AgentHub/server/lib/pty-run.js

/**
 * @file pty-run.js
 * @description Owns the tmux-backed run lifecycle for the dashboard's
 * terminal-run feature: Start (create-or-adopt), Resume (`--resume`), Kill,
 * and List. Unlike the old run-spawner.js, there is no in-memory handle Map —
 * `listRuns`/`getRun` are computed fresh from `tmux list-sessions` on every
 * call, the same "computed fact, never a stored one" principle this repo
 * already applies to lane runtime up/down (see CLAUDE.md). A session killed
 * out-of-band (crash, manual `tmux kill-session`, host reboot) self-corrects
 * on the next read instead of leaving a ghost "running" row.
 *
 * Start/Resume is create-or-reuse: when the lane's tmux session already
 * exists but its pane sits at a shell prompt, the argv is typed into that
 * pane instead of being dropped on the floor by a silent adopt.
 *
 * Every session is named `agenthub-lane-<laneId>` so a real terminal can attach
 * to the exact same session (`tmux attach -t agenthub-lane-<id>`, or
 * `agenthub lanes shell`) — that's the whole point: the dashboard both creates
 * the session (one-click Start/Resume) and is just one of possibly several
 * attached clients tmux already keeps in sync.
 *
 * Spawns `claude` or `codex` in the tmux pane based on the tool parameter.
 *
 * @author Nguyễn Ngọc Trí Vĩ <vinnt@smartgift.vn>
 */

const tmux = require("./tmux");
const { routerArgvForSpawn, routerEnvForSpawn } = require("./router-apply");

let dashboardRuns = null;
try {
  dashboardRuns = require("./dashboard-runs");
} catch {
  /* db-less environment, skip persistence */
}

const RUN_ID_RE = /^agenthub-lane-(\d+)$/;
const EFFORT_LEVELS = new Set(["low", "medium", "high", "xhigh", "max"]);
const ALLOWED_PERMISSION_MODES = new Set(["acceptEdits", "default", "plan", "bypassPermissions"]);
// Pane commands that mean "idle shell prompt, safe to type a command into".
const SHELL_COMMANDS = new Set(["sh", "bash", "zsh", "fish", "dash", "ksh", "csh", "tcsh"]);
const VALID_TOOLS = new Set(["claude", "codex"]);
const CODEX_SANDBOXES = new Set(["read-only", "workspace-write", "danger-full-access"]);
const CODEX_APPROVALS = new Set(["on-request", "never"]);
// Codex has no `--effort` flag; reasoning effort is a config override.
const CODEX_EFFORTS = new Set(["low", "medium", "high", "xhigh"]);

function runIdForLane(laneId) {
  return `agenthub-lane-${laneId}`;
}

function laneIdFromRunId(id) {
  const m = typeof id === "string" ? id.match(RUN_ID_RE) : null;
  return m ? Number(m[1]) : null;
}

function makeErr(code, message) {
  const err = new Error(message);
  err.code = code;
  return err;
}

/**
 * Build the pane's command argv for Claude. Unlike the old stream-json spawner
 * there is no headless/conversation split — every run is a live interactive
 * pane, so an initial prompt (when given) is a trailing POSITIONAL argument:
 * `claude` treats a bare positional as the first turn's message and stays
 * interactive afterward (unlike `-p`, which forces print-and-exit and closes
 * stdin).
 */
function buildClaudeArgv({ model, permissionMode, effort, resumeSessionId, initialPrompt }) {
  const argv = ["claude"];
  argv.push("--permission-mode", permissionMode || "acceptEdits");
  // Puts "bypass permissions" INTO the pane's shift+tab cycle without turning
  // it on: `claude --help` — "Enable bypassing all permission checks as an
  // option, without it being enabled by default." Without it the cycle is only
  // plan/auto/manual/acceptEdits, so a run started in any non-bypass mode could
  // never reach bypass from inside the pane, while a real terminal could.
  // The starting mode is still whatever --permission-mode says above.
  argv.push("--allow-dangerously-skip-permissions");
  if (model) argv.push("--model", model);
  if (effort && EFFORT_LEVELS.has(effort)) argv.push("--effort", effort);
  if (resumeSessionId) argv.push("--resume", resumeSessionId);
  if (initialPrompt) argv.push(initialPrompt);
  return argv;
}

/**
 * Codex's CLI shape differs from Claude's in two ways that matter here:
 * `resume` is a SUBCOMMAND (so it precedes every flag) and reasoning effort has
 * no flag of its own — it rides on a `-c` config override. The prompt stays a
 * trailing positional, same as Claude, so the pane stays interactive.
 */
function buildCodexArgv({
  model,
  effort,
  sandbox,
  approval,
  resumeSessionId,
  resumeLast,
  initialPrompt,
  routerProfile,
}) {
  const argv = ["codex"];
  // Before the subcommand and before the prompt: `-p` layers
  // `$CODEX_HOME/<name>.config.toml` over the base config, and an option that
  // lands after the positional prompt is not an option any more.
  argv.push(...routerArgvForSpawn({ tool: "codex", router_profile: routerProfile }));
  if (resumeSessionId) argv.push("resume", resumeSessionId);
  else if (resumeLast) argv.push("resume", "--last");
  if (model) argv.push("-m", model);
  if (sandbox) argv.push("-s", sandbox);
  if (approval) argv.push("-a", approval);
  if (effort && CODEX_EFFORTS.has(effort)) argv.push("-c", `model_reasoning_effort=${effort}`);
  if (initialPrompt) argv.push(initialPrompt);
  return argv;
}

function buildArgv(args) {
  return args.tool === "codex" ? buildCodexArgv(args) : buildClaudeArgv(args);
}

/**
 * @param {object} args
 * @param {number} args.laneId
 * @param {string} args.cwd
 * @param {string} [args.tool="claude"]
 * @param {string} [args.model]
 * @param {string} [args.permissionMode]
 * @param {string} [args.effort]
 * @param {string} [args.sandbox]
 * @param {string} [args.approval]
 * @param {string} [args.resumeSessionId]
 * @param {boolean} [args.resumeLast]
 * @param {string} [args.initialPrompt]
 * @param {string} [args.routerProfile] AI Router profile the LANE runs against
 *   (`lanes.router_profile`), never a request field.
 */
function spawnRun(args) {
  const {
    laneId,
    cwd,
    tool = "claude",
    model,
    permissionMode,
    effort,
    sandbox,
    approval,
    resumeSessionId,
    resumeLast,
    initialPrompt,
    routerProfile,
  } = args || {};
  if (typeof laneId !== "number" || !Number.isInteger(laneId)) {
    throw makeErr("EBADLANE", "laneId must be an integer");
  }
  if (typeof cwd !== "string" || !cwd) {
    throw makeErr("EBADCWD", "cwd is required");
  }
  if (!VALID_TOOLS.has(tool)) {
    throw makeErr("EBADTOOL", `tool must be one of: ${Array.from(VALID_TOOLS).join(", ")}`);
  }
  if (tool === "codex") {
    if (sandbox != null && sandbox !== "" && !CODEX_SANDBOXES.has(sandbox)) {
      throw makeErr(
        "EBADSANDBOX",
        `sandbox must be one of: ${Array.from(CODEX_SANDBOXES).join(", ")}`
      );
    }
    if (approval != null && approval !== "" && !CODEX_APPROVALS.has(approval)) {
      throw makeErr(
        "EBADAPPROVAL",
        `approval must be one of: ${Array.from(CODEX_APPROVALS).join(", ")}`
      );
    }
  }
  if (permissionMode != null && !ALLOWED_PERMISSION_MODES.has(permissionMode)) {
    throw makeErr(
      "EBADMODE",
      `permissionMode must be one of: ${Array.from(ALLOWED_PERMISSION_MODES).join(", ")}`
    );
  }
  if (effort != null && effort !== "" && !EFFORT_LEVELS.has(effort)) {
    throw makeErr("EBADEFFORT", `effort must be one of: ${Array.from(EFFORT_LEVELS).join(", ")}`);
  }
  if (
    resumeSessionId != null &&
    (typeof resumeSessionId !== "string" || !/^[A-Za-z0-9-]{8,}$/.test(resumeSessionId))
  ) {
    throw makeErr("EBADSESSION", "resumeSessionId is not a valid session id");
  }

  // A run IS a tmux session, so an absent binary is a precondition failure, not
  // a spawn failure. Without this the caller gets `spawnSync tmux ENOENT` from
  // three directories down — a message that names no fix — and the route maps
  // it to 400, blaming the request for the host's missing package. Checked here
  // rather than in each route because both spawn paths (POST /api/run and
  // POST /api/lanes/:id/start) come through this function.
  if (!tmux.isTmuxAvailable()) {
    throw makeErr(
      "ETMUXMISSING",
      "tmux is not installed, and every run the dashboard starts is a tmux session. " +
        "Install it (sudo apt install tmux / brew install tmux; on Windows run AgentHub under WSL2), " +
        "then restart the dashboard. Everything else keeps working without it."
    );
  }

  const id = runIdForLane(laneId);
  const startedAt = Date.now();
  const argv = buildArgv({
    tool,
    model,
    permissionMode,
    effort,
    sandbox,
    approval,
    resumeSessionId,
    resumeLast,
    initialPrompt,
    routerProfile,
  });

  const record = () => {
    if (!dashboardRuns) return;
    dashboardRuns.recordRun({
      id,
      sessionId: resumeSessionId || null,
      // `dashboard_runs.mode` is NOT NULL. Passing null made every insert throw
      // inside recordRun's best-effort catch, so no tmux-backed run was ever
      // persisted: publicRun then read its cwd/model/prompt/startedAt off a row
      // that did not exist (a blank row dated Jan 1 1970 in the runs modal), and
      // session_id never landed either, so a resumed session stayed listed as a
      // separate `external` row. Every run here is one interactive pane.
      mode: "interactive",
      cwd,
      tool: tool === "codex" ? "codex" : "claude",
      model: model || null,
      permissionMode: permissionMode || "acceptEdits",
      effort: effort || null,
      resumeSessionId: resumeSessionId || null,
      prompt: initialPrompt || "",
      status: "running",
      startedAt,
      endedAt: null,
      exitCode: null,
      laneId,
    });
  };

  // Every pane the dashboard starts carries its lane id. A `agenthub stage` call
  // from inside it then resolves the lane exactly, instead of matching on cwd —
  // which is ambiguous now that one directory can back both a `claude` lane and
  // a `codex` lane (see bin/agenthub.js's resolveLaneArg).
  // A codex lane on an AI Router also needs the key its layer's `env_key`
  // names. A claude lane needs nothing here — its settings file already carries
  // the router — and a lane on the vendor's own API needs nothing either.
  const env = {
    AGENTHUB_LANE_ID: String(laneId),
    ...routerEnvForSpawn({ tool, router_profile: routerProfile || null }),
  };

  if (!tmux.hasSession(id)) {
    tmux.newSession({ name: id, cwd, argv, env });
    record();
  } else if (SHELL_COMMANDS.has(tmux.paneCommand(id) || "")) {
    // The session exists but its pane is sitting at a bare shell prompt — a
    // `agenthub lanes shell`, or a `claude` that already exited. Adopting it
    // silently here would swallow the whole request: a Resume would spawn no
    // `--resume` and an initial prompt would never be typed, while the API
    // still answered 200. Run the argv in the pane the user already sees
    // instead of erroring or opening a second session.
    tmux.sendCommand(id, argv, env, cwd);
    record();
  }
  // Pane is running something (a live `claude`, an editor, a build): adopt
  // silently, same convention as this repo's server port-adoption logic — no
  // error, no duplicate session. Attaching shows the user what is running.
  //
  // Adopting is only honest when the request asked for nothing more than a
  // pane. A Resume or an initial prompt cannot be delivered into a pane that is
  // already running an agent (typing over it is what the shell-pane branch
  // above exists to avoid), so answering 200 there reports work that never
  // happened: the user sees "resumed" while no `--resume` was ever spawned.
  else if (resumeSessionId || initialPrompt) {
    throw makeErr(
      "ERUNLIVE",
      `lane ${laneId}'s pane is already running \`${tmux.paneCommand(id) || "a process"}\` — stop it before resuming a session or sending a prompt`
    );
  }

  return getRun(id);
}

function killRun(id) {
  if (!id || !tmux.hasSession(id)) return false;
  tmux.killSession(id);
  if (dashboardRuns) {
    dashboardRuns.patchRun({ id, status: "killed", endedAt: Date.now() });
  }
  return true;
}

function publicRun(id) {
  const laneId = laneIdFromRunId(id);
  const live = tmux.hasSession(id);
  const row = dashboardRuns ? dashboardRuns.getRun(id) : null;
  return {
    id,
    laneId,
    status: live ? "running" : "gone",
    tool: row?.tool || "claude",
    cwd: row?.cwd || null,
    model: row?.model || null,
    permissionMode: row?.permission_mode || null,
    effort: row?.effort || null,
    resumeSessionId: row?.resume_session_id || null,
    sessionId: row?.session_id || null,
    startedAt: row?.started_at || null,
    promptPreview: row?.prompt_preview || null,
  };
}

function getRun(id) {
  if (!id) return null;
  return publicRun(id);
}

/** Computed fresh from tmux state every call — see file header. */
function listRuns() {
  return tmux.listSessions("agenthub-lane-").map(publicRun);
}

module.exports = {
  spawnRun,
  killRun,
  getRun,
  listRuns,
  laneIdFromRunId,
  runIdForLane,
  SHELL_COMMANDS,
};