What can you do with this server?

The aiterm-mcp server allows an AI to drive persistent terminal sessions and orchestrate other coding agents (Claude Code, Codex, Grok, Composer) programmatically and headlessly. * Persistent PTY: Open, send commands/keys, read output (with intelligent token reduction), close, and list tmux-backed sessions that survive server restarts. * Agent orchestration: Launch agents in isolated sessions; dispatch tasks non-blocking with an event_cursor; use aiterm-wait to asynchronously monitor completion via a 5-layer detection system (process exit, marker, until pattern, quiescence, timeout). * Token reduction: Output is automatically folded, elided, and can use per-command reducers (e.g., grep, pytest) to save context. * Structured Claude management: claude_turn for durable issue/recover operations and claude_approval for secure inspection and response to approval prompts with digest correlation. * Safety tripwire: Blocks destructive commands (e.g., rm -rf /) unless force: true. * Human co-driving: Any session can be attached via tmux for live observation or takeover. * Diagnostics: Read-only health check returning machine-readable JSON without exposing credentials or paths. * Cross-platform: Linux, macOS, WSL2, and Windows (core PTY tools).

Which integrations are available for this server?

Allows spawning and interacting with Composer agents via persistent terminal sessions, enabling AI-driven orchestration of Composer's coding capabilities.

aiterm-mcp

by kitepon

Overview Schema Related Servers Score Discussions

JavaScript

Local

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

Aiterm

npm weekly downloads node license: MIT

(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

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

Sixteen tools: six PTY tools — pty_open / pty_send / pty_read / pty_key / pty_close / pty_list — to open, drive, and read one persistent terminal; one canonical agent launcher, agent_launch, which selects claude-code, codex-cli, grok-cli, or cursor-cli as the execution harness; agent_steer for an active Codex or Grok turn; four deprecated launcher aliases kept for migration; agent_configure; 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. Grok Composer is a Grok CLI model preset, not another harness: use harness: "grok-cli", model: "grok-composer-2.5-fast". The old four 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.

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: every send to an agent session is a non-blocking dispatch, completion is one universal aiterm-wait waiter whose exit codes mirror the receipt outcome (0=done / 3=timeout, not finished / 4=closed / 5=running for a zero-time observation), and a launch with an initial prompt returns a ready-made wait_command in its structured receipt. 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 v0.29.30 · 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 npm install -g aiterm-mcp@latest. To roll back, install a known-good immutable version, for example npm install -g "aiterm-mcp@<known-good-version>", then restart the MCP client. 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.

Two ways to use it

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.

The human-readable launch text is accompanied by an aiterm.agent-launch-result.v1 structured receipt containing the canonical harness; the old provider field remains for compatibility. The same harness is carried by agent dispatch, aiterm-wait, agent_configure, and agent rows in pty_list, while their old vendor/provider/agent fields remain compatibility fields. Codex completion comes from its normal durable rollout transcript, Grok CLI from its normal session events, Claude Code from a launch-specific Stop hook settings addition, and Cursor from its normal agent transcript's terminal turn_ended record. Sending to any agent session is a non-blocking dispatch — the call returns immediately with an opaque, harness-specific integer event_cursor, and completion arrives via aiterm-wait. The Cursor adapter translates submit into the current CLI's extended keyboard protocol. If submitted text remains in Cursor's composer, its dispatch fails instead of returning a successful receipt.

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.

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
$ aiterm-wait --session codex1 --cursor <event_cursor>   # never in the parent's foreground; exit 0=done, 3=timeout (not done), 4=closed
pty_read("codex1", { agent_transcript: true })           → collect the full answer

The canonical harness choices are:

`harness`	Launches	Notes
`claude-code`	Claude Code CLI	Claude model and effort controls; correlated Stop hook
`codex-cli`	Codex CLI	OpenAI model and effort controls; durable rollout completion
`grok-cli`	Grok Build CLI	Grok or Composer model selected with `model`; live catalog check
`cursor-cli`	Cursor Agent CLI	GPT, Claude, Grok, or another Cursor catalog model; normal transcript completion

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. There is no implicit whole-environment copy, backend-server restart, retry, or fallback. 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, folded by the per-command reducer to a count header plus just the hits:

→ pty_send("demo", "grep -rn capture-pane src/ test/")
→ pty_read("demo", { wait: true, rtk: true })
← 2 matches in 1 files:

  src/core.ts:159:// maxBuffer defaults to 1 MiB; capture-pane (large scrollback) … (line truncated here)
  src/core.ts:335:const args = ["capture-pane", "-p", "-J", "-t", name];
  [aiterm demo: rtk:grep applied / ~46 tok (raw ~53 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)

Restart Claude Code, then verify the connection:

/mcp        # aiterm should show as connected, exposing 16 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.

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 remains a Grok CLI model preset.
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 · agent_launch · agent_steer · agent_configure · claude_turn · claude_approval<br/>legacy launcher aliases · diagnostics"| S["aiterm-mcp<br/>stdio MCP · 16 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.

Command	Built-in shell	aiterm	Verdict
`git log --oneline -5`	1 call, ~7 s	2 calls, ~13 s	shell (fewer round-trips)
`npm test` (203 tests)	~4,292 tok	~607 tok	aiterm (~7.1× fewer, verdict kept)
`find node_modules -type f`	~500 tok¹	~456 tok	tokens tie; aiterm keeps head and tail + `line_range`
`grep -rn "session" src/`	~2,989 tok	~1,096 tok	aiterm (~2.7×; long lines get clipped²)

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 truncates long lines (~80 chars) and folds the overflow into [+N more], which suits scanning; use the built-in tool when you need every full line.

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.

	aiterm-mcp	one-shot shell MCP(e.g. `mcp-server-commands`)	terminal / SSH / tmux MCPs(e.g. `iterm-mcp`, `ssh-mcp`, `tmux-mcp`)	shared-tmux agent-to-agent(e.g. `smux`)
Persistent session	✅ tmux / psmux, survives restarts	❌ new shell every call	⚠️ varies	✅ tmux
SSH / containers / REPLs	nest with one `pty_send`	reconnect every command	⚠️ often separate tools	✅ tmux (human drives)
Launch another agent in one call	✅ `agent_launch(harness=…)`	❌	❌	⚠️ agents join a human-run tmux via a CLI + skills
Headless (no human at a tmux)	✅ MCP-driven, programmatic	✅	⚠️ varies	❌ built around a human in the tmux
MCP-native (any MCP client)	✅ one `claude mcp add`	✅	✅ (they are MCPs)	❌ tmux config + CLI + Agent Skills
Token-reduced reads	✅ per-command reducers	❌ raw output	⚠️ rarely	❌ raw tmux
Completion detection	5-layer: exit / `mark` / `until` / quiescence / timeout	n/a (blocks per call)	⚠️ prompt-match, fragile	❌ agent reads the pane
Human can co-drive	✅ shared socket / namespace (`attach`)	❌	⚠️ varies	✅ (its core model)

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:

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.
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.
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

Tool	Role	Key args
`pty_open`	Grab one terminal, return a `session_id`	`name?`, `shell="bash"`
`pty_send`	Send text; on an agent session this is a non-blocking dispatch returning an `event_cursor`	`session_id`, `text`, `enter=true`, `mark`, `force`, `rtk`, `raw`
`pty_read`	Read output, token-reduced (incremental by default)	`session_id`, `wait`, `until`, `until_regex`, `timeout`, `screen`, `full`, `lines`, `line_range`, `raw`, `rtk`, `agent_transcript`, `operation_id`
`pty_key`	Send a control key	`session_id`, `key` (`C-c`/`Enter`/`Up`…)
`pty_close`	Close idempotently; return `closed` / `already_closed`	`session_id`
`pty_list`	List sessions (agent rows carry canonical `harness=<id>` plus compatibility `agent=<kind>`)	(none)
`agent_launch`	Canonical agent launch; harness and model are independent	`harness`, `prompt?`, `model?`, `reasoning_effort?`, `cwd?`, `write_scope?`, `throughline_source_session?`, `throughline_supplement_file?`
`agent_steer`	Inject text into the active Codex or Grok turn; return `idle` without sending when no turn is active	`session_id`, `text`
`claude_agent` / `codex_agent` / `grok_agent` / `composer_agent`	Deprecated compatibility aliases	legacy launcher arguments
`agent_configure`	Change model/effort in a running Claude, Codex, Grok, Composer, or Cursor session without restarting it	`session_id`, `model?`, `reasoning_effort?`
`claude_turn`	Issue (dispatch-only) or recover one correlated Claude operation	`action`, `session_id`, `operation_id`, `text?`
`claude_approval`	Inspect or answer the current correlated Claude approval prompt	`action`, `session_id`, `operation_id?`, `approval_choice?`, `observed_prompt_digest?`
`diagnostics`	Read-only factory readiness as machine-readable JSON	(none)

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.

Local runtime error snapshot

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.

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? }) changes a running Claude, Codex, Grok, Composer, or Cursor TUI through the harness's standard controls, preserving the PTY and conversation context.

`harness`	Launches	Model behavior
`claude-code`	Claude Code CLI	Claude catalog model; native effort controls
`codex-cli`	Codex CLI	OpenAI catalog model; native effort controls
`grok-cli`	Grok Build CLI	Grok/Composer catalog model; Composer is `model: "grok-composer-2.5-fast"`
`cursor-cli`	Cursor Agent CLI	Cursor catalog model, including GPT/Claude/Grok; effort uses model parameter override

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.

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/Composer return 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. Missing or ambiguous attribution remains an explicit error.

Completion detection (5 layers)

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. 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/Composer observe 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`)

As of v0.16 a parent agent never blocks on aiterm — there is no wait parameter anywhere (v0.17 makes the waiter's exit codes mirror its outcome). The whole flow is dispatch + one universal waiter:

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.
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).
The parent never runs the waiter in its own foreground. Waiting is correct — but the waiter is a separate process, not the parent's turn. A harness that re-invokes its agent when a background task exits (Claude Code) runs the waiter in the background and gets woken with zero polling. So that this is not left to interpretation, aiterm reads clientInfo.name from the MCP initialize handshake and its receipts name the concrete invocation for the detected host — for Claude Code, literally Bash(command: "aiterm-wait …", run_in_background: true). Unknown or undeclared hosts get the generic "start it as a process that does not block the parent's turn" wording; nothing else about the contract changes. Every receipt leads with the same rule: dispatch and let go, then go do something else or end the turn.
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.

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).
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

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 is in the foreground is encoded as one newline-free eval input: 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 use the tmux-compatible bracketed-paste operation (paste-buffer -p): panes that requested bracketed-paste mode receive each chunk wrapped in ESC[200~/201~, 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

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.

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 0.42.0 on test counts, the rule line, 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 0.42.0 truncates the reason at the first " - " — a readability choice, so those lines are intentionally not byte-identical to rtk. The [full output: …] tee-pointer line rtk appends on large output is not reproduced on the read side.
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

Development uses focused local tests first. GitHub Actions runs the suite on the self-hosted linux-workstation runner for every push, adds windows-native only when Windows-specific files change, and runs all three runners (macos-native, linux-workstation, windows-native) once a week as a health check. 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 reducer is ported to match upstream rtk 0.42.0, except the deliberate FAILED-line difference noted above, and is locked by regression tests).

Try it

One command, no clone, no build:

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

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 still comes from the persistent shell session. When a caller needs a value that belongs to the current MCP process rather than the older persistent multiplexer server, every harness accepts env_vars: ["NAME", ...]. Only those names are refreshed at launch; this is a narrow per-launch overlay, not a replacement environment or configuration snapshot.

License

MIT

Available Tools

16 tools

agent_configureA

起動済みのClaude／Codex／Grok／Composer／Cursor agent sessionを再起動せず、会話contextを保ったままmodel／reasoning effortを変更する。各harnessのCLI標準model操作を使う。Cursorのreasoning_effort変更はmodelと同時指定する。

ParametersJSON Schema

Name	Required	Description
`model`	No	変更後のmodel。省略時はmodelを変更しない
`session_id`	Yes
`reasoning_effort`	No	変更後のreasoning effort。省略時はeffortを変更しない

Output Schema

ParametersJSON Schema

Name	Required	Description
`model`	Yes
`schema`	Yes
`harness`	Yes
`provider`	Yes
`session_id`	Yes
`reasoning_effort`	Yes

TDQS

A4.2/5.0

Behavior4/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

With no annotations provided, the description carries the burden of behavioral disclosure. It explains that the operation is non-restarting, preserves conversation context, and mutates model/reasoning settings, with a useful Cursor-specific constraint. It does not mention failure modes or side effects, but the core behavior is transparent.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness5/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is compact and front-loaded: it states the core behavior first, then the method, then the special Cursor caveat. Every sentence adds information and there is no filler.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness4/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

Given three parameters and available output schema, the description is sufficiently complete. It explains the operation context, the non-restart behavior, and the key harness-specific coupling. It could add a bit more guidance on failure conditions or where to get valid model identifiers, but the essential calling context is present.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters3/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

The schema already documents model and reasoning_effort with 'omit means no change' semantics, and 67% of parameters have descriptions. The description adds the cross-parameter Cursor caveat that reasoning_effort must be changed together with model, but it does not provide value formats or allowed model identifiers beyond what the schema implies.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description clearly identifies the action (change model/reasoning effort), the resource (already-started agent sessions for multiple harnesses), and the key constraint (no restart, context preserved). This distinguishes it from agent_launch and other session-oriented siblings.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines4/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

It gives clear usage context: target existing sessions, avoid restarting, and use each harness's standard CLI model operation. It does not explicitly name alternatives or state when not to use the tool, but '起動済み' and '再起動せず' effectively separate it from launch/restart flows.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

agent_launchA

エージェントを単一の標準入口から永続sessionへ起動する。harnessはagent loop・認証・hook・transcriptを所有する実行基盤、modelはそのharnessが選ぶ推論モデルであり別軸。Cursor harnessからGPT／Claude／Grok等を選んでも完了相関はCursor方式のまま。Grok Composerは別harnessではなく harness=grok-cli と model=grok-composer-2.5-fast で指定する。通常CLIと同じHOME・cwd・project/user/local設定・MCP・plugin・skill・permission/trustを共有する。aitermは完了相関stateだけをlaunch単位で所有する。起動されたagentにはsub-agent自己認識、親session、delegation depth/lineage、delegation_allowed=trueを注入し、必要な追加委譲は許可する。起動して投げたら投げっぱなしでよい＝親はここで待たない。完了通知は起動応答またはpty_send dispatch receiptの wait_processを、親のターンを塞がない別プロセスAPIへexecutable／argsの境界を保ったまま渡して受ける（PowerShell 7のStart-Processはwindows_start_process_argument_listを使う）（exit 0=done / 3=timeout（既定600秒・未完了） / 4=closed。receiptのoutcomeが正で、done以外は未完了。ポーリング不要・foreground実行はしない）。wait_commandは人間向け互換表示。結果回収は pty_read(agent_transcript:true)。

ParametersJSON Schema

Name	Required	Description
`cwd`	No	作業ディレクトリ（絶対パス・任意）
`model`	No	harnessが選ぶモデル。provider名ではなくlive catalog上のmodel ID
`prompt`	No	起動時に渡す初手プロンプト（任意）。送信後は待たずに即返る
`harness`	Yes	agent loop・session・hook・transcript・認証を所有する実行基盤
`env_vars`	No	現在のMCP processから継承する環境変数名
`write_scope`	No	能力宣言。read-onlyは対応harnessの標準read-only面で実効禁止する
`session_name`	No	Aiterm session名（省略で自動採番）
`reasoning_effort`	No	harness adapterが標準CLI表現へ変換する思考レベル。Cursorではmodel同時指定が必要
`launch_operation_id`	No	Claude Codeのpromptなしexact replay相関だけで使用
`throughline_source_session`	No	同一端末のThroughline sessionから読み取り専用contextを初手へ注入する
`throughline_supplement_file`	No	Throughline 0.10.8以降へそのまま渡すproject束縛済み長期記憶・知識の補足JSON path

Output Schema

ParametersJSON Schema

Name	Required	Description
`schema`	Yes
`harness`	Yes
`provider`	Yes	旧互換field。新規連携はharnessを使う
`session_id`	Yes
`write_scope`	No
`event_cursor`	Yes
`wait_command`	Yes
`wait_process`	Yes
`submit_residue`	Yes
`managed_completion`	Yes
`write_scope_enforcement`	No

TDQS

A4.4/5.0

Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

With no annotations present, the description carries full behavioral disclosure burden. It comprehensively details the asynchronous fire-and-forget behavior, the shared environment (HOME, cwd, MCP, plugins), the injection of sub-agent self-awareness and delegation_allowed=true, the exact meaning of exit codes (0, 3, 4), the default timeout of 600 seconds, no polling/no foreground execution, and the return value collection via pty_read(agent_transcript:true). No contradictions exist.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is lengthy but information-dense; every sentence contributes to understanding the tool's behavior and architecture. The core purpose is front-loaded, followed by clarifying details. Some minor over-specification occurs (e.g., the PowerShell 7 Start-Process note) but it serves to clarify platform-specific behavior. Overall, it earns its length, though slightly more focus would improve conciseness.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

Given the tool's complexity, the description is remarkably complete. It covers the launch flow, async behavior, completion mechanisms, exit codes, environment sharing, permission injection, and result retrieval. Although there is a declared output schema (not shown), the description also explains the receipt outcome semantics, making the agent fully equipped to invoke and interpret results without gaps.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters3/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema description coverage is 100%, so all 11 parameters are documented in the input schema itself. The description adds conceptual context (e.g., model is a separate axis from harness), but does not provide per-parameter elaboration beyond what the schema already offers. The baseline of 3 is appropriate since the schema does the heavy lifting.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description explicitly states the tool's action: 'エージェントを単一の標準入口から永続sessionへ起動する' (launch agents to persistent sessions from a single standard entry point). It clearly distinguishes the conceptual axes of harness vs model, and explains how Grok Composer is specified via harness=grok-cli, which differentiates it from potential sibling-specific launch tools.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines4/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

The description provides clear usage context: this is the single standard entry point for launching any agent, including via specific harnesses like Cursor or Grok Composer. It details the fire-and-forget pattern (parent does not wait), how completion notifications are received via wait_process or dispatch receipt, and explains exit codes and result retrieval via pty_read. It does not explicitly name alternative tools, but the '単一の標準入口' phrasing strongly implies it is the primary launch tool.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

agent_steerA

実行中のCodex/Grok agentへ追加メッセージを差し込み、現在のターンを誘導する。独立した次ターンを始める用途ではなく、idle時は文字を送らずdelivery=idleを返す。

ParametersJSON Schema

Name	Required	Description	Default
`text`	Yes	現在のターンへ追加する文字列。UTF-8で最大64KiB
`session_id`	Yes

Output Schema

ParametersJSON Schema

Name	Required	Description
`schema`	Yes
`vendor`	Yes
`harness`	Yes
`delivery`	Yes
`launch_id`	Yes
`session_id`	Yes

TDQS

A4.4/5.0

Behavior4/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

With no annotations provided, the description carries the full burden. It discloses the key behavior: it injects a message into the active turn and returns delivery=idle without sending characters when idle. It could add more detail about side effects or error conditions, but the core behavioral traits are covered.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness5/5

Is the description appropriately sized, front-loaded, and free of redundancy?

Two compact sentences in Japanese: the first states the purpose, the second adds the critical exclusions and idle behavior. No filler or redundant phrasing; each sentence carries essential information.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness4/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

For a simple two-parameter tool with an output schema, the description covers purpose, usage constraints, and a specific return behavior (delivery=idle). It lacks details on session_id semantics and edge cases, but the core operational guidance is sufficient.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters3/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

The schema already describes 'text' well (string to add to the current turn, UTF-8 max 64KiB), but 'session_id' is undocumented. The description's phrase 'additional message' aligns with 'text' but adds no new parameter-level meaning, and it does not compensate for the missing session_id documentation.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description states a specific action: inserting an additional message into a running Codex/Grok agent to steer the current turn. It clearly distinguishes itself from starting an independent next turn, which separates it from launch/turn tools in the sibling list.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines5/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

It explicitly says when to use (while an agent is running, to steer the current turn) and when not to use (when idle or wanting to start a new turn). The idle behavior with 'delivery=idle' further clarifies the intended usage window.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

claude_agentA

【旧互換alias。新規連携は agent_launch(harness=claude-code)】Claude Codeの対話エージェントTUIを永続端末に起動する。claude -pではなく、同じ利用者可視sessionへpty_sendで継続入力する。通常CLIと同じHOME・cwd・project/user/local設定・MCP・plugin・skill・permission/trustを共有する。aitermは完了相関stateだけをlaunch単位で所有する。起動されたagentにはsub-agent自己認識、親session、delegation depth/lineage、delegation_allowed=trueを注入し、必要な追加委譲は許可する。通常settingsへlaunch固有Stop hook settingsを加算する。起動前に共有認証を構造化確認し、未認証ならsessionを作らない。起動して投げたら投げっぱなしでよい＝親はここで待たない。完了通知は起動応答またはpty_send dispatch receiptの wait_processを、親のターンを塞がない別プロセスAPIへexecutable／argsの境界を保ったまま渡して受ける（PowerShell 7のStart-Processはwindows_start_process_argument_listを使う）（exit 0=done / 3=timeout（既定600秒・未完了） / 4=closed。receiptのoutcomeが正で、done以外は未完了。ポーリング不要・foreground実行はしない）。wait_commandは人間向け互換表示。結果回収は pty_read(agent_transcript:true)。Claude の durable turn は claude_turn でも回収できる。

ParametersJSON Schema

Name	Required	Description
`cwd`	No	作業ディレクトリ（対象リポのルート等・任意）
`model`	No	起動モデル（例: claude-sonnet-4-6）。省略時はClaude CLI既定
`prompt`	No	起動時に渡す初手プロンプト（任意）。送信後は待たずに即返る
`env_vars`	No	起動したagentへ現在のMCP processから継承する環境変数名。値はtool引数へ渡さない
`session_name`	No	セッション名（省略で自動採番）
`reasoning_effort`	No	Claude Code reasoning effort。low/medium/high/xhigh/max。省略時はCLI既定
`launch_operation_id`	No	promptなしClaude launchのexact replay相関ID。session_name必須
`throughline_source_session`	No	同一端末のThroughline sessionから所有権を変えずに記憶を読み、promptのmissionより前へ注入する
`throughline_supplement_file`	No	Throughline 0.10.8以降へそのまま渡すproject束縛済み長期記憶・知識の補足JSON path

Output Schema

ParametersJSON Schema

Name	Required	Description
`schema`	Yes
`harness`	Yes
`provider`	Yes
`session_id`	Yes
`event_cursor`	Yes
`wait_command`	Yes
`wait_process`	Yes
`submit_residue`	Yes
`managed_completion`	Yes	後方互換field。trueはaiterm完了相関が有効という意味で、project/user環境の隔離を意味しない

TDQS

A4.6/5.0

Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

With no annotations provided, the description carries full behavioral burden and does so thoroughly: fire-and-forget launch, shared HOME/cwd/settings/MCP/plugin/permissions, injected delegation metadata, Stop-hook additions, pre-launch auth verification, and exact exit codes 0/3/4 with the rule that only exit 0 means done. This is unusually complete operational disclosure.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is long but dense; every sentence carries operationally relevant information and the alias/redirect note is front-loaded. Some parenthetical chains could be formatted more cleanly, but there is little redundancy for a tool with this level of behavioral complexity.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

Given 9 parameters, no required parameters, and an output schema, the description still covers launch semantics, async handoff, completion receipts, auth preconditions, result collection, and sibling routing. With the output schema covering return structure, almost nothing an agent needs to invoke this correctly is missing.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters3/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema description coverage is 100%, so every parameter is already documented in the schema. The description adds general context such as cwd/HOME/settings inheritance but does not attach new meaning to individual parameters like throughline_source_session or launch_operation_id. Baseline 3 is appropriate given that the schema does the heavy lifting.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description opens by identifying itself as an old-compatibility alias and redirects new integrations to agent_launch(harness=claude-code), then states it launches the Claude Code interactive-agent TUI on a persistent terminal. This is a specific verb+resource+scope framing that clearly distinguishes it from siblings and from claude -p.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines5/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

It gives explicit when/when-not guidance: not claude -p, not a blocking wait, no polling, no foreground execution; continue input via pty_send, retrieve results via pty_read(agent_transcript:true) or claude_turn, and use agent_launch for new integrations. It also explains the async completion path and exit-code meanings.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

claude_approvalA

aiterm相関付きClaudeのactive turn中に表示された権限確認UIを、turn相関を保ったまま検査・応答する専用面。inspectで画面digestと安全な単発Yes/Noだけを取得し、respondは同じoperation・同じdigestが現在も表示中の場合だけ送信する。

ParametersJSON Schema

Name	Required	Description
`action`	Yes
`session_id`	Yes
`operation_id`	No	durable operationのID。通常pty_send由来の匿名turnでは省略する
`approval_choice`	No	respondだけに指定する
`observed_prompt_digest`	No	直前のinspectが返したdigest。respondだけに指定する

Output Schema

ParametersJSON Schema

Name	Required	Description
`at`	Yes
`action`	Yes
`schema`	Yes
`status`	Yes
`choices`	Yes
`session_id`	Yes
`operation_id`	Yes
`prompt_digest`	Yes
`selected_choice`	Yes

TDQS

A4.1/5.0

Behavior4/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

With no annotations, the description carries the full burden. It discloses that inspect returns only the screen digest and safe single Yes/No, and that respond is conditional on matching operation and current digest. This provides meaningful behavioral context, though it does not cover all error cases or side effects of responding.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is a single dense sentence that efficiently conveys the main purpose, actions, and safety condition without wasted words. It is compact and front-loaded, though the multiple clauses make it slightly complex to parse.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness4/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

An output schema exists, so return values need not be described. The description covers the two actions, what inspect returns, and the precondition for respond, which is sufficient for a specialized approval tool. It lacks details on mismatch handling or how to obtain session_id, but these are partially addressed by the schema hints.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters3/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema description coverage is 60%, and the schema already documents the optional parameters (operation_id, approval_choice, observed_prompt_digest) with useful descriptions. The tool description adds context for the action enum (inspect vs respond) but does not elaborate on individual parameters beyond what the schema provides. It adds some value but not extensive.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description clearly states the tool's function as a dedicated surface for inspecting and responding to permission confirmation UI during an active Claude turn, preserving turn correlation. It distinguishes itself from sibling PTY tools by focusing on the approval context and the specific inspect/respond actions.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines4/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

The description specifies the intended scenario (permission UI during active Claude turn) and the safe usage pattern: inspect only retrieves a digest and simple Yes/No, while respond is only sent when the same operation and digest are still displayed. It implies this is the correct tool for approvals but does not explicitly name alternatives like pty_send or state exclusions.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

claude_turnC

aiterm相関付きClaude sessionのdurable operationを構造化issue／recoverするmachine-caller専用面。pending／unknown／completedを人間向けerror文字列の解析なしで返し、Observer固有ロジックは持たない。

ParametersJSON Schema

Name	Required	Description
`text`	No	issueだけに指定するbounded turn本文
`action`	Yes
`session_id`	Yes
`operation_id`	Yes

Output Schema

ParametersJSON Schema

Name	Required	Description
`action`	Yes
`reason`	Yes
`schema`	Yes
`status`	Yes
`raw_output`	Yes
`session_id`	Yes
`operation_id`	Yes
`submit_residue`	Yes

TDQS

C2.9/5.0

Behavior3/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Without annotations, the description carries the behavioral burden. It discloses that the tool returns structured statuses (pending/unknown/completed) and that it does not contain Observer-specific logic, which is useful. Still, it does not explain the side effects or durability semantics of 'issue' and 'recover' actions, or any preconditions.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness3/5

Is the description appropriately sized, front-loaded, and free of redundancy?

It is a single sentence, so it is compact, but the density of technical terms (aiterm, durable operation, Observer) without elaboration makes it less readable. It is not verbose, but borderline under-specified rather than genuinely concise.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness2/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

Given no annotations, four parameters, and an output schema that is not referenced, the description leaves many gaps: what actions are available, what correlates with aiterm, what 'recover' means, and what the returned statuses imply. It gives only a vague overview without enough operational detail for an agent to confidently invoke the tool.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters1/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

The description contains zero mention of the four parameters. Schema coverage is only 25% (only 'text' has a description), yet the description does nothing to compensate for the missing meaning of 'action', 'session_id', or 'operation_id'.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose4/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description clearly states a specific function: 'issue/recover durable operations on Claude sessions' for machine callers. It also notes the structured status returns (pending/unknown/completed) and the lack of Observer-specific logic, helping it stand apart from agent-focused sibling tools like claude_agent. However, the heavy use of jargon (aiterm, durable operation) slightly detracts from immediate clarity.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines3/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

The phrase 'machine-caller dedicated surface' implies its intended audience, and the statement about returning pending/unknown/completed 'without parsing human-oriented error strings' suggests a use case. Yet there is no explicit 'when to use' or comparison with alternative tools, leaving guidance largely implicit.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

codex_agentA

【旧互換alias。新規連携は agent_launch(harness=codex-cli)】Codexの対話エージェント TUI を永続端末に起動する。実装・レビュー・調査を対話で回す。通常CLIと同じHOME・cwd・project/user/local設定・MCP・plugin・skill・permission/trustを共有する。aitermは完了相関stateだけをlaunch単位で所有する。起動されたagentにはsub-agent自己認識、親session、delegation depth/lineage、delegation_allowed=trueを注入し、必要な追加委譲は許可する。委譲契約を使う完全な呼び出し例: codex_agent({"prompt":"<依頼>","model":"gpt-5.6-sol","reasoning_effort":"high","cwd":"/absolute/path/to/repo","write_scope":"read-only"})。turn は pty_send で送る（自動で非ブロック dispatch になる）。起動して投げたら投げっぱなしでよい＝親はここで待たない。完了通知は起動応答またはpty_send dispatch receiptの wait_processを、親のターンを塞がない別プロセスAPIへexecutable／argsの境界を保ったまま渡して受ける（PowerShell 7のStart-Processはwindows_start_process_argument_listを使う）（exit 0=done / 3=timeout（既定600秒・未完了） / 4=closed。receiptのoutcomeが正で、done以外は未完了。ポーリング不要・foreground実行はしない）。wait_commandは人間向け互換表示。結果回収は pty_read(agent_transcript:true)。model / reasoning_effort を引数で指定可（省略時は端末 config／CLI 既定を継承。実効値は起動応答に明示）。

ParametersJSON Schema

Name	Required	Description
`cwd`	No	作業ディレクトリ（対象リポのルート等・任意）
`model`	No	起動モデル（例: gpt-5.6-sol / gpt-5.6-terra / gpt-5.6-luna）。省略時は端末 config／CLI 既定を継承（端末側のピンがそのまま効く。実効値は起動応答に明示される）
`prompt`	No	起動時に渡す初手プロンプト（任意）。送信後は待たずに即返る
`env_vars`	No	起動したagentへ現在のMCP processから継承する環境変数名。値はtool引数へ渡さない
`write_scope`	No	能力宣言。read-only、または書込みを許可するパスの説明文字列。対応harnessのread-onlyはCLI標準のread-only面で実効禁止する
`session_name`	No	セッション名（省略で自動採番）
`reasoning_effort`	No	reasoning effort（思考レベル）。low/medium/high/xhigh/max/ultra（CLI／model 版依存）。ultra は max 推論＋proactive 自動委譲 ON＝使用量急増注意（明示要求時のみ）。省略時は端末 config／CLI 既定。
`throughline_source_session`	No	同一端末のThroughline sessionから所有権を変えずに記憶を読み、promptのmissionより前へ注入する
`throughline_supplement_file`	No	Throughline 0.10.8以降へそのまま渡すproject束縛済み長期記憶・知識の補足JSON path

Output Schema

ParametersJSON Schema

Name	Required	Description
`schema`	Yes
`harness`	Yes
`provider`	Yes
`session_id`	Yes
`write_scope`	No
`event_cursor`	Yes
`wait_command`	Yes
`wait_process`	Yes
`submit_residue`	Yes
`managed_completion`	Yes	後方互換field。trueはaiterm完了相関が有効という意味で、project/user環境の隔離を意味しない
`write_scope_enforcement`	No

TDQS

A4.8/5.0

Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

没有annotations，描述承担了全部行为披露责任。描述详尽展示了工具的行为模式：TUI启动方式、共享配置、注入的自我认知与委托权限、turn发送机制、完成通知的退出码（0/3/4）、结果获取路径、模型参数继承规则等。这些远超基本要求，没有任何隐藏行为。

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

描述较长，但信息密度高，结构层次分明：先身份，再功能，然后共享配置、行为细节、调用示例、退出码、结果接收。开头即点明新老替代关系，有前置性。虽篇幅偏长，但对一个复杂交互式代理启动工具而言，每句都有实际价值。

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

工具复杂度高（9参数、有输出schema），描述覆盖了启动、配置共享、委托链、退出码、结果获取、模型指定方式、调用示例，甚至包含了与pty_send/pty_read的交互模式。唯一未提的可能需要预先pty_open，但可通过兄弟工具推断。对于该工具，上下文已相当齐全。

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters4/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema覆盖100%，每个参数都有基础描述。描述额外增加了重要语义：prompt发送后即返、write_scope的read-only实际强制方式、model/reasoning_effort的继承及实効値确认、throughline注入等。虽然描述没有逐个重复参数，但补充了关键行为细节，超越了schema基线。

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

描述第一句即明确该工具是旧兼容alias，新集成应使用agent_launch，随后说明其核心功能是启动Codex对话代理TUI到持久终端，并用于实现、审查、调查。动词（起動）、资源（Codex TUI）和用途清晰，且与兄弟工具agent_launch的区分直接点明，不存在混淆。

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines5/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

描述明确指出新規連携は agent_launch(harness=codex-cli)，给出了当not-to-use的条件和替代方案。同时通过调用示例和说明其适用场景（実装・レビュー・調査），提供了何时使用的上下文。使用指南完整且无歧义。

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

composer_agentA

【旧互換alias。新規連携は agent_launch(harness=grok-cli, model=grok-composer-2.5-fast)】Grok BuildのComposerモデルを永続端末に起動する。通常CLIと同じHOME・cwd・project/user/local設定・MCP・plugin・skill・permission/trustを共有する。aitermは完了相関stateだけをlaunch単位で所有する。起動されたagentにはsub-agent自己認識、親session、delegation depth/lineage、delegation_allowed=trueを注入し、必要な追加委譲は許可する。turn は pty_send で送る（自動で非ブロック dispatch になる）。起動して投げたら投げっぱなしでよい＝親はここで待たない。完了通知は起動応答またはpty_send dispatch receiptの wait_processを、親のターンを塞がない別プロセスAPIへexecutable／argsの境界を保ったまま渡して受ける（PowerShell 7のStart-Processはwindows_start_process_argument_listを使う）（exit 0=done / 3=timeout（既定600秒・未完了） / 4=closed。receiptのoutcomeが正で、done以外は未完了。ポーリング不要・foreground実行はしない）。wait_commandは人間向け互換表示。結果回収は pty_read(agent_transcript:true)。model／reasoning_effortを引数で指定可。live catalogにComposer modelがなければGrokへfallbackせず明示エラー。read-only sandboxとagent_configureに対応。

ParametersJSON Schema

Name	Required	Description
`cwd`	No	作業ディレクトリ（対象リポのルート等・任意）
`model`	No	起動モデル。省略時は grok-composer-2.5-fast。既定／explicit modelを起動前にlive catalogへ照合し、不在ならfallbackせずエラー
`prompt`	No	起動時に渡す初手プロンプト（任意）。送信後は待たずに即返る
`env_vars`	No	起動したagentへ現在のMCP processから継承する環境変数名。値はtool引数へ渡さない
`write_scope`	No	能力宣言。read-only、または書込みを許可するパスの説明文字列。対応harnessのread-onlyはCLI標準のread-only面で実効禁止する
`session_name`	No	セッション名（省略で自動採番）
`reasoning_effort`	No	Grok Build reasoning effort。利用可能値はCLI／modelのlive catalogに従う。省略時はCLI／model既定。
`throughline_source_session`	No	同一端末のThroughline sessionから所有権を変えずに記憶を読み、promptのmissionより前へ注入する
`throughline_supplement_file`	No	Throughline 0.10.8以降へそのまま渡すproject束縛済み長期記憶・知識の補足JSON path

Output Schema

ParametersJSON Schema

Name	Required	Description
`schema`	Yes
`harness`	Yes
`provider`	Yes
`session_id`	Yes
`write_scope`	No
`event_cursor`	Yes
`wait_command`	Yes
`wait_process`	Yes
`submit_residue`	Yes
`managed_completion`	Yes	後方互換field。trueはaiterm完了相関が有効という意味で、project/user環境の隔離を意味しない
`write_scope_enforcement`	No

TDQS

A4.8/5.0

Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

With no annotations provided, the description carries the full behavioral burden and delivers thoroughly. It discloses non-blocking dispatch, that the parent does not wait, completion semantics with exit codes (0/3/4), no polling, no fallback to Grok, and environment/settings sharing. It also covers agent injection details and read-only sandbox support, leaving little behavior implicit.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is long and dense, but it is front-loaded with the most important routing information and every sentence adds operational value. Some points are reiterated for emphasis, and the density makes it harder to parse quickly, but the length is justified by the tool's complexity and the absence of annotations.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

The description covers the full lifecycle: launch, parameter handling, async dispatch, completion notification via another process API, exit-code interpretation, result retrieval through pty_read, and error behavior. Since an output schema exists, return-value details need not be in the description. Nothing essential for correct invocation is missing.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters4/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema coverage is 100%, so the baseline is 3. The description adds meaningful semantic context beyond the schema, especially for model (validated against live catalog with explicit error, no fallback), reasoning_effort (CLI/model-dependent), and the async dispatch relationship between prompt and pty_send. It does not elaborate on every parameter, but the schema already covers those adequately.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description clearly states the tool's function: launching the Grok Build Composer model on a persistent terminal. It also explicitly identifies itself as a legacy-compatibility alias and names agent_launch as the replacement, distinguishing it from sibling tools without requiring schema inspection.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines5/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

The description explicitly says to use agent_launch for new integrations, reserving composer_agent for legacy compatibility. It further explains the full usage flow: send turns via pty_send, receive completion via wait_process, retrieve results via pty_read, and avoid polling or foreground execution. This gives an agent clear when-to-use and when-not-to-use guidance.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

diagnosticsA

Factory 向け read-only 診断。安全な状態語彙だけを機械可読 JSON で返す（PTY 内容・認証情報・path・環境値は返さない）。

ParametersJSON Schema

Name	Required	Description	Default
No parameters

TDQS

A4.3/5.0

Behavior4/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

The description discloses read-only nature and explicitly states what data is not returned (PTY content, credentials, paths, environment values). Given no annotations, this provides sufficient behavioral transparency for a zero-parameter diagnostic tool.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness5/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is two concise sentences with no wasted words. The key point ('read-only diagnostics returning safe vocabulary') is front-loaded.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

For a zero-parameter, no-output-schema tool with no annotations, the description is complete. It clearly defines scope, safety guarantees, and exclusions.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters4/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

No parameters exist (schema coverage 100%), so the description does not need to add parameter meaning. The baseline of 4 applies as it adds no confusion.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description clearly states 'Factory 向け read-only 診断' (Factory-oriented read-only diagnostics) and specifies that it returns only safe vocabulary in machine-readable JSON, explicitly listing excluded data types. This distinguishes it from sibling tools like pty_open or pty_list.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines3/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

The description implies usage for Factory contexts via '向け' but does not provide explicit when-to-use or when-not-to-use guidance, nor does it name alternative tools. Usage context is implied but not fully elaborated.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

grok_agentA

【旧互換alias。新規連携は agent_launch(harness=grok-cli)】Grok BuildのGrokモデル（既定 grok-4.6）の対話エージェント TUIを永続端末に起動する。通常CLIと同じHOME・cwd・project/user/local設定・MCP・plugin・skill・permission/trustを共有する。aitermは完了相関stateだけをlaunch単位で所有する。起動されたagentにはsub-agent自己認識、親session、delegation depth/lineage、delegation_allowed=trueを注入し、必要な追加委譲は許可する。turn は pty_send で送る（自動で非ブロック dispatch になる）。起動して投げたら投げっぱなしでよい＝親はここで待たない。完了通知は起動応答またはpty_send dispatch receiptの wait_processを、親のターンを塞がない別プロセスAPIへexecutable／argsの境界を保ったまま渡して受ける（PowerShell 7のStart-Processはwindows_start_process_argument_listを使う）（exit 0=done / 3=timeout（既定600秒・未完了） / 4=closed。receiptのoutcomeが正で、done以外は未完了。ポーリング不要・foreground実行はしない）。wait_commandは人間向け互換表示。結果回収は pty_read(agent_transcript:true)。model／reasoning_effortを引数で指定可。read-only sandboxとagent_configureに対応。

ParametersJSON Schema

Name	Required	Description
`cwd`	No	作業ディレクトリ（対象リポのルート等・任意）
`model`	No	起動モデル。省略時は grok-4.6。explicit modelを起動前にlive catalogへ照合し、不在ならfallbackせずエラー
`prompt`	No	起動時に渡す初手プロンプト（任意）。送信後は待たずに即返る
`env_vars`	No	起動したagentへ現在のMCP processから継承する環境変数名。値はtool引数へ渡さない
`write_scope`	No	能力宣言。read-only、または書込みを許可するパスの説明文字列。対応harnessのread-onlyはCLI標準のread-only面で実効禁止する
`session_name`	No	セッション名（省略で自動採番）
`reasoning_effort`	No	Grok Build reasoning effort。利用可能値はCLI／modelのlive catalogに従う。省略時はCLI／model既定。
`throughline_source_session`	No	同一端末のThroughline sessionから所有権を変えずに記憶を読み、promptのmissionより前へ注入する
`throughline_supplement_file`	No	Throughline 0.10.8以降へそのまま渡すproject束縛済み長期記憶・知識の補足JSON path

Output Schema

ParametersJSON Schema

Name	Required	Description
`schema`	Yes
`harness`	Yes
`provider`	Yes
`session_id`	Yes
`write_scope`	No
`event_cursor`	Yes
`wait_command`	Yes
`wait_process`	Yes
`submit_residue`	Yes
`managed_completion`	Yes	後方互換field。trueはaiterm完了相関が有効という意味で、project/user環境の隔離を意味しない
`write_scope_enforcement`	No

TDQS

A4.1/5.0

Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

The description is exceptionally transparent about behavior: it shares HOME/cwd/env, owns only completion-state, injects delegation metadata, sends turns via pty_send as non-blocking, does not wait, defines exit codes (0/3/4), states no polling needed, and explains how to collect results via pty_read. This goes far beyond any annotations, which are absent.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness3/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is information-dense but long and sprawling, packed into one continuous paragraph with mixed Japanese and technical details. While every sentence carries meaningful behavioral info, the lack of clear segmentation or prioritization makes it harder for an agent to quickly extract the key points. Not concise enough nor well-structured for maximum clarity.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

With an output schema present and parameters fully documented in the schema, the description provides extensive behavioral context: lifecycle, delegation, completion signaling, exit codes, and result retrieval. It even addresses compatibility and alternatives. Nothing critical for a caller is missing, making this complete.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters3/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema description coverage is 100%, so the input schema already fully documents all 9 parameters. The description mentions model/reasoning_effort can be specified and that env_vars are inherited, but this mostly restates schema info without adding deeper meaning, so a baseline 3 is appropriate.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose4/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description explicitly states it launches Grok Build's Grok model conversational agent TUI into a persistent terminal, identifying the verb and resource. It also clearly differentiates itself as an old-compatibility alias for agent_launch, which helps distinguish it from siblings, though the alias framing adds a bit of ambiguity about its intended use.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines4/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

The description explicitly says '旧互換alias。新規連携は agent_launch(harness=grok-cli)' meaning new integrations should use agent_launch instead, which provides clear when-not-to-use guidance. It also describes how to interact with the launched agent (via pty_send, no waiting, completion notifications), though it does not contrast with other sibling agent tools beyond agent_launch.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

pty_closeA

セッションを閉じ、ログ／読取位置を破棄する。同じsession_idへの再試行は安全で、closed／already_closedのstructured receiptを返す。

ParametersJSON Schema

Name	Required	Description	Default
`session_id`	Yes

Output Schema

ParametersJSON Schema

Name	Required	Description
`schema`	Yes
`outcome`	Yes
`session_id`	Yes

TDQS

A4.5/5.0

Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

With no annotations provided, the description fully carries the behavioral disclosure burden. It reveals that closing destroys logs and read position (destructive side effects), that retries are idempotent and safe, and that the tool returns a structured receipt with states 'closed' or 'already_closed'. This is rich, honest behavioral transparency.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness5/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is two sentences long, front-loaded with the core action, and every phrase earns its place. It covers purpose, side effects, retry behavior, and return type without any filler.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

Given the tool's simplicity (one required parameter) and the presence of an output schema (structured receipt), the description is sufficiently complete. It covers the action, the destructive side effect, the idempotency behavior, and the return format, leaving no major gaps.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters4/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

The schema only defines session_id as a string with no description (0% coverage). The description compensates by explaining that session_id is used for retry safety, and that the same value will yield a receipt indicating whether it was newly closed or already closed. This adds meaningful semantics beyond the raw schema.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description clearly states the tool's action: 'Close the session and discard logs/read position.' It uses a specific verb ('close') and resource ('session'), and distinguishes itself from sibling tools like pty_open, pty_send, pty_read, and pty_list by focusing solely on the closing lifecycle.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines3/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

The description implies when to use the tool (to close a PTY session) but does not explicitly contrast it with alternatives or provide exclusion criteria. It adds a safety note about retrying with the same session_id, but lacks explicit 'when to use vs. not use' guidance.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

pty_keyA

制御キーを送る（C-c, C-d, Enter, Tab, Up, Down... の別名に対応）。aiterm相関付きClaude sessionではturn相関を守るためC-cだけを許可し、承認UIはclaude_approvalで操作する。

ParametersJSON Schema

Name	Required	Description	Default
`key`	Yes	キー名（例 "C-c", "Enter", "Up"）
`session_id`	Yes

TDQS

A4.6/5.0

Behavior4/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

With no annotations provided, the description carries the full burden of behavioral disclosure. It reveals a significant behavioral trait: in Claude sessions, only C-c is permitted to preserve turn correlation, and that approval actions should be routed to claude_approval. However, it does not discuss error handling or consequences of sending disallowed keys, though the restriction itself is valuable transparency.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness5/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is two sentences, front-loaded with the core action and immediately followed by a key usage rule. Every word serves a purpose: examples clarify the key parameter, and the restriction clarifies boundary conditions. No redundancy or filler.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness4/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

For a simple tool with no output schema and no annotations, the description covers purpose, parameter examples, and usage restrictions, and relates it to sibling tools. It does not explain session_id or return behavior, but these are less critical given the tool's straightforward nature and the presence of sibling tools like pty_list and pty_open that likely establish session context.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters4/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

The schema covers only 50% of parameters (key has a description, session_id does not). The description adds meaning to the key parameter by listing acceptable aliases (C-c, C-d, Enter, Tab, Up, Down) and specifying a contextual restriction (only C-c in Claude sessions), compensating for the schema gap. session_id remains undocumented in both, but its purpose is implied as a session identifier.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description clearly states the tool's function with a specific verb ('送る' / send) and resource ('制御キー' / control keys), and provides concrete examples (C-c, C-d, Enter, Tab, Up, Down) that distinguish it from sibling tools like pty_send (which likely sends arbitrary text) and claude_approval (which handles approval UI).

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines5/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

The description explicitly states when to use this tool and provides an exclusion: 'In aiterm-correlated Claude sessions, only C-c is allowed to maintain turn correlation, and the approval UI is operated with claude_approval.' This names an alternative tool (claude_approval) and clarifies a key restriction, giving clear usage guidance.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

pty_listB

握っているセッション一覧（名前 / 現在の前面コマンド / attach 状態 / サイズ / agent 情報）。

ParametersJSON Schema

Name	Required	Description	Default
No parameters

TDQS

B3.3/5.0

Behavior2/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

No annotations are provided, so the description carries full burden. It does not state if the operation is read-only, destructive, or requires authentication. The list nature implies a safe read, but this is not explicit.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness5/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is a single sentence that efficiently conveys the purpose and output fields. No unnecessary words.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness4/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

Given the tool has no parameters and no output schema, the description is largely complete. It explains what is listed. It could potentially mention the return format, but it's adequate for a simple list.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters4/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

There are no parameters, so the schema is trivial. The description adds value by listing the fields returned in the list, which is meaningful beyond the empty schema.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose4/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description clearly states that the tool lists sessions with specific fields (name, current command, attach status, size, agent info). It is a list operation, distinct from sibling tools like pty_open which likely opens sessions.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines2/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

No usage context or when-to-use guidance is provided. It simply describes what the tool does without mentioning when it should be used versus alternatives like diagnostics or pty_open.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

pty_openA

ローカル永続端末（POSIXはtmux、Windows nativeはpsmux 3.3.8以上）を1個開き、session_id を返す。backend server常駐ゆえ本サーバやクライアントが再起動してもセッションは生存する。リモート操作は専用ツールにせず、開いた端末の中で pty_send(session_id, "ssh host") と打って入る。

ParametersJSON Schema

Name	Required	Description	Default
`name`	No	セッション名（省略時は t1, t2... を自動採番）
`shell`	No	起動シェル（既定 bash）	bash

TDQS

A4.4/5.0

Behavior4/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

With no annotations, the description carries the full burden, and it does well: it discloses that sessions persist across server/client restarts, that the backend is resident, and that platform-specific backends (tmux/psmux) are used. It does not mention cleanup, failure modes, or resource implications, but the main behavioral surprise—persistence—is clearly highlighted.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness5/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is three dense, purposeful sentences: purpose and return value first, persistence semantics second, and a concrete usage example third. There is no filler or redundant restatement.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness4/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

Given the low complexity, no annotations, and no output schema, the description covers the essential invocation contract: what is created, what is returned (session_id), why sessions persist, and how to interact with the terminal afterward. It could add exact response shape and failure conditions, but these are not critical for selecting and invoking the tool correctly.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters3/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

The input schema already documents both parameters (name and shell) with 100% schema_description_coverage, so the baseline is 3. The description adds no additional semantic detail about the parameters themselves, which is acceptable given the schema coverage.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description states the exact operation: opens one local persistent terminal (tmux on POSIX, psmux on Windows) and returns a session_id. It clearly stands apart from siblings like pty_read, pty_send, pty_list, and pty_close by describing the creation action.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines5/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

The description explicitly tells the agent how to perform remote operations: do not use a dedicated tool; instead launch a terminal with pty_open and then run pty_send(session_id, "ssh host") inside it. This gives clear when-to-use and how-to-use guidance relative to the pty_* sibling tools.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

pty_readA

セッションの出力をトークン削減して読む（既定は前回読取位置からの増分）。削減: 制御文字除去 / 反復圧縮 / head+tail 折りたたみ＋復元ヒント＋メタ併記。agent_transcript:true は agent session の直近完了ターンの最終 assistant メッセージを公開されたharness記録から平文で返す。長い回答が screen tail で切れた時の回収用。

ParametersJSON Schema

Name	Required	Description
`raw`	No	削減せず生テキスト
`rtk`	No	直前コマンド別の自前 reducer(git/grep/pytest 等)で縮約
`full`	No	増分でなく全文
`wait`	No	完了まで待つ（dead / mark sentinel 自動検出 / until / 出力静止∧シェル復帰 / timeout）
`lines`	No	末尾 N 行のみ
`until`	No	この文字列が出たら完了とみなす（既定はリテラル部分一致。`$ ` や `[..]` もそのまま探せる）
`screen`	No	描画済みスクリーン(TUI 向け)
`timeout`	No	wait の最大待ち秒数
`line_range`	No	全文からの行範囲 "A:B"
`session_id`	Yes
`until_regex`	No	until を正規表現として扱う（既定 false＝リテラル部分一致。メタ文字を使いたい時のみ true）
`operation_id`	No	Claude operationの期待ID。agent_transcript:true時だけ指定し、古い別operationの結果を拒否する
`agent_transcript`	No	agent session の直近完了ターンの最終 assistant メッセージを返す。Claudeはlaunch相関付きStop hook result、他harnessは通常transcript／session historyを使う。長い回答がscreen tailで切れた時の回収用

Output Schema

ParametersJSON Schema

Name	Required	Description
`mode`	Yes
`text`	Yes
`schema`	Yes
`vendor`	Yes
`harness`	Yes
`turn_id`	Yes
`raw_chars`	Yes
`session_id`	Yes

TDQS

A4.1/5.0

Behavior4/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

With no annotations provided, the description carries the full burden and does substantial work: it discloses the token-reduction pipeline (control-character removal, repetition compression, head+tail folding, restoration hints), the incremental read-offset default, and agent_transcript source semantics with operation_id rejection. It does not explicitly flag side effects like advancing the read cursor, but it is far from opaque.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is one dense paragraph but stays focused: main action, default behavior, reduction details, and the special agent_transcript mode. Each clause earns its place, and the core read action is front-loaded. It could be broken into bullets for scannability, but there is no filler.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness4/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

The tool has 13 parameters and an output schema, and the description supplies essential high-level context that the schema alone does not: the reduction pipeline, the default cursor behavior, and the recovery purpose of agent_transcript. It does not walk through every wait/until/timeout combination, but the schema's 92% coverage and existing output schema compensate.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters3/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema coverage is 92%, so the baseline is 3. The description reinforces the purpose of key parameters such as agent_transcript and the incremental-read default that clarifies full, but it mostly echoes what the detailed schema descriptions already state rather than adding new per-parameter semantics.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description opens with 'セッションの出力をトークン削減して読む' – a clear verb (read), resource (session output), and qualifier (token-reduced). It goes on to define the default incremental behavior and the agent_transcript special mode, distinguishing it from pty_send/key/list which act on the session rather than read from it. The action–resource pairing is unambiguous even without naming a sibling.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines4/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

The description states a clear default (incremental from last read position) and names a concrete recovery scenario for agent_transcript: '長い回答が screen tail で切れた時の回収用'. It provides clear context without explicit exclusions or named alternatives, which fits the 'clear context, no exclusions' level.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

pty_sendA

セッションへテキストを送る。通常PTYへは送信のみ（出力は pty_read で取得）。agent session（launcher起動）への send は自動で dispatch になる: TUI の ready gate と submit 分離を通して即返り、receipt の event_cursor を返す。dispatch した子は投げっぱなしでよい＝親はここで待たない。完了通知はreceiptの wait_process.executable と wait_process.args をそのまま親のターンを塞がない別プロセスAPIへ渡して受け、PowerShell 7のStart-Processだけは windows_start_process_argument_list を単一文字列として渡す。exit を完了通知として扱う（exit 0=done / 3=timeout（既定600秒・未完了） / 4=closed。receiptのoutcomeが正で、done以外は未完了。ポーリング不要）。wait_command は人間向け互換表示でありprocess境界へ使わない。foreground実行で親のターンを塞がない。結果回収は pty_read(agent_transcript:true)、Claude の durable turn は claude_turn を使う。force:true は非Claude agent sessionへの手動介入用の素送信。aiterm相関付きClaudeの承認UIはclaude_approvalを使う。

ParametersJSON Schema

Name	Required	Description
`raw`	No	送信前サニタイズを無効化
`rtk`	No	既知コマンドを rtk 形へ委譲して送る（rtk 不在なら素通し）
`mark`	No	完了 sentinel(終了コード付き)で包む。pty_read(wait:true) が until 無しでも自動検出して完了確定する（POSIX shell と PowerShell に対応。PowerShell の rc は成功0／失敗1。fish/csh/tcsh は未対応として送信前に拒否）。 enter:false と併用すると sentinel が実行されず完了検出が発火しない（送信後に pty_key("Enter") で実行される）。
`text`	Yes	送る文字列（コマンド／prompt）。UTF-8で最大64KiB
`enter`	No	末尾で Enter を送る（agent dispatch では常に submit）
`force`	No	非Claude agent sessionでは自動dispatchせず素送信する。aiterm相関付きClaudeのactive turnには使えない
`session_id`	Yes

Output Schema

ParametersJSON Schema

Name	Required	Description
`mode`	Yes
`schema`	Yes
`vendor`	Yes
`harness`	Yes
`launch_id`	Yes
`session_id`	Yes
`event_cursor`	Yes
`wait_process`	Yes
`submit_residue`	Yes

TDQS

A4.8/5.0

Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

With no annotations available, the description carries the full burden of behavioral disclosure, and it does so thoroughly: it explains send-only semantics, dispatch behavior with event_cursor receipt, fire-and-forget semantics, exit codes (0/3/4), no polling needed, foreground non-blocking execution, and the force:true caveat. This is far beyond what schema or annotations could convey.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is a dense single paragraph rather than a structured output, which hurts scannability. However, nearly every clause carries actionable information — alternatives, exit codes, constraints — so the length is earned. It starts with the core purpose and then layers on detail, but a bulleted structure would improve readability.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

The description covers the full calling context: normal PTY vs agent dispatch, output retrieval, completion semantics, exit codes, sibling routing, and caveats like foreground execution and force mode. Since an output schema exists, return values don't need to be explained in prose. Nothing critical is missing for an agent to invoke this tool correctly.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters4/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

The schema already covers 86% of parameters, so the baseline is 3. The description adds cross-parameter context that the schema does not: the mark+enter interplay for completion detection, force:true being restricted to non-Claude agent sessions and incompatible with active turns, and the relationship between receipt fields and process API usage. These are meaningful semantics beyond the schema.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description opens with a clear statement of what the tool does: sends text to a session. It also explicitly differentiates itself from siblings by stating that output must be retrieved via pty_read, and that agent-session sends become automated dispatch, distinguishing it from pty_key, claude_turn, and claude_approval.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines5/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

The description gives concrete routing guidance: use pty_read for result collection, use claude_turn for durable turns, use claude_approval for the approval UI, and do not use wait_command across process boundaries. It also clarifies when force:true is appropriate, which tells the agent exactly when this tool is or isn't the right choice.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

Install Server

RayrunBuild, deploy, and operate MCP servers without becoming an infrastructure team.ray.runAd

TDQS

A3.6/5.0

Disambiguation3/5

The pty_*, claude_*, and agent_* tools are largely distinct, but the four *_agent legacy aliases overlap directly with agent_launch. The explicit 'old compatibility alias' notes help an agent disambiguate, so the overlap is real but mitigated.

Naming Consistency3/5

Naming is readable but mixed: pty_* uses an object-verb pattern, agent_* uses a prefix pattern, the legacy *_agent tools use a suffix pattern, and diagnostics stands alone. There is a visible logic to the clusters, but the conventions are not uniform.

Tool Count4/5

15 tools is at the upper edge of the reasonable range for a persistent terminal and developer-agent orchestration server. The count is somewhat inflated by four legacy aliases, but the core set would be comfortably sized without them.

Completeness4/5

The surface covers terminal lifecycle, agent launch/configure, Claude durable operations and approvals, reading transcripts, listing sessions, and diagnostics. The main gap is that agent completion waiting is delegated to the external aiterm-wait command rather than exposed directly as an MCP tool.

Maintenance

ActivityActive

ResponsivenessNo issues

Resources

Need Help?

Related Servers

Unclaimed servers have limited discoverability.

Looking for Admin?

If you are the server author, to access and configure the admin panel.

Related MCP Connectors

Apuchat A
io.github.opcastil11
Real-time chat for AI agents. Claude Code, Cursor, Cline and Codex join channels over MCP.
Communication Agent Orchestration
181
Remote Desktop Commander
app.desktopcommander
Hosted MCP server connecting claude.ai, ChatGPT and other AI apps to your own computer
Shell Access OS Automation
214
mcp
dev.augments
Augments MCP Server - A comprehensive framework documentation provider for Claude Code
Developer Tools Documentation Access
2128
Quallaa
com.quallaa
Talk to your public-facing AI from any MCP client — Claude, ChatGPT, Cursor, Cline, Windsurf.
AI & Machine Learning Communication
2

Related MCP Servers

claude-code-mcp
Agent Orchestration Code Execution
joschi655
A
license
A
quality
C
maintenance
An MCP server that gives orchestrator agents fine-grained control over interactive Claude Code sessions running inside tmux, enabling mid-session steering, interruption, and token-efficient result extraction.
Updated a month ago (2026-08-04 07:39 UTC)
15
MIT
claude-tmux
Coding Agents Autonomous Agents Agent Orchestration
Ilm-Alan
A
license
Not graded
quality
D
maintenance
MCP server for orchestrating multiple Claude Code instances via tmux, enabling spawning, reading, sending, listing, and killing sessions.
Updated 7 months ago (2026-01-21 20:31 UTC)
18
2
MIT
claude-tmux-bridge
Coding Agents Shell Access
PiMPStudios
A
license
A
quality
D
maintenance
An MCP server that connects Claude Desktop to an interactive Claude Code session running in a tmux terminal.
Updated 4 months ago (2026-05-03 20:21 UTC)
1
MIT
codex-mcp-bridge
Agent Orchestration Coding Agents Shell Access
eyalk11
A
license
A
quality
C
maintenance
MCP server that lets Claude Code drive the local Codex CLI as a sub-agent for concurrent queries and optional file/shell actions, using the CLI's existing login and sessions.
Updated a month ago (2026-08-08 05:10 UTC)
4
81
1
MIT

Latest Blog Posts

Who's Calling? MCP Hosts Are an Identity Blind Spot (And the Spec Knows It)
By Om-Shree-0709 on July 25, 2026.
mcp
Agent Identity
OAuth 2.1
Your AI Chatbot Just Exposed Your CEO's Salary to an Intern
By Om-Shree-0709 on July 2, 2026.
Agent Identity
MCP Security
OAuth Delegation
Why MCP Servers Need Execution Sandboxing (And Why Your Current Stack Isn't Enough)
By Om-Shree-0709 on June 30, 2026.
Agentic Ai
Prompt Injection
WebAssembly

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/kitepon/aiterm-mcp'

If you have feedback or need assistance with the MCP directory API, please join our Discord server

Aiterm

Install in your MCP client

Claude Code

Claude Desktop

Cursor

Update and rollback

Why now

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

Two ways to use it

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

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

Demo

First run (≈60 seconds)

Headless: no human at the terminal

How it works

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

vs. the alternatives

Where aiterm fits

Tools

Local runtime error snapshot

Interactive agent harnesses

Completion detection (5 layers)

Completion push for parent agents (aiterm-wait)

Token reduction

Input and output

A human can watch

Requirements

Known constraints (by design, not bugs)

Development

Try it

Shared agent environment

License

Available Tools

Output Schema

Output Schema

Output Schema

Output Schema

Output Schema

Output Schema

Output Schema

Output Schema

Output Schema

Output Schema

Output Schema

Output Schema

Maintenance

Resources

Looking for Admin?

Related MCP Connectors

Related MCP Servers

Latest Blog Posts

MCP directory API

Completion push for parent agents (`aiterm-wait`)