ち、receiptのpane_input_recoveryにgrok_rate_limit_dialog_dismissedを記録する。ターン未終了・harness不在はGROK_RATE_LIMIT_RECOVERY_BLOCKED、解除後の入力受付失敗はGROK_RATE_LIMIT_RECOVERY_FAILEDとなり、本文は未送信。上限の継続はrate_limitedとして返し、過去promptは再送しない。Grokの上限観測には現在の画面だけを使う。Claude Codeの上限は現在の画面の入力欄の下に出る知らせで、Codexの上限はturnを終えた記録（task_completeのcodex_error_info: "usage_limit_exceeded"）で見分ける。pane logや道具の出力に残る上限の文字では判定しない。

Grokの待機中にerror eventが先に現れた場合は、見回り一回分後に現在の上限panelを読み直す。上限が無ければ元のerrorを返し、次turnや消失したlaunchへ古いerrorを結び付けない。timeout=0の即時照会は待たず、入力やAPI要求は再送しない。

When a Cursor pre-submit hook (beforeSubmitPrompt, or a Claude Code UserPromptSubmit hook that Cursor loads for compatibility) rejects the prompt, Cursor drops it and no turn or completion follows. Aiterm recognizes the rejection: an initial prompt returns initial_prompt=failed, and pty_send returns an error instead of a success receipt, both with USER_HOOK_BLOCKED and the hook's output. A rejection that comes after the 3-second start check is reported by the completion wait as outcome=error (aiterm-wait exit 7).

A current Cursor Error: containing a complete JSON object with error.code equal to prepaid_balance_exhausted ends the wait as outcome=error with that error_kind, even without a new user record. Quoted diagnostic codes, malformed JSON and old errors followed by a composer or busy display do not establish this terminal. Aiterm does not resend or close the session.

Grokがread-only sandboxの適用を拒否した場合、prompt送信時にGROK_SANDBOX_STARTUP_FAILEDとCLIの原因を返す。hookパスのシンボリックリンクなど、CLIが示した原因を設定の管理元で修正し、対象sessionをpty_closeして起動し直す。Aitermはsandboxを解除したりhookをコピーしたりしない。

この判定はGrok専用アダプターが所有する。初回prompt付きのagent_launchと通常のpty_sendで、入力受付待ち中に拒否を検出すると未送信のエラーを返す。promptなし・trust_project指定なしの起動応答は入力受付を保証しない。trust_project:trueでは入力受付まで確認し、startup.statusを返す。Grokのprivacy notice起動設定も同アダプターが所有する。実装の責務分担はDESIGNを参照。

Codex 0.155.1の「Approaching rate limits」model切替dialogは、通常のpty_sendとagent_configureで同じsessionのまま一時的な2. Keep current modelだけを選ぶ。入力受付を再確認してから本文または設定変更を進め、dispatch receiptのpane_input_recoveryにはcodex_rate_limit_model_switch_kept_currentを記録する。model切替と今後の表示抑止は選ばない。入力受付へ戻らなければCODEX_RATE_LIMIT_MODEL_SWITCH_RECOVERY_FAILEDとなり、本文・設定変更は未送信。このdialogはagent_approvalの対象ではなく、inspectはreason="rate_limit_model_switch"だけを返し、prompt digestとchoicesを出さない。

For a correlated Claude turn stopped at Do you want to proceed?, use claude_approval(action: "inspect", ...) to capture the active operation and SHA-256 screen digest, review the displayed command, then call respond with that exact digest and either approve_once or deny. The relay rechecks the operation and screen under the send lock, never exposes arbitrary input or permanent approval, keeps the active marker intact, and records a prompt-free owner-only receipt. pty_send(force: true) does not bypass this boundary.

agent_launch({ harness: "codex-cli", session_name: "codex1", cwd: "/repo",
              prompt: "port test/legacy.py to vitest",
              model: "gpt-5.6-sol", reasoning_effort: "high",
              write_scope: "test/ only; no commit" })
                                    → { session_id: "codex1", … }   # Codex now live in a persistent terminal
pty_read("codex1", { screen: true })   → read what it's doing (token-reduced)
pty_send("codex1", "also fix the imports it broke")
                                    → non-blocking dispatch; receipt carries event_cursor
# Codex／Claude Code親には回答が自動で届く。それ以外の親:
$ aiterm-wait --session codex1 --cursor <event_cursor>   # never in the parent's foreground; exit 0=done, 3=timeout (not done), 4=closed, 7=error (turn aborted by an API error)
pty_read("codex1", { agent_transcript: true })           → collect the full answer

The canonical harness choices are:

Antigravity CLI 1.4.10 uses its normal user authentication and project configuration. Install and sign in with the official agy CLI before launching. Model IDs include their effort; pass the exact catalog ID. Configuration changes keep the same conversation and restore the user default model. Its global MCP file is ~/.gemini/config/mcp_config.json; Aiterm adds its named completion hooks in ~/.gemini/config/hooks.json. A session-only queuedMessages: "send-immediately" override gives running-turn sends the native immediate-send behavior. An Antigravity parent uses the returned wait_process and pty_read(agent_transcript:true). Native /boost is a separate multi-agent workflow, not the fast speed switch. agent_goal currently returns native_goal_adapter_not_implemented; this does not mean the CLI lacks /goal. Forced relogin is refused without issuing input because official /logout purges shared credentials and cache.

A new terminal gets the environment of the MCP process that opened it (since 0.50.0, ADR 0081). Values from another caller that happened to start the tmux server first no longer appear in it. AITERM_SESSION_ID and AITERM_AGENT_* are not inherited; aiterm sets them per terminal. If a harness passes only a few variables to its MCP process, the terminal has only those (Codex passes HOME, PATH, SHELL, and TERM by default; forward more with env_vars under [mcp_servers.aiterm]).

env_vars is an allowlist of environment-variable names, not a name/value map. At launch, aiterm reads each valid name from its current MCP process, shell-quotes present values, and places them on that one harness launch command. Missing names are omitted; invalid shell variable names fail before session creation. Only the named values can be read back with env_keys in pty_list. Values do not enter the MCP tool arguments, but they are delivered through the PTY launch command and retained in aiterm's per-session .lastcmd; the launched harness and other processes with access to the same OS user may read them. Use this for non-secret seat identity and workflow variables, not as a secret transport.

The selected harness CLI must be installed and authenticated. Aiterm resolves CLAUDE_BIN / CODEX_BIN / GROK_BIN / CURSOR_AGENT_BIN, then the documented default binary, then PATH. Cursor resolution deliberately uses cursor-agent, never the ambiguous agent name. Claude and Cursor authentication are checked before a PTY exists, so a failed preflight leaves no session. All harnesses use their normal harness-owned credential and configuration stores in place. For Grok, Aiterm does not lock, inspect, or modify the credential. A non-empty inherited GROK_AUTH_PATH must be absolute and exist; Aiterm passes it unchanged to Grok. Grok owns its contents, permissions, and link handling. Absence of the default auth file is accepted only when XAI_API_KEY is set.

Portable fork is optional. When throughline_source_session is present, prompt is the required new mission and launch_operation_id cannot be combined with it. aiterm resolves Throughline via THROUGHLINE_BIN and then PATH, runs throughline handoff-context --session <id> --json, and places its returned context before a fixed separator and the mission. This route requires throughline >= 0.9.0; throughline_supplement_file requires Throughline 0.10.8 or later. Aiterm appends --supplement-file <path> without reading or interpreting the file. Throughline owns its project binding, validation, and shared context budget. The route reads source memory without changing database session ownership. No Throughline dependency is needed when the field is omitted.

Harness adapters translate model and reasoning_effort into each CLI's public controls. Explicit Grok models are checked against grok models; Cursor combines a base model such as gpt-5.6-luna with a separate effort such as high, checks the resulting current catalog ID, and uses Cursor's standard model picker for in-session changes. Missing models are errors, with no cache, retry, or fallback. Claude adds only launch-local Stop-hook settings, Codex reads its normal rollout store, Grok reads its normal session event/history, and Cursor binds its normal agent transcript with the launch ID. Pass an absolute cwd; ~ is not expanded.

There is no hidden protocol between agents: every launched harness is another user-visible persistent terminal session. The MCP client drives that TUI with ordinary PTY operations, and a human can attach to watch or take over.

Demo

Real captured output — each block below was just run through aiterm in this repo; the numbers, the elision marker, and every is_complete verdict are the tool's own, not mocked. The bracketed meta line is what pty_read appends; its labels are Japanese in the actual output, translated here for readability (the Japanese README shows them verbatim).

A long output folded head+tail — the middle is elided by the reducer, not by me (166 → 56 tokens):

→ pty_send("demo", "seq 1 150")
→ pty_read("demo", { wait: true })
← 1
  2
  3
  ⋮  (head runs to line 29 — abbreviated in this README)
  … ⟨102 lines elided · full=true, or line_range="A:B"⟩ …    ← the tool's own marker
  ⋮  (tail resumes at line 132 — abbreviated in this README)
  149
  150
  [aiterm demo: 51 lines / ~56 tok (raw 152 lines / ~166 tok); 102 lines hidden] [is_complete=True via quiescent]

A grep comes back exactly as grep printed it while it stays within rtk's caps (200 lines in all, 25 per file). Past a cap, the per-command reducer groups the hits by file and says what it left out:

→ pty_send("demo", "grep -rn session src/")
→ pty_read("demo", { wait: true, rtk: true })
← 550 matches in 23 files:

  src/agent-resolver.ts:211:// …(long lines keep ~80 chars around the pattern)
  …
  src/remote.ts:268:session_id: session, agent_transcript: true, raw: true,
    +4 more in src/remote.ts
  +7 more files
  [aiterm demo: rtk:grep applied / ~4286 tok (raw ~12546 tok)] [is_complete=True via quiescent]

Nesting is just text you send in — here a Python REPL inside the same PTY (an ssh host, a docker exec -it … bash, or a launched coding-agent TUI nests exactly the same way):

→ pty_send("demo", "python3")
→ pty_read("demo", { until: ">>>" })                # nested prompt = "the inner shell is ready"
→ pty_send("demo", "print(sum(range(1_000_000)))")
→ pty_read("demo", { wait: true, until: ">>>" })
← 499999500000                                      [is_complete=True via until]

The only edits to the captures above are the two ⋮ lines (a long head/tail run abbreviated for the README) and one over-long grep line truncated to fit — the ⟨…⟩ marker, the token counts, and every is_complete verdict are exactly what the tool printed. (Use until: ">>>" without a trailing space — the captured prompt is trimmed, so ">>> " would miss and fall through to timeout.) While nested, pass until (the inner prompt) or mark: true, because quiescence cannot fire there by design — see Completion detection and Known constraints. A human can attach to the same multiplexer backend and watch any of this live (see A human can watch).

First run (≈60 seconds)

aiterm-setup --jsonがreadyになったら、利用するMCP clientを再起動して接続を確認する。Claude Codeの場合:

/mcp        # aiterm should show as connected, exposing 19 tools

Your first session — four calls, one persistent terminal:

pty_open()                          → { session_id: "t1", attach: "<platform attach command>" }
pty_send("t1", "echo hello")        → command sent into the PTY
pty_read("t1", { wait: true })      → "hello"   (token-reduced, completion detected)
pty_close("t1")                     → terminal released

pty_close is idempotent and returns a structured closed / already_closed receipt, so durable callers can retry the same session_id after losing the MCP response. On Windows, closing a Claude Code session first asks Claude Code to exit on its own (Ctrl-C twice, then a wait of up to 4 seconds) so that its SessionEnd hooks run; psmux stops a session without the hangup signal that tmux sends on POSIX. The session is stopped afterwards either way.

That's it. The terminal in t1 is real and persistent — ssh, docker exec, a REPL, or a launched agent's TUI are just things that live inside it. To launch a worker agent instead, one call does it: agent_launch({ harness: "codex-cli" }) returns a session_id you drive with the same pty_read / pty_send.

Prefer a global install, or a different client?

# install globally, then register the command name
npm i -g aiterm-mcp
claude mcp add --scope user --transport stdio aiterm -- aiterm-mcp

This registers it in ~/.claude.json; you'll get an approval prompt the first time. For client-specific JSON, see Install in your MCP client.

Headless: no human at the terminal

Because an MCP client drives aiterm programmatically over stdio, everything above can run with nobody sitting at the terminal. Any MCP-capable orchestrator can call agent_launch — including a harness matching itself — then pty_read the result and act on it unattended. That makes aiterm a fit for exactly the places a human-driven terminal isn't:

• Multi-agent orchestration — an orchestrator hands sub-tasks to Claude Code / Codex / Grok / Cursor harnesses, each in its own persistent session, and reads them all back. Composer is one of the models Cursor selects (composer-2.5-fast).

• CI — a job step can spin up an agent, drive it, and tear it down.

• cron — a scheduled run can launch an agent and collect its output.

The terminal is real and shared, so a human can jump in (A human can watch) — but nothing requires one to.

How it works

flowchart LR
    AI["AI / MCP client<br/>(the orchestrator)"] -->|"pty_send · pty_observe · agent_launch · agent_models · agent_configure · agent_auth · agent_approval · claude_turn · claude_approval<br/>legacy launcher aliases · diagnostics"| S["aiterm-mcp<br/>stdio MCP · 19 tools"]
    S -->|"pty_read<br/>token-reduced"| AI
    S -->|"tmux / psmux<br/>send · capture"| P["persistent PTYs<br/>survive restarts"]
    P -->|"ssh · docker · repl"| R["nested<br/>remote · container · REPL"]
    P -->|"launches a fresh PTY per agent"| A["another coding-agent harness<br/>Claude Code · Codex CLI · Grok CLI · Cursor CLI"]

One PTY is the only primitive. Everything else — SSH, containers, REPLs, and the launched agent TUIs — is just something interactive running inside a persistent terminal, driven with the same pty_send / pty_read. Each launcher opens its own fresh PTY. Because the PTYs live in tmux on POSIX or psmux on native Windows, sessions outlive the MCP server and the AI client.

When to reach for it vs. the built-in shell

Your MCP client already has a shell tool, and it wins on some jobs. aiterm wins on others. We measured both on the same commands in this repo, counting tokens the same way on each side (characters ÷ 4, aiterm's own estimator), so the comparison is apples-to-apples.

Start with the built-in tool for a light one-shot. git log --oneline -5 is one round-trip; aiterm is two — pty_send then pty_read — and that second round-trip costs more than a light command saves (~7 s vs ~13 s).

The second round-trip pays for itself once the output runs long, or the state has to outlive the call.

In the recorded 203-test benchmark the reduction is real and safe. The built-in tool drops the whole 223-line log — ~4,292 tokens — into context. aiterm folds its own capture of the run down to ~607:

[aiterm demo: 51 行 / ~607 tok (raw 223 行 / ~4292 tok); 172 行 hidden] [is_complete=True via mark]

行 = lines; the meta line is quoted verbatim from aiterm's real output.

That is about 7.1× fewer tokens reaching the model, and the verdict survives the fold: the tail still carries ℹ tests 203 / ℹ pass 203 / ℹ fail 0. The reduction drops the noise and keeps the line you opened the log for. Wall-clock effectively ties, so on a run this long the extra round-trip is a small part of the total.

aiterm also holds state across calls. The built-in tool runs each call in a fresh shell, so cwd resets between calls and the environment doesn't carry. Send cd /tmp && export BENCH_VAR=hello123, then read it back in a second, separate call:

built-in shell  →  var=                   # empty; env dropped, cwd back at project root
aiterm          →  cwd=/tmp var=hello123  # one persistent PTY holds both

cd then set env then build, ssh once then run ten commands on the authenticated session, drive a live REPL or a launched agent's TUI turn by turn — one persistent PTY holds all of it. Reach for aiterm when the terminal has to remember something.

¹ Today's harness auto-offloads the ~192 KB dump to a file and previews only a ~2 KB head, so the token counts nearly tie; aiterm reports the accurate line count and lets line_range="A:B" pull any slice later, head or tail. ² The rtk grep reducer returns grep's output unchanged while it fits rtk's caps (200 lines, 25 per file). Past a cap it groups by file, keeps ~80 chars of each long line around the pattern, and folds the rest into +N more in <file> / +N more files; use the built-in tool when you need every full line of a large search.

vs. the alternatives

aiterm sits at the intersection of two families: terminal-driving MCP servers, and the newer "agents talk to each other through a shared terminal" idea (see Where aiterm fits). Here's how the axes line up — honestly, including where the others are strong.

Where aiterm fits

"AIs talking to each other through a shared terminal" is becoming its own category — and it's a genuinely good idea. The terminal is a universal interface every coding agent already speaks, so no bespoke agent-to-agent protocol is needed; the shell is the shared surface. smux (by @shawn_pana) popularized this framing as a one-command shared tmux environment a human sets up, that agents then join via a tmux-bridge CLI and Agent Skills. It's good at the in-the-loop, shared-pane workflow it's built for, and it has real traction.

aiterm takes the same core insight — the terminal as the meeting point — and makes three deliberate, different choices:

1. Headless by construction. Because aiterm is driven programmatically over MCP, an AI can launch and drive another agent with no human sitting in the tmux — from an orchestration loop, a CI step, or a cron job. The shared-tmux tools lead with a human at the keyboard (their docs center on interactive pane navigation), so unattended operation isn't their native mode; aiterm's is.

2. MCP-native, not a workflow you adopt. aiterm is a stdio MCP server: one claude mcp add line and it works as structured tools in any MCP client that speaks stdio (tested in Claude Code; Cursor, Cline, and Claude Desktop speak the same protocol and should work the same way). It doesn't ask you to adopt a tmux config, learn pane navigation, or install skills into your setup — the client already knows how to call tools.

3. Launching an agent is one tool call — an orchestration primitive. agent_launch({ harness: "codex-cli" }) spawns Codex in a persistent terminal and returns a session you drive immediately. You don't arrange panes or paste between them by hand; the launch, the steering, and the reads are all tool calls the orchestrating model can make on its own.

On top of that sits a productized layer a raw tmux bridge doesn't have: token-reduced reads and 5-layer completion detection. None of this makes the human-in-the-tmux model wrong — it's a different, complementary bet on where the human is standing.

Tools

Session observation and startup

pty_open defaults to bash on POSIX and PowerShell 7 on Windows. Ordinary terminals and agents receive AITERM_SESSION_ID. A terminal inherits the environment of the MCP process that opened it. Pass names in env_vars to register them on the session. pty_list({ env_keys: ["JOB_OWNER"] }) returns only registered non-secret values in environment; unregistered and missing values are null. Its aiterm.pty-list-result.v1 receipt contains observed_at and sessions, whose entries include session_id, current_command, attached, width, height, harness, and environment. Existing text remains available. After the last session closes, session listing may briefly re-observe an exiting server for up to 500 ms. Managed native operations retain unknown unless the listing or exact scoped absence can be confirmed; this never repeats a close or send.

pty_observe({ session_id, cursor? }) returns aiterm.pty-observe-result.v1 with exists, observed_at, state (busy/idle/blocked/dead/missing/unknown/hibernated), reason, pane_alive, and harness_alive. pane_process and harness_process are separate identities. process_identity selects the harness for agents, or the unique child process-group leader for an ordinary terminal, using the pane when no child leader exists. An identity contains pid, process_group_id, started_identity, and argv_digest; unresolved identities are null. Windows PIDs are native and its process-group field is null. Start identity uses POSIX LC_ALL=C ps lstart or Windows UTC ISO milliseconds; the argv digest is SHA-256 hex. New managed Windows launches of Claude, Codex, Grok, and Antigravity keep missing native processes unknown until their launch shell records that the same CLI invocation returned. Pending input uses the existing readiness deadline; it is still checked before paste and Enter. Older launch metadata keeps its previous behavior. Input admission uses fresh native ownership and its dedicated readiness screen; an unrelated activity-capture failure does not block it. Public pty_observe still requires the full activity observation. Native process-read errors include the probe kind, OS error code, child exit status/signal and output byte counts. Raw output, arguments and credentials are excluded; the existing refusal and exit code remain unchanged.

Pass activity.cursor into the next observation to obtain output_changed and cpu_delta_seconds; first observations and recreated panes return null differences. cpu_seconds is the current subtree's cumulative CPU. If a process disappeared between observations, the delta covers only observed increments and cpu_delta_complete is false. background_cpu_seconds, background_cpu_delta_seconds, and background_cpu_delta_complete apply the same measurement only to descendants created at least 60 seconds after the pane, excluding startup MCP processes. token_hint is the latest displayed token count or null. Callers do not need raw argv or pane-text parsing.

A confirmed hibernation keeps the same logical session in pty_list, with the saved harness, conversation ID, and explicitly registered environment values. pty_observe returns state: "hibernated", reason: "idle_timeout", and exists: true, while both liveness fields are false and process identities and token hint are null. Pending child deliveries and post-startup process count are zero; CPU totals are zero and activity cursors and differences are null. Stopping, resuming, failed, or conflicting native generations remain unknown. Closing a hibernated seat removes it without restarting the CLI. An explicit same-name agent_launch resumes the saved conversation and settings; conflicting harness, settings, project, or registered environment names return AGENT_RESUME_CONFLICT before input. pty_open cannot replace a dormant agent seat. Native pty_read, control keys and approval actions refuse dormant or uncertain generations with AGENT_RESUME_UNCONFIRMED; transcript reads remain available through the existing transcript option. The stop button returns idle for a confirmed hibernated seat without waking or closing it, and refuses uncertain or conflicting generations. Initial managed Claude/Codex launches on all supported operating systems also retain the actual terminal creation identity; a missing proof prevents automatic hibernation, and automatic naming preserves logical seat names. Idle shutdown also requires confirmation of the running CLI’s settings; mismatched or unknown settings keep the seat alive. Managed Claude and Codex conversations on Linux, macOS and Windows hibernate after 30 minutes from their last completed turn and resume the same conversation on the next send. Set AITERM_IDLE_MINUTES to a positive finite duration to change the threshold. Busy or blocked turns, attached terminals, pending child deliveries, active descendants, missing native creation evidence and uncertain settings remain alive. Other harnesses, ordinary PTYs and mounted execution graphs are excluded. aiterm-idle-check --session <name> --launch <32hex> runs the same guarded check once.

Codex can start a native queued turn during resume. At standard speed, Aiterm accepts that busy state only with the saved conversation proof, a new task_started after the hibernation cursor and the same live native launch. A busy screen alone is insufficient. This path does not capture user work as startup scaffolding: its post-startup process count remains unknown and automatic hibernation stays disabled for that launch. Fast restoration still requires an idle composer. Failed attempts remain blocked. Resume failures return isError: true and structuredContent with schema aiterm.agent-resume-error.v1, code AGENT_RESUME_UNCONFIRMED or AGENT_RESUME_FAILED, session_id, launch_id, input: "not_sent" and recovery: "inspect_session". The input status describes only this request's body; it does not prove that the CLI was never started or that other native input was not processed. Inspect the seat; do not automatically resend or restart it.

Two fields tell a caller what closing an agent session would lose. activity.post_startup_process_count is the number of processes in the session that did not exist when agent_launch finished startup (before the first prompt), so background work started in the first minute is counted too. Two kinds of process are not counted: Codex's own codex-code-mode-host helper, and the direct children of an mcp-lazy relay (the MCP server it starts on first use and its wake predicate). What runs below them is counted. A startup process that the harness restarted (a process with the same parent and the same arguments as a startup process that has exited, such as a reconnected MCP server) is not counted either, and neither are the processes Aiterm itself starts to read the process table (ps, or PowerShell and its console host on Windows). pending_child_deliveries is the number of sub-agent results this session is still waiting for as a parent, whoever the caller is. Both are null when Aiterm cannot tell (ordinary terminals, agents launched by 0.48.0 or earlier for the process count, or an unresolved harness process), and may be absent when a remote host runs an older Aiterm. Treat null or absent as unknown, not as zero.

Another product that needs to deliver an answer to a Codex parent can ask Aiterm with aiterm-parent-delivery instead of registering its own hooks. Aiterm's registered hooks and the official queue deliver it. codex verify --thread <uuid> checks the parent and reports configuration in steer (enabled: the hooks are registered and trusted and the parent started after they were installed; disabled: official queue only, so the text arrives after the turn ends). codex submit --thread <uuid> --delivery <uuid> --text-file <file|-> hands the text over once and returns the queue's acceptance id. codex state --thread <uuid> --delivery <uuid> reports how it arrived (hook:"emitted" with turn_id: the hook put the text into that turn; queued: whether it is still in the official queue). Each result is one JSON line; failures are {ok:false, code, message, outcome_unknown} with exit 1, and a reused delivery id is refused with PARENT_DELIVERY_DUPLICATE. Only Codex parents are supported. aiterm-setup records the command's location in ~/.config/aiterm-mcp/delivery-provider.json. Node products can call it through aiterm-steer-delivery (0.3.0 or later), for example submitCodexParentAnswerViaAiterm.

When Aiterm is registered behind a relay that starts the MCP server only on first use, a sleeping server cannot take over the deliveries of an owner that has exited. aiterm-delivery-wake --parent <codex|claude|cursor> answers whether one needs to start: exit 0 when an exited owner still holds a delivery for that parent kind, exit 1 otherwise, exit 2 for bad arguments. It prints nothing, starts no child process, and marks the delivery so that only one of the seats running the same check starts its server (the mark is retaken after 30 seconds if nobody took the delivery over). For mcp-lazy, pass it as the wake predicate with the same environment as the server: MCP_LAZY_WAKE_COMMAND='["/absolute/path/to/node","/absolute/path/to/aiterm-mcp/dist/delivery-wake-cli.js","--parent","claude"]'. Owner start times are checked through /proc on Linux; elsewhere an existing PID is treated as a live owner.

aiterm-setup does not rewrite a registration that wraps the same server in the mcp-lazy relay. When the registered command is an executable whose name starts with mcp-lazy and its args are the direct registration's command followed by its args (relay flags before -- are allowed), the Claude Code, Codex, Grok, and Cursor registrations are left as they are. If the wrapped server path differs, setup writes the direct registration.

agent_launch({ harness, cwd, trust_project: true }) completes known workspace, project-hook, and project-MCP startup consent even without a prompt, then verifies input readiness and harness liveness before returning startup.status="ready". Startup checks wait for concurrent metadata writers before continuing. Each terminal recovery or startup-menu key still checks the current generation and stop state; an uncertain native launch remains available for status inspection without automatic replay. For Claude Code's first-run text-style menu, it confirms the item already selected on screen before continuing startup. If the CLI then requests an account login method, the launch reports vendor_onboarding_required; complete that choice in the official interactive Claude Code CLI. The input-readiness wait is 30 seconds. Only when it expires while the agent has not drawn anything yet (a busy host), the launch waits up to 50 seconds from startup. If nothing is drawn by then, it still returns initial_prompt=not_sent and keeps the session. Positive readiness deadlines are checked again after screen and transcript reads; a late ready or busy result cannot authorize input. A prompt-free launch without this option retains startup.status="not_checked". initial_prompt.status distinguishes not_requested, not_sent, submitted_unconfirmed, and started. Failure responses retain structured session information. WindowsのCodexもhook確認を認識し、npm shim経由の起動を一つのharnessとして識別する。 submitted_unconfirmed (the prompt left the composer, but the turn start was not observed within the confirmation window) is not a tool error: isError is not set, the text carries an unconfirmed-start note, the session is alive and the parent delivery stays registered. Do not resend an unconfirmed prompt or relaunch the agent; observe with pty_observe or wait using its returned cursor.

For a live Codex approval, inspect with agent_approval({ action: "inspect", session_id }), review prompt and choices, then respond with observed_prompt_digest and approval_choice (approve_once or deny). Unknown or changed dialogs return status="blocked" and isError:true without sending input. Permanent approval is not exposed. Correlated Claude approvals continue to use claude_approval.

diagnostics never starts a PTY or agent. It reports package version, MCP call readiness, a read-only PTY-list summary, bounded runtime-error-store status, and optional vendor-launcher availability. It deliberately excludes paths, environment values, credentials, command text, PTY output, and raw logs; normal unset optional dependencies are not_applicable, while an indeterminate probe is unverified.

diagnostics({format:"v2"}) keeps the two text items and uses aiterm-mcp.factory-diagnostics.v2 for the first. Its pty_list adds reason_code (listed, socket_absent, dependency_unavailable, permission_denied, timeout, output_limit, signal, or native_failure) and timeout_ms:5000. Binary resolution and one read-only native listing share that 5000 ms subprocess budget and a 1 MiB output limit; no PTY is created. An absent result requires exit 1, empty stdout, and the exact selected-socket absence diagnostic. Other sockets, permission/refused connections, conflicting output and unknown failures stay unverified. Unknown text is never echoed. {} or {format:"v1"} retains the existing fields and schema, with the corrected strict absence classification. Inspect the connected tool's advertised format before opting into v2 and require the returned v2 schema: older servers can discard unknown input fields. Vendor lookup and runtime-store checks retain their own finite deadlines; the 5000 ms field describes the PTY subprocess budget, not the entire MCP call or a hard real-time OS guarantee.

The result carries two text items. The first is the factory JSON described above (aiterm-mcp.factory-diagnostics.v1), whose fields are fixed. The second reports the parent-delivery hooks (aiterm-mcp.parent-delivery-diagnostics.v1): a status (ready / setup_required / not_applicable / unverified) and reason_code for Claude Code and Cursor, plus caller_status for the calling client. When caller_status is setup_required, agent dispatch from that client is rejected; run aiterm-setup. Hook status is not folded into overall in the first item.

Local runtime error snapshot

snapshotのproduct_versionは各recordの最終実発生時の版を表す。store v2は旧v1を読み取り、単発記録の版を保持し、複数回の旧集約の版はunknownにする。読取りでは状態JSONを書き戻さず、次のロック内更新でv2を保存する。consumerを先に更新し、旧writerの終了後に新writerを使う。v2保存後の旧版への切替は、製品のバックアップ復元を伴う。

aiterm-runtime-errors snapshot exposes a machine-readable, product-owned local snapshot for the dotagents factory adapter. Collection is fail-closed unless the canonical dotagents factory-reporter config is schema-exact, its host profile matches the executing OS, and it contains the JSON boolean collection.enabled: true; reporting fields are schema-validated but endpoints and credential files are never contacted, and the store performs no network I/O. The only accepted observations are three fixed codes owned by the core boundary (PTY dependency, persistence, and optional vendor launcher). Stored data is limited to fixed templates and aggregate metadata (SHA-256 fingerprint, count, first/last seen, status, and monotonic sequence); exceptions, stderr/stdout, stacks, prompts, terminal/transcript/event bodies, paths, and arbitrary context cannot enter the API. Persisted JSON is revalidated with exact top/record fields and a recomputed fingerprint before explicit DTO projection.

Consumer flow is aiterm-runtime-errors snapshot, then aiterm-runtime-errors ack --cursor N after durable ingestion. Operators can use resolve|reopen --fingerprint SHA256. MCP collection and diagnostic reads run in timeout-bounded child processes, so a FIFO or stalled filesystem cannot block terminal work; child failure emits only the fixed store diagnostic. Store mutation uses a bounded bakery ticket queue: every waiter owns a never-reused ticket containing PID, process-start identity, and an owner token, so dead owners are removed by unique filename without fixed-path reclaim ABA. The queue deadline measures lack of progress by the same head owner, not total wait behind healthy predecessors; normal polling uses the native process-liveness check and validates process-start identity only when a blocker stalls. Worker deadlines use forced termination so a SIGTERM-ignoring child cannot mutate state after timeout. POSIX state is atomically replaced under $XDG_STATE_HOME/aiterm-mcp/ (default ~/.local/state/aiterm-mcp/) with owner/mode rechecked on every read. Windows native uses %LOCALAPPDATA%\aiterm-mcp\; each DACL is rebuilt and read back as one non-inherited FullControl ACE for the current SID. Windows path/DACL/timeout behavior is covered by pure tests in this change; no new Windows integration success is claimed.

Reporting to BugHub is off by default. Aiterm sends runtime errors nowhere unless both hold: the user ran aiterm-runtime-errors reporting enable, and a credential file placed by the BugHub owner exists (~/.config/bughub/product-credentials/aiterm-mcp.json, or %LOCALAPPDATA%\bughub\product-credentials\aiterm-mcp.json on Windows; it is read only when it is a regular file readable by its owner alone). The destination comes from that file. The payload is the cumulative error codes, counts, first/last timestamps, versions, severity, and resolution marks; prompts, paths, and stacks are neither stored nor sent. A separate process does the sending, never the MCP process, and there is no polling: a report is attempted when an error is recorded, when a record is resolved or reopened, and when the MCP server starts (only while something is unreported, at most once per hour), or on aiterm-runtime-errors report (once, at most once per minute). A record becomes acknowledged only after the response signature is verified; otherwise the next attempt resends the then-current totals. aiterm-runtime-errors reporting status shows the switch, the credential state, the unreported count, and the last attempt; reporting disable turns it off. Enabling reporting also enables collection on that host.

Interactive agent harnesses

agent_launch starts a selected harness's interactive coding-agent TUI inside a fresh persistent PTY and returns its session_id. The harness owns the agent loop, authentication, hooks, session, and transcript; model is independent. The TUI is a full-screen app, so read it with pty_read({ screen: true }) for the rendered view.

agent_configure({ session_id, model?, reasoning_effort?, fast? }) changes a running Claude, Codex, Grok, or Cursor TUI through the harness's standard controls, preserving the PTY and conversation context.

New consumers can opt into an original request binding (UUID, expected launch, settings digest) and use action: "reconcile" to read the saved outcome without input or replay. Negotiate the exact tool schema before the first call. Handler completion does not prove that previously issued native input cannot act later; current settings inspection remains unsupported. Tracked apply is POSIX-only and uses a local target connection. Existing incidents without the original binding cannot be recovered this way. See the request/reconcile contract.

For an active Codex v1 goal, agent_configure temporarily pauses the goal, waits for the current native turn to end (up to 30 minutes), applies settings through the same send queue, and resumes that goal. Steering remains available during the wait. A changed goal is preserved; a failed or uncertain restoration is reported with the configuration outcome. Other harnesses retain the normal readiness limit: stop or complete a continuously running goal before configuring them.

Codex model-only changes explicitly preserve the current reasoning effort read from the native picker. Unreadable or unsupported current effort fails without selecting another value.

Long-lived Codex sessions recognize native model display names containing spaces in the status footer, so a ready composer remains available after the launch header scrolls away.

Model IDs come from each harness's native catalog. Claude aliases and differing Codex picker labels use that catalog's display name; Cursor matches complete native IDs or their exact supported effort variants. No model allowlist or alternate-model fallback is added. Codex success requires a fresh native change notice matching the requested ID and effort.

Configuration failures keep isError: true and the original error text, with structuredContent: { schema: "aiterm.agent-configure-error.v1", session_id, outcome, effect } for all five harnesses. not_applied / not_started means this request failed before a model, effort, or fast change could be issued. outcome_unknown / possible means a change may have been issued, including partial application before a later failure; do not replay it automatically. This receipt does not certify the session's previous configuration or clear unrelated uncertainty. Older servers without this receipt must be treated as unconfirmed on failure. Successful results retain aiterm.agent-configure-result.v1.

ファスト（fast）。 各harnessの、費用や利用量を多く払って応答を速くする切り替え（Claude Codeの/fast、Codexの速度の段、GrokとCursorのfast版のmodel）を、席ごとに切り替える（ADR 0114）。

• 席は必ず通常の速度で起きる。agent_launchのfastを省いた時とfalseは同じで、利用者の設定にファストが残っていても通常で始まる。fast: trueは通常で起こしてから、その席を切り替える。

• agent_configure({ session_id, fast })は、動いている席を起こし直さずに切り替える。fastだけでも渡せる。省略は、その席の今の状態を変えない。

• 切り替えは席だけに留める。harnessの切り替えの命令は利用者の既定（~/.claude/settings.jsonのfastMode、~/.codex/config.tomlのservice_tier、~/.grok/config.tomlの[models]）も書き換えるので、Aitermが切り替えの直前の値へ戻す。Grokは起動の--modelでも既定を書くので、起動の前後でも戻す。

• 返りのfastは、切り替えた後の実際の状態。agent_modelsのmodelごとのfastが、そのmodelで切り替えられるかを示す（GrokとCursorのfast版は別のmodelとして並ばず、元のmodelのfast: trueになる）。最上位のfast_unavailableは、アカウントの都合でどのmodelでも使えない時の訳。

• 断りは文の先頭の符号で分かる: FAST_MODE_NOT_SUPPORTED（そのmodelは対応していない）、FAST_MODE_UNAVAILABLE（アカウントや契約で使えない）、FAST_MODE_MODEL_CONFLICT（modelへfast版のIDを書いてfast: falseを渡した）、FAST_MODE_UNCONFIRMED（切り替えた後の状態を確かめられない）。起動でFAST_MODE_UNAVAILABLE・FAST_MODE_UNCONFIRMEDになった時は、起動を断らずに通常の速度で起こし、返りをfast: false、fast_error: { code, message }にする。

Claude Codeの席では、/modelの選ぶ画面のs（use this session only）で替える。/model <名前>・/effort <値>は選んだ値を端末の~/.claude/settings.jsonへ新しい会話の既定として保存するので、打たない（ADR 0111）。その選び方が無い版、選ぶ画面に無いmodel、そのmodelで選べないeffortは、何も変えずに断る（CLAUDE_MODEL_SESSION_ONLY_UNAVAILABLE・CLAUDE_MODEL_NOT_IN_PICKER・CLAUDE_EFFORT_NOT_SELECTABLE）。席を閉じて立て直すと端末の既定へ戻るので、立て直す時はagent_launchへ同じmodel・reasoning_effortを渡す。

agent_auth({ harness, action:"start"|"status"|"cancel"|"respond", session_id?, cwd?, env_vars?, relogin?, expected_binding?, response? })は、各harnessの公式CLIで認証を進める。Claudeはclaude auth login、CodexとGrokはlogin --device-auth、CursorはNO_OPEN_BROWSER=1 cursor-agent loginを使い、資格情報は各CLIだけが保存する。Aitermは資格情報を読取り・copy・編集せず、独自OAuthも実装しない。remoteは他toolと同じ標準対応。

認証UIは追加のauth_uiを使う。stageはopen_url / device_code / paste_code / select / waiting / complete / failed / unavailable。公式URL・表示code・具体的な選択ラベル・入力種別をNativeが返し、input_requiredだけから入力内容や端末キーを推測しない。respondは直近のauth_ui.bindingをexpected_bindingへ渡し、expected_prompt_digestには同じauth_ui.prompt_digestを渡す。response:{kind:"authorization_code",value}または{kind:"choice",id}を同じ認証flowへ送る。認証コードをpty_sendへ渡さない。専用入力はlastcmd・通常turn・paneログ・argv・一時fileへコードを保存せず、CLIの画面echoもpaneログへ保存しない。未発行と入力後不明は構造化errorのeffect:not_issued|unknownで区別し、自動再送しない。legacy flowのlaunch_id:nullは実native識別情報で照合し、UUIDを補わない。Windowsの専用入力は前面所有確認が未受入のためnot_issuedで拒否する。

コード入力後は同じ認証世代の状態を最大3秒確認する。画面更新後のwaitingは処理待ちで、認証成功を意味しない。入力待ちが変わらない・世代が変わる・状態を読めない場合はunknownを保ち、入力を再送しない。

認証processの端末切断が終了コード確定に先行した時は、同じ世代の終了コード・signalを3秒の期限付きで確認する。signal終了や非0は失敗を保ち、0でも公式statusを確認する。期限まで不明、session消失・交代は成功にせず、ログインの再起動や入力再送は行わない。

startは、公式の状態が認証済みなら何も起こさずauthenticated（session_id:null）を返す。relogin:trueを付けると、認証済みに見えても公式ログインを開始し、session_id付きの結果を返す。別のアカウントへ入り直す時と、ログインの期限切れを状態から見抜けない時に使う。Aitermは資格情報を消さず、置き換えは公式CLIが行う。ただし、公式CLIがログインを始めた時点で元のログインを消す事がある（Codex 0.160.0のcodex login --device-authは、始めた時点でauth.jsonを消す。途中でcancelしても戻らない）。Claude CodeとCursorは、途中でcancelすれば元の資格情報が残る（無効な資格情報で確認）。

status（sessionなし）は、公式CLIに今の状態を聞く。

結果はaiterm.agent-auth-result.v1。statusはwaiting／authenticated／blocked／failed、session_id・url・user_code・input_required・messageを返す。startのsession IDを保存してstatusへ渡す。公式HTTPS URLと明示device codeだけを返し、CLIの生出力・token・OAuth callback codeを結果へ載せない。auth_uiの安全な案内だけを表示し、入力は同じbindingとprompt_digestを持つ専用respondへ渡す。生画面や通常PTY入力へ認証コードを中継しない。

Grokには公式認証status commandが無いため、session付きの確認は公式loginのexit 0を正本にし、session無しの確認は公式modelsの認証要求を照合し、不明なら成功扱いしない。他harnessは公式statusも照合する。Claudeは認証後の公式初回案内を同じsessionで進め、選択待ちはblocked／input_required:trueで返す。authenticatedは認証結果であり、agent_launchの起動準備完了は別途確認する。cancelは指定した認証sessionだけを閉じ、資格情報を削除しない。

PTYが消失した場合は、相関記録の有無にかかわらずstatusがfailed、cancelが既に終了・取消済みを示すblockedを返し、どちらもsession_id:nullとなる。保存したsession IDを解除してstartで再開始できる。生存中の通常PTYやharness不一致、記録の破損・読取り失敗はエラーを返す。

agent_models({ harness, cwd?, include_hidden? }) returns the model IDs and reasoning efforts the installed harness offers right now, so a UI can build its choices from the machine that actually runs the agents. It reads each harness's own catalog and never sends a prompt or starts a turn:

The result (aiterm.agent-models.v1) is { harness, source, harness_version, default_model, efforts, adapter_efforts, models: [{ id, display_name, efforts, default_effort, hidden }] }. Every id and effort can be passed to agent_launch and agent_configure as is; efforts at the top is the union across models. An unavailable catalog is MODEL_CATALOG_UNAVAILABLE and a malformed one is MODEL_CATALOG_INVALID; Aiterm never falls back to another list. With remote, the catalog comes from the harness on that host.

{ "name": "agent_models", "arguments": { "harness": "grok-cli" } }

The selected harness CLI must be installed and authenticated. Use each product owner's official installer and updater; Aiterm does not distribute alternate CLI tarballs. For Cursor Agent CLI, use curl https://cursor.com/install -fsS | bash on macOS/Linux/WSL or irm 'https://cursor.com/install?win32=true' | iex on native Windows, authenticate once with agent login, and update with agent update; Aiterm invokes the unambiguous cursor-agent binary. Missing binaries, invalid model/effort values, unavailable Grok catalog models, and nonexistent cwd fail before a session exists.

Claude authentication preflight refusals retain the existing error text and add structuredContent with schema aiterm.agent-launch-error.v1, harness:"claude-code", phase:"authentication_preflight", effect:"not_issued", and session_created:false. code distinguishes CLAUDE_AUTH_NOT_AUTHENTICATED, CLAUDE_AUTH_STATUS_UNCONFIRMED, and CLAUDE_AUTH_STATUS_TIMEOUT. These describe the official status check, not the cause of a credential failure or a confirmed login expiry. Native status bodies and stderr are omitted.

Set throughline_source_session together with a non-empty mission in prompt to prepend Throughline's read-only handoff context. This optional route requires throughline >= 0.9.0, cannot be combined with launch_operation_id, and leaves the source session's database ownership unchanged. Optional throughline_supplement_file is passed unchanged to Throughline and requires throughline_source_session and Throughline 0.10.8 or later; Aiterm does not read or classify the supplement. Throughline is resolved through THROUGHLINE_BIN and then PATH; a missing or invalid export fails before the PTY exists instead of silently launching clean.

When an agent's answer is longer than the on-screen tail (pane height ≈ 24 lines), callers recover it in full with pty_read({ agent_transcript: true }). It returns the most recently completed turn's final assistant message in plain text with no re-prompting. The existing human-readable content keeps its diagnostic suffix; machine callers read the answer alone from structuredContent.text in aiterm.pty-read-result.v1. Claude reads the bounded owner-only result captured by the launch-correlated Stop hook and verifies its digest/byte count; it never reads Claude's private transcript. Durable machine callers should use claude_turn: issue sends once, recover never sends, pending is distinct from unsafe or malformed state, and only completed carries the exact verified raw_output. Codex uses the normal rollout transcript's task_complete.turn_id; Grok returns the last non-empty assistant message after the last real user row, excluding tool-use preambles; Cursor uses the normal agent transcript bound to the launch ID and current turn. A turn that completed with an empty answer is not an error: text is the empty string and answer_empty is true, so callers can tell it apart from an answer that could not be read. Missing or ambiguous attribution, including an answer that cannot be located in the record, remains an explicit error.

Completion detection (5 layers)

For PowerShell over SSH, mark:true recognizes the current standard PS ...> prompt and emits PowerShell syntax even when Aiterm runs on macOS or Linux. A prompt left in earlier output is not used to select the syntax.

On native Windows, a wrapped current Git Bash prompt is reconciled with the measured pane columns before choosing shell syntax. Only full-width continuation rows of the current empty prompt are joined; earlier commands, other prompts and unknown widths do not authorize a correction. Ordinary and marked sends use the same runtime decision.

pty_read({ wait: true }) decides "is the command done?" via five layers: process exit / a mark:true sentinel / an until match / output quiescence with shell return / timeout. Quiescence requires new output after the read cursor; an empty stream stays unconfirmed until its deadline. Each new mark carries a submission ID, so an older command's late sentinel cannot complete a newer read; legacy pending marks remain readable. If its ID cannot be saved, the command is refused before input. mark emits the shell's exit status on POSIX shells and 0 (success) or 1 (failure) on PowerShell; fish/csh/tcsh are rejected before send because they do not share either status syntax. When mark or until is active, that requested evidence takes precedence and a momentarily quiet shell cannot complete the read as quiescent. Agent sessions add a sixth exact layer: Codex observes normal rollout task_complete; Grok observes normal session turn_ended; Claude observes its additive launch-correlated Stop event; Cursor observes turn_ended(status:"success") in the launch-bound normal agent transcript. aiterm-wait --cursor performs that harness-specific observation without the parent blocking or polling. Pre-send readiness failures are MCP errors, and late completion remains recoverable without resending.

Completion push for parent agents (aiterm-wait)

Codex／Claude Code／Cursor親には子の回答本文が自動で届く。 Codex／Claude Codeでは、子を起動・dispatchした後は別作業へ進むか親のturnを終える。Aitermが完了を観測し、加工前の本文を保存して親へ渡す。waiter、pty_readによる回答回収、子への返送指示は不要。子は全対応harnessから選べる。

Codex／Claude Codeの自動配送ではreceiptにparent_deliveryが付き、wait_process／wait_commandはnullになる。pty_observeのparent_deliveriesでwaiting、ready、sending、submitted、failed、unknownを確認できる。submittedはCodexの公式受信口での受付またはClaudeのhookへの本文出力を示し、modelの読了ではない。MCP再接続後は未送信の記録を再開し、出力中断で結果が分からない場合は本文を保持してunknownとする。自動再送はしない。

For queue delivery, use a Codex runtime that supplies MCP _meta.threadId and the official thread/queue API (verified with Codex CLI 0.154.0). aiterm-setup checks the installed queue entry point; Aiterm verifies the requesting thread before each dispatch. Codex native sub-agents reject external queue input and cannot be automatic-delivery parents. Steer相当の選択時も公式キューへ投入し、専用hookが同一ターンへ取り込みます。

Claude Codeは2.1.259以上の対話sessionに対応する。aiterm-setupが専用のPreToolUse、PostToolUse、SessionEndを登録するため、Channelsの起動flagは不要。公式asyncRewake hookだけが裏で待ち、親はその間も次のturnへ進める。回答はStop hook feedbackとして届く。hookのexit 2は親の再開信号であり、子の成功・失敗は本文のoutcomeで区別する。

/clear等の会話終了後は未送信の旧回答を送らず、本文を保存する。受信hookの上限は24時間。hookの終了・出力失敗・無効化を成功扱いせず、別の待機経路へ黙って切り替えない。Claude Desktopのチャット、Web、agent_id付きの会話（--agent起動とnative subagent）はこの受信契約に含めない。

hookを持たない旧版、または0.52.2以前へ戻す時は、install前にaiterm-setup --remove-claude-parent-hooksを実行する。Aiterm専用hookだけを解除し、他製品のhookと設定は保持する。

Cursor parents (clientInfo.name of cursor-vscode) use the same completion capture. aiterm-setup adds afterMCPExecution and postToolUse to ~/.cursor/hooks.json and keeps every other hook and its position. A missing registration fails the dispatch with CURSOR_PARENT_HOOK_UNAVAILABLE before the child is sent. While the parent keeps calling tools, the answer is injected through additional_context on the next tool result. If the parent ends the turn, start the receipt wait_process in the background first; that receiver exits when the answer arrives. wait_command is null. submitted means the hook or the receiver claimed the text, not that the model has read it. No claim within 24 hours is failed, the text is kept, and nothing is resent. Remove only Aiterm's entries with aiterm-setup --remove-cursor-parent-hooks. Cursor Cloud Agents and Background Agents are outside this contract.

Claudeをリンク経由のcwdから起動した場合も、実体パスに対応する会話記録を参照する。

For other parent hosts, dispatch and start the receipt's waiter in a separate process:

1. Launch the child with agent_launch({ harness: ... }); every launch shares the normal project/user environment and adds only completion correlation plus lineage. Send a turn with plain pty_send (or claude_turn issue for durable Claude operations). The call returns immediately with an event_cursor in its structured receipt. If the child is still working when you send, Aiterm steers the text into the running turn instead (mode: "agent_steer"); the original request's completion then covers it, so no new cursor or delivery is created.

2. Pass the receipt's wait_process.executable and wait_process.args unchanged to a true argv process API. PowerShell 7's Start-Process is the exception because it joins -ArgumentList arrays; pass windows_start_process_argument_list as its one ready-made argument string instead. This invokes the bundled waiter through the exact Node runtime that is already running aiterm, including on native Windows where npm's human-facing bin is a PowerShell script shim and install paths may contain spaces. wait_command remains a compatibility display string for humans. The waiter observes the harness-owned completion source, plus Claude's additive launch hook, as a pure reader and exits with a one-line aiterm.agent-wait-result.v1 receipt. Exit ≠ done: the receipt's outcome is authoritative (0 = done, 3 = timeout, 4 = closed, 1 = error).

3. 親自身のforegroundでwaiterを実行しない。 receiptのprocess起動情報を、そのhostが持つバックグラウンドprocess APIへ渡す。親は別作業へ進むかturnを終え、process終了の通知で続行する。

4. Collect the result exactly as before: pty_read(agent_transcript: true), or claude_turn recover for durable Claude operations. The waiter carries the signal, never the payload.

If your host has no completion push (no mechanism that re-invokes the agent when a background process exits), --timeout 0 is a one-shot check instead of a wait: it scans the event file once and returns running (exit 5) when the turn is still in flight, done (exit 0) when it finished, closed (exit 4) when the session is gone. It is deliberately absent from the receipts and tool descriptions — a host that does get pushed should be woken, not poll. An unknown session name is an error, never running, so a typo cannot masquerade as a child that is still working.

aiterm-wait takes no locks, never writes session state, and never dispatches — any number can run beside the MCP server and each other, and pty_close/concurrent sends are unaffected.

Launch an agent on another machine in one call (remote)

Add remote to agent_launch and the same launch runs in the Aiterm of another machine reached over SSH. Entering the machine and starting its agent become one call, and completion reaches the parent exactly as it does for a local child. The intended use is sending the same task to Linux, macOS, and Windows machines in parallel.

agent_launch({ "harness": "codex-cli", "remote": { "host": "rabbit" }, "cwd": "/home/kite/project", "prompt": "..." })

• remote takes host, user, port, identity_file, passphrase or passphrase_env, and ssh_options (Key=Value). A bare host is used as an ~/.ssh/config alias. Aiterm does not store or manage destinations; the caller owns where to connect and with which key.

• A plain passphrase stays in the calling AI's conversation log. Prefer ssh-agent or passphrase_env (an environment variable name). A received passphrase lives only in the MCP process memory and reaches ssh through SSH_ASKPASS; it is never written to state files or logs.

• The remote machine needs aiterm-mcp, tmux (psmux on Windows), and the harness CLI. The remote shell family (POSIX, PowerShell, or cmd) is detected on first contact. On POSIX machines Aiterm takes only PATH from the login shell, so CLIs under ~/.local/bin, Homebrew, or nvm are found; on Windows it starts aiterm-mcp with the user's PATH as is.

• Pass the same remote to later pty_send, pty_read, pty_close, pty_observe, and so on. Session names belong to the remote machine and never collide with local sessions of the same name.

• Codex, Claude Code, and Cursor parents receive the answer automatically, as with a local child. Completion is observed with ssh <host> aiterm-wait, reconnecting at the same cursor if SSH drops. Other parents get an ssh-based wait_process.

• Calls to the same destination share one SSH connection through ControlMaster. image attachments and claude_turn issue are not yet supported with remote.

• POSIX SSH control sockets use a short private directory even with a long TMPDIR. Connections stay isolated by state root; OpenSSH closes idle masters and removes their sockets after 600 seconds.

Token reduction

• pty_read by default strips control characters, collapses repeated lines, and folds long output into head+tail (with a restore hint and a meta line).

• Carriage returns followed only by cursor visibility controls preserve printed output and completion marks.

• pty_read({ rtk: true }) further shrinks the observed output with a per-command reducer (git status/git log/grep/pytest and more) — a self-contained reimplementation that needs no rtk binary.

• pty_send({ rtk: true }) rewrites a known command into rtk form before sending, so reduction happens at the source if rtk exists there (passthrough otherwise).

Input and output

Ordinary delivery and parent-return hooks do not acquire or wait for the shared execution tracking lock. The legacy native entry now serves ordinary delivery plus leaf-first agent_stop, without agent_execution registration. Hosts using that entry must remove mandatory ExecutionBridge registration; recording failure must not prevent delivery.

pty_send does not interpret command or prompt meaning; it delivers the requested text to the terminal. By default it sanitizes ESC and bracketed-paste terminators, while pty_read neutralizes control characters in returned output (raw: true keeps them unchanged). The shell, remote endpoint, or launched harness owns command authorization.

Each pty_send accepts at most 64 KiB of UTF-8 text. Sends to the same session are serialized across aiterm processes so chunks cannot interleave. Every OS pastes through its multiplexer in UTF-8-safe 256-byte chunks with a 10 ms drain interval; macOS, Linux, and WSL2 have all demonstrated silent middle/trailing loss when a long input is pushed without that boundary. Sanitized multiline text sent while a POSIX shell or PowerShell is in the foreground is encoded as one newline-free input (eval for POSIX shells; a dot-sourced, UTF-8 Base64-decoded scriptblock for PowerShell): the shell receives the complete script before it runs the first line, so a pager or REPL started mid-script cannot consume later lines as interactive keystrokes. Single-line input, raw:true, and non-shell frontends remain direct PTY pastes. Agent dispatches additionally wrap the whole text once in ESC[200~/201~ and stream the wrapped text in the same chunks, so the agent TUI sees one paste (an image path is never split across pastes), hardening prompt injection against mid-word key-interpretation corruption and dropped submits. If a later chunk fails, aiterm reports the partial-send state and does not press Enter automatically. A lock left by a terminated sender fails closed before sending; use pty_list to confirm the affected session, close it with pty_close, then recreate the same session ID. There is no public kill-all tool.

A human can watch

On native Windows, non-raw commands sent to a PowerShell shell use UTF-8 for Console input/output and native pipelines. Runtime additions to the caller PATH are passed to new panes; raw PTY input is unchanged. Agent launches use a PowerShell 7 pane and pass the harness argv directly; accepted shebang wrappers use noninteractive Git Bash in that same pane. The owned startup script is sent as one ASCII line, while agent TUI bodies retain their literal text.

Sessions live on a shared tmux socket on POSIX or a shared psmux namespace on native Windows. The attach line printed by pty_open and agent_launch lets a human attach to the same terminal and intervene, including a Claude/Codex/Grok/Cursor harness session: tmux -S … attach -t <id> on POSIX, or psmux -L <namespace> attach -t <id> on native Windows. On POSIX, the first agent_launch can create the tmux server when its socket does not yet exist. Aiterm verifies the exact created session and server before accepting that launch; an uncertain result is reported without retrying the input.

Requirements

• Node.js >= 18

• tmux or psmux (platform runtime prerequisite)

  • macOS / Linux / WSL2 run tmux directly. On macOS install it with brew install tmux (stock macOS ships none). If your MCP client is launched from the GUI rather than a terminal, Homebrew's bin (/opt/homebrew/bin on Apple Silicon, /usr/local/bin on Intel) may be off its PATH; aiterm auto-searches those locations, or set AITERM_TMUX=/path/to/tmux to point at it explicitly.

  • Native Windows has no tmux, so aiterm drives psmux — a tmux-CLI-compatible native terminal/session multiplexer — with a per-install -L namespace. psmux is not a shell. pty_open defaults to PowerShell 7 (pwsh.exe) and never falls back to Windows PowerShell 5.1, PowerShell 6, or cmd.exe; if only 5.1 is installed, use Microsoft's official installer or package manager first. Install psmux 3.3.8 or newer (winget install marlocarlo.psmux; 3.3.8 is the first release whose pipe-pane file sink, byte-exact paste-buffer wire, and foreground #{pane_current_command} behave the way aiterm's capture/dispatch paths rely on). Git for Windows remains required for the explicit Bash shell used internally by harness launchers; System32's bash.exe is the WSL launcher and is deliberately not used. Override multiplexer/Bash resolution with AITERM_PSMUX / AITERM_BASH. Other products consume persistent terminals through Aiterm's public API instead of depending on psmux directly.

• For agent harnesses: the selected CLI, installed and authenticated through its product owner's official path — claude, codex, grok, or Cursor's cursor-agent. Portable fork additionally needs throughline >= 0.9.0; ordinary clean launch does not. (Not needed if you only use the PTY tools.)

• Optional: the rtk binary (used by pty_send's rtk: true delegation; works fine without it)

Known constraints (by design, not bugs)

• While nested (ssh / docker / REPL / a launched agent TUI), quiescence cannot fire by design, because the foreground command is no longer in the shell set (bash/sh/zsh/fish/dash). When nested with no until and no mark, pty_read({ wait: true }) returns early as is_complete=False via nested (rather than burning the full timeout, since no signal can confirm completion there) with a note to pass until (a literal substring by default; until_regex: true for a regex) or mark: true (an exit-code sentinel, auto-detected) for a confirmed completion. For a full-screen agent TUI, read { screen: true } once its output settles.

• is_complete=False is not a failure. It means "completion was not observed within timeout." For long commands, raise timeout or use until/mark.

• Agent harnesses run their real TUI; aiterm doesn't proxy the model API. The selected harness owns model choice, authentication, and behavior. There is no hidden inter-agent protocol; the MCP client drives the Claude/Codex/Grok/Cursor TUI with ordinary send/read operations.

• pty_send({ rtk: true }) is single-line only and needs the external rtk binary (passthrough without it). The pty_read({ rtk: true }) reducer, by contrast, is self-contained and rtk-independent.

• The pytest reducer matches rtk 0.50.0 on test counts and FAILURES-block formatting (locked by regression tests). It deliberately preserves the full failure reason on the FAILED summary lines (emitted under -ra/-rf), whereas rtk 0.50.0 truncates the reason at the first " - " — a readability choice, so those lines are intentionally not byte-identical to rtk. The [full output: …] recall-pointer line rtk appends on large output is not reproduced on the read side.

• The grep reducer matches rtk 0.50.0: within the caps it returns the output unchanged, and past a cap its grouped form is byte-identical to rtk grep without the recall hints. git log keeps every commit when you set a count (-n) or a range (A..B); otherwise it shows 10 and ends with [+N more commits]. Like rtk's never_worse, a reducer whose result would cost more tokens than the output is skipped.

• tmux is started with -f /dev/null, so it does not read ~/.tmux.conf (to keep behavior reproducible across machines).

• All sessions share one multiplexer endpoint (claude.sock on POSIX, one psmux namespace on native Windows). The platform's kill-server command removes them all.

Development

npm install
npm run build      # tsc → dist/
npm test           # build, then the node:test regression suite (requires tmux or psmux)
npm link           # put `aiterm-mcp` on PATH locally

開発中は変更に直結する試験を先に実行する。GitHub Actionsは共通実装・CI自身・未分類の変更をMac・Linux・Windowsで検証し、 Windows固有だけの変更はLinuxとWindowsを選ぶ。版番号だけの変更はLinuxの配布情報・pack確認、文書だけなら文書検査を行う。 試験内容とOSの選択、週次・手動実行の範囲は公開手順に従う。 npm run release -- <version> syncs the version, commits, tags, and publishes the GitHub Release in one command; tag-triggered npm publishing checks only that the tagged commit is on origin/main and does not wait for another CI run. The native Windows runner needs psmux ≥ 3.3.8 and Git for Windows on its PATH, and must run as an interactive Windows user; NETWORK SERVICE lacks the per-user environment the pane shell and harness CLIs rely on and is not a valid runner identity.

Logic lives in src/core.ts (tmux control, reduction, completion detection, safety, agent launch) and src/rtk.ts (per-command reducers); src/index.ts is the MCP surface. The current architecture is in docs/DESIGN.md, the release procedure is in docs/RELEASE.md, and prototype/python/ remains the reducer's historical porting source (the pytest and grep reducers are ported to match upstream rtk 0.50.0, except the deliberate FAILED-line difference noted above, and is locked by regression tests).

Try it

公開packageを導入して、検出したAIへ登録する。cloneやビルドは不要:

npm install -g aiterm-mcp@latest
aiterm-setup --json

If aiterm let your AI hand a task to another agent — or saved you a round-trip of tokens — star the repo. It's the cheapest way to help others find it.

• npm: https://www.npmjs.com/package/aiterm-mcp

• Issues / bug reports: https://github.com/kitepon/aiterm-mcp/issues

Shared agent environment

All harnesses use the caller's normal project and user environment. Aiterm does not copy, symlink, filter, or replace harness configuration, authentication, MCP, plugin, skill, permission, trust, memory, or history stores. Cleanup removes only aiterm-owned launch metadata and completion correlation files.

The ordinary environment comes from the MCP process that opened the terminal, not from whichever caller started the persistent multiplexer server (since 0.50.0). Every harness still accepts env_vars: ["NAME", ...]; those names are placed on the launch command and registered on the session.

License

MIT

Claude authentication URL receipts recognize the official claude.com/cai/oauth/authorize PKCE authorization request and its nonsecret code=true display flag. Callback authorization codes and tokens remain excluded; the common URL credential filter is unchanged.

Native agent goals

agent_goal accepts session_id, the original 32-character launch_id, and action: inspect, create, edit, pause, resume, or clear. expected_thread_id compares a previously observed native binding; it never selects a different thread. create requires objective; edit requires objective or a positive integer token_budget. Other actions do not accept those fields.

The first adapter attaches to an existing Codex native control socket on POSIX and verifies the thread is loaded there. It never starts a worker or resumes a stored thread to make an unavailable Bot appear running. The launch must still have an identifiable live harness. Cursor adapters and native Windows Codex transport return explicit unsupported status; a missing binding, disabled feature, or missing connection returns unavailable.

If the original CLI uses an embedded worker without an attachable control endpoint, inspection uses the official read-only native API and pause/resume/clear run once through that same CLI. This mode also supports budget-only edits through the official goal service. It advertises create:true with {objective:true, token_budget:false}, and edit:true. Creation refuses an existing goal, an existing draft, terminal control characters, or an objective exactly equal to the native reserved controls clear, edit, pause, or resume (case-insensitive). Budget-bearing creation is rejected before either field is issued, rather than creating an unbudgeted goal. For a literal existing objective, field_capabilities.edit is {objective:true, token_budget:true, combined:false}. Creation uses bounded ordered literal paste events so native large-paste attachment folding does not replace the objective; complete canonical objective and active-status readback remain required. Objective-only edits use the current CLI’s native editor, including reserved words and multiline text. File-backed native objectives keep objective editing disabled. Combined objective/budget edits are rejected before either mutation; clients must not split them automatically. Clients must honor both action and field capabilities. Missing field capabilities on older servers retain the older action-wide meaning; missing combined retains the existing field-wide meaning. The TUI path verifies that the original PID and process group own foreground input on the same TTY, with explicit raw/no-echo flags; unknown, stopped, background, cooked or changed ownership rejects before input, and ownership lost after input remains unknown. It never resumes a job, changes terminal mode, or clears a draft to pass this check. The TUI path distinguishes styled placeholders from existing input, confirms cleared and pasted editor contents before one Enter, and preserves an existing draft. A proven input refusal reads fresh native state and reports not_issued; unconfirmed submission or cancellation remains unknown without replay. Slash-command reserved words, native editor input, and token budgets are not interchangeable with the full native API.

Only an authenticated host action may stamp MCP metadata {"aiterm/goal":{"origin":"user"}} on the existing local bellteam client connection. BellTeam must reject seat-origin changes and authenticate the Owner before stamping this provenance. This metadata is not a new authentication mechanism. Other clients and missing provenance remain automatic and cannot mutate a goal. Embedded hosts may supply a stricter goalOrigin resolver to the server constructor.

aiterm.agent-goal-result.v1 includes native binding, per-action capabilities, availability, operation outcome, mutation acknowledgment, and readback status. A missing goal field is unknown; available with goal:null is an authoritative empty goal. All explicitly requested native fields, including the active status requested by create, must match the readback. A changed result remains unknown with the observed native state. Unknown mutations are never retried. clear does not interrupt a running turn, resume requires paused, and a new objective resets Codex usage accounting. For an explicit Owner whole-application stop, the native delivery entry accepts agent_stop with pause_goals:true and the same user metadata. Each selected descendant completes goal pause and one ordinary interruption key before its parent; independent roots proceed concurrently. Unconfirmed goal pause stays partial even when a key was sent. Ordinary per-Bot interruption is unchanged, and enabling AI does not resume goals.

Claude Code 2.1.310 supports inspect and user-authorized create, objective-only edit, and clear on the same live TUI. It opens the native goal dialog from a strictly empty composer, verifies the original TTY owner, combines the fresh dialog with matching native root-session attachments, and dismisses only the recognized dialog. A draft or unknown viewport returns unavailable without clearing it. Wrapped conditions that cannot be matched exactly remain unavailable. Claude replies use aiterm.agent-goal-result.v2: goal is null or {kind:"claude-code",phase:"active",native,reported_turns?}, with an optional separate native_history for achieved versus explicitly cleared goals. Missing native accounting stays missing. Codex replies retain v1. Claude pause/resume remain unsupported; whole-app stop must retain an active Claude goal as unconfirmed when native pause is unsupported.

Claude authorization receipts recognize the official claude.com/cai/oauth/authorize PKCE request and its nonsecret code=true display flag. Callback authorization codes and tokens remain excluded.

Claude clear submits the native /goal clear once and requires a fresh empty dialog; a previously active goal must also have matching native cleared history. It leaves an in-flight model request running and removes further goal continuation. Lost ownership, replaced generation, missing readback or another goal's terminal event remains unknown without replay. Refusal proved before submission refreshes status without restoring an older goal.

Claude objective creation and replacement accept literal lines that can be fully observed in both the composer and native goal dialog. Blank/indented/trailing-space inner lines, control text, viewport overflow and the case-insensitive reserved objective clear are unsupported before submission. Ordinary pause, resume, status and --budget text remain literal objectives. Create refuses an existing goal; edit requires one. Both advertise objective support with token_budget:false; combined edits remain false and budget-bearing requests never issue a partial update. Exact newlines and native objective readback are preserved. Native dim queued-input hints are distinguished from an actual draft containing the same text by their styling.

Native Claude goal input also verifies the official per-PID current-session registration, including native session ID, process birth ticks and PID namespace before each input boundary. A switched conversation is refused before goal input; ownership loss after input remains unknown. No conversation is switched back automatically. This registration birth format is currently verified on Linux; unverified OS/registry data remains unavailable.

The Linux Grok 1.0.70 inspector binds the native current-session registry to the original PID, birth ticks, cwd and conversation. A fresh local status reply confirms no goal or an existing goal when the composer is idle. goal.native is the persisted snapshot; goal.current_status holds the fresh worker status, phase, token count and exact elapsed display. Stored phase and accounting values are never substituted for these current values. An already fullscreen TUI can also expose its live Goal detail without stopping the worker: inspection opens and closes that native view, preserving an empty composer. Its current_status.source is goal detail; phase and exact tokens_used remain null while phase_display, status_display and budget_display preserve the UI text, including rounded usage. Inspection never changes the screen mode. Busy minimal views remain unavailable without queuing a command. A user-origin clear is supported only on the same worker with fresh UserPaused/Idle status and an empty composer: it submits the official local command once, requires its fresh durable-clear reply and deleted native state, then confirms a fresh no-goal status. Uncertain effects remain unknown without retry or restoring the goal. Other mutations remain unsupported. The ordinary launcher has been checked without manually editing its metadata. This adapter is limited to the verified Linux contract; other operating systems remain unavailable until their native registration and input ownership are accepted.

Offline resume checkpoint

aiterm-resume-state checkpoint --session ID --launch LAUNCH emits a private bundle on stdout and a safe receipt on stderr. aiterm-resume-state restore --session ID --launch LAUNCH --sha256 SHA reads that bundle on stdin and emits its receipt on stdout. This Linux/Codex scope requires the explicit persistent AITERM_STATE_BASE, the same UID/absolute paths and retained vendor rollout identity. The caller must stop all source writers and prevent restart before checkpoint; restore runs under startup admission before bootstrap or native startup. It does not recover already-lost legacy tmpfs state or migrate pending deliveries. The bundle is limited to 16 MiB including LF; the safe receipt is at most 512 bytes. Success exits 0; refusal exits 2 with aiterm: RESUME_STATE_UNCONFIRMED plus LF (33 bytes). See the contract and diagram.