@ours.network/fleet 0.9.4 → 0.9.5

package/dist/monitor.js

@@ -1,4 +1,4 @@
-import { existsSync, readFileSync, writeFileSync } from 'node:fs';
+import { existsSync, readFileSync, renameSync, writeFileSync } from 'node:fs';
 import { homedir } from 'node:os';
 import { join } from 'node:path';
 // Code constants (not config — YAGNI, design §2).
@@ -242,12 +242,16 @@ function stillInComposer(pane, line) {
 }
 export class Monitor {
     name;
+    identity;
     cfg;
     deps;
     ep;
     statusPath;
     cursorPath;
+    statePath;
     cursor = null;
+    deliveredCursor = null;
+    pendingState = null;
     fatal = false;
     stopped = false;
     bootDeadline = 0;
@@ -258,16 +262,25 @@ export class Monitor {
     turnFailThreshold;
     constructor(o) {
         this.name = o.name;
+        this.identity = o.identity ?? o.name;
         this.cfg = o.cfg;
         this.deps = o.deps;
         this.ep = resolveEndpoint(o.deps.env);
         this.statusPath = join(o.agentDir, '.monitor-status');
         this.cursorPath = join(o.agentDir, '.notify-cursor');
+        this.statePath = join(o.agentDir, '.monitor-state.json');
         const n = o.cfg.turn_fail_threshold;
         this.turnFailThreshold = typeof n === 'number' && n >= 1 ? n : DEFAULT_TURN_FAIL_THRESHOLD;
     }
-    /** Prime at the stream tip (or resume a persisted cursor if the daemon is down). */
+    /** Resume the last delivered cursor; only a brand-new monitor primes at stream tip. */
     async prime() {
+        const persisted = this.readPersistedCursor();
+        if (persisted !== null) {
+            this.cursor = persisted;
+            this.deliveredCursor = persisted;
+            this.setStatus('armed');
+            return;
+        }
         try {
             const body = await this.doFetch('tip', LONGPOLL_TIMEOUT_MS);
             this.cursor = typeof body.cursor === 'number' ? body.cursor : 0;
@@ -280,7 +293,7 @@ export class Monitor {
                 this.setStatus(`failed: ${e.message}`);
             }
             else {
-                this.cursor = this.readPersistedCursor();
+                this.cursor = null;
                 this.setStatus(`degraded: prime failed (${msg(e)})`);
             }
         }
@@ -291,6 +304,7 @@ export class Monitor {
             return;
         this.bootDeadline = this.deps.now() + BOOT_GRACE_MS;
         let backoff = 0;
+        const pending = [];
         while (!this.stopped) {
             if (!this.deps.isAlive(pid)) {
                 this.setStatus('degraded: session offline');
@@ -314,12 +328,35 @@ export class Monitor {
                 await this.deps.sleep(backoff);
                 continue;
             }
-            this.advance(body.cursor);
+            this.advance(body.cursor, false);
             const batch = filterEvents(body.events ?? [], this.cfg.wake_sources);
-            if (batch.length === 0)
+            pending.push(...batch);
+            if (pending.length === 0) {
+                this.persistCursor();
                 continue;
-            await this.coalesce(batch);
-            await this.deliver(pid, batch);
+            }
+            this.pendingState = {
+                count: pending.length,
+                eventTypes: uniq(pending.map(event => event.event ?? 'unknown')),
+                attempts: (this.pendingState?.attempts ?? 0) + 1,
+            };
+            this.persistState();
+            await this.coalesce(pending);
+            // Do not durably commit this cursor until the session explicitly accepts
+            // the wake. If delivery fails or the runner crashes, the daemon replays
+            // from the last committed cursor and the wake is attempted again.
+            let accepted = false;
+            try {
+                accepted = await this.deliver(pid, pending);
+            }
+            catch (e) {
+                this.setStatus(`degraded: delivery failed (${msg(e)})`);
+            }
+            if (accepted) {
+                pending.length = 0;
+                this.pendingState = null;
+                this.persistCursor();
+            }
         }
     }
     stop() {
@@ -336,12 +373,22 @@ export class Monitor {
             return;
         try {
             const more = await this.doFetch(String(this.cursor ?? 0), COALESCE_HOLD_MS);
-            this.advance(more.cursor);
+            this.advance(more.cursor, false);
             batch.push(...filterEvents(more.events ?? [], this.cfg.wake_sources));
         }
         catch { /* no stragglers / abort — deliver what we have */ }
     }
     async deliver(pid, batch) {
+        const line = formatNotificationLine(batch);
+        if (this.deps.delivery) {
+            const result = await this.deps.delivery.submit(line);
+            if (!result.accepted) {
+                this.setStatus(`degraded: ACP prompt not accepted${result.detail ? ` (${result.detail})` : ''}`);
+                return false;
+            }
+            this.recordTurn('completed');
+            return true;
+        }
         const state = await this.awaitInjectable(pid);
         if (state !== 'ready') {
             if (state === 'offline')
@@ -349,9 +396,8 @@ export class Monitor {
             else if (state === 'modal')
                 this.setStatus(`degraded: modal wedge — pane held a dialog for ` +
                     `${MODAL_GIVE_UP_MS / 1000}s, wake not injected`);
-            return; // events remain covered by unread.json / SessionStart backlog
+            return false;
         }
-        const line = formatNotificationLine(batch);
         await this.clearComposer(); // start from an empty composer
         await this.deps.tmux.sendText(this.name, line); // send-keys -l + Enter
         let delivered = false;
@@ -360,8 +406,12 @@ export class Monitor {
         // dead pane makes safeCapture return '' ⇒ not-in-composer ⇒ breaks, no wasted Enter.
         for (let i = 0; i < MAX_ENTER_RETRIES; i++) {
             await this.deps.sleep(POST_VERIFY_MS);
-            const pane = await safeCapture(this.deps.tmux, this.name);
-            if (!stillInComposer(pane, line)) {
+            const capture = await safeCapture(this.deps.tmux, this.name);
+            if (!capture.ok) {
+                this.setStatus('degraded: capture failed during injection verification');
+                return false;
+            }
+            if (!stillInComposer(capture.pane, line)) {
                 delivered = true;
                 break;
             }
@@ -369,12 +419,13 @@ export class Monitor {
         }
         if (!delivered) {
             this.setStatus('degraded: injection unverified');
-            return;
+            return false;
         }
         // The wake landed and a turn started; observe how that turn terminates so a
         // refusal-wedge (every turn dies with `API Error:` while delivery stays green)
         // becomes visible in `.monitor-status` instead of masquerading as armed (#19).
         await this.observeTurnOutcome(pid);
+        return true;
     }
     /**
      * Watch the pane until the just-triggered turn settles, then fold its outcome
@@ -389,12 +440,16 @@ export class Monitor {
                 return; // shutting down — leave status
             if (!this.deps.isAlive(pid) || !(await this.deps.tmux.has(this.name)))
                 return; // loop marks offline
-            const pane = await safeCapture(this.deps.tmux, this.name);
-            if (looksApiError(pane)) {
+            const capture = await safeCapture(this.deps.tmux, this.name);
+            if (!capture.ok) {
+                this.setStatus('degraded: capture failed during turn observation');
+                return;
+            }
+            if (looksApiError(capture.pane)) {
                 this.recordTurn('api-error');
                 return;
             }
-            if (!looksRunning(pane)) {
+            if (!looksRunning(capture.pane)) {
                 this.recordTurn('completed');
                 return;
             }
@@ -443,8 +498,13 @@ export class Monitor {
                 await this.deps.sleep(this.bootDeadline - now);
                 continue;
             }
-            const pane = await safeCapture(this.deps.tmux, this.name);
-            if (!looksModal(pane))
+            const capture = await safeCapture(this.deps.tmux, this.name);
+            if (!capture.ok) {
+                this.setStatus('degraded: capture failed while checking session readiness');
+                await this.deps.sleep(MODAL_RETRY_MS);
+                continue;
+            }
+            if (!looksModal(capture.pane))
                 return 'ready';
             if (modalWaits++ >= MAX_MODAL_WAITS)
                 return 'modal';
@@ -457,7 +517,7 @@ export class Monitor {
         const timer = this.deps.timers.set(() => ctrl.abort(), holdMs);
         let resp;
         try {
-            resp = await this.deps.fetch(`${this.ep.url(this.name)}?since=${since}`, { headers: this.ep.headers, signal: ctrl.signal });
+            resp = await this.deps.fetch(`${this.ep.url(this.identity)}?since=${since}`, { headers: this.ep.headers, signal: ctrl.signal });
         }
         finally {
             this.deps.timers.clear(timer);
@@ -469,16 +529,22 @@ export class Monitor {
             throw new Error(`daemon returned HTTP ${resp.status}`);
         return resp.json();
     }
-    advance(cursor) {
+    advance(cursor, persist = true) {
         if (typeof cursor === 'number' && cursor !== this.cursor) {
             this.cursor = cursor;
-            this.persistCursor();
+            if (persist)
+                this.persistCursor();
+            else
+                this.persistState();
         }
     }
     persistCursor() {
         try {
-            if (this.cursor !== null)
+            if (this.cursor !== null) {
+                this.deliveredCursor = this.cursor;
                 writeFileSync(this.cursorPath, `${this.cursor}\n`);
+                this.persistState();
+            }
         }
         catch (e) {
             this.deps.log(`[${this.name}] monitor: failed to persist cursor: ${msg(e)}`);
@@ -486,15 +552,46 @@ export class Monitor {
     }
     readPersistedCursor() {
         try {
+            if (existsSync(this.statePath)) {
+                const state = JSON.parse(readFileSync(this.statePath, 'utf8'));
+                if ((state.identity !== undefined && state.identity !== this.identity)
+                    || (state.profileKey !== undefined && state.profileKey !== this.ep.origin))
+                    return null;
+                if (typeof state.deliveredCursor === 'number') {
+                    this.deliveredCursor = state.deliveredCursor;
+                    return state.deliveredCursor;
+                }
+            }
             if (!existsSync(this.cursorPath))
                 return null;
             const n = parseInt(readFileSync(this.cursorPath, 'utf8').trim(), 10);
+            if (Number.isFinite(n))
+                this.deliveredCursor = n;
             return Number.isFinite(n) ? n : null;
         }
         catch {
             return null;
         }
     }
+    /** Atomically persist body-free delivery state; restart always resumes from deliveredCursor. */
+    persistState() {
+        try {
+            const tmp = `${this.statePath}.${process.pid}.tmp`;
+            writeFileSync(tmp, JSON.stringify({
+                version: 1,
+                identity: this.identity,
+                profileKey: this.ep.origin,
+                observedCursor: this.cursor,
+                deliveredCursor: this.deliveredCursor,
+                pending: this.pendingState,
+                updatedAt: new Date(this.deps.now()).toISOString(),
+            }, null, 2) + '\n', { mode: 0o600 });
+            renameSync(tmp, this.statePath);
+        }
+        catch (e) {
+            this.deps.log(`[${this.name}] monitor: failed to persist state: ${msg(e)}`);
+        }
+    }
     setStatus(s) {
         try {
             writeFileSync(this.statusPath, `${s}\n`);
@@ -511,10 +608,10 @@ export function createMonitor(o) {
 }
 async function safeCapture(tmux, name) {
     try {
-        return await tmux.capture(name);
+        return { ok: true, pane: await tmux.capture(name) };
     }
     catch {
-        return '';
+        return { ok: false, pane: '' };
     }
 }
 const msg = (e) => e?.message ?? String(e);

package/dist/runner.js

@@ -3,7 +3,7 @@ import { join } from 'node:path';
 import { randomUUID } from 'node:crypto';
 import { parse } from 'yaml';
 import { agentDir, home, stateRoot } from './paths.js';
-import { loadConfig, findRole } from './config.js';
+import { loadConfig, findRole, resolvePermissions } from './config.js';
 import { getAdapter } from './harness/registry.js';
 import { Tmux } from './tmux.js';
 import { createMonitor } from './monitor.js';
@@ -11,6 +11,9 @@ import { realExec, shq } from './exec.js';
 import { resolveIsolation } from './isolation/policy.js';
 import { selectIsolationBackend } from './isolation/registry.js';
 import { resourceArgs, cpuControllerDelegated } from './isolation/resources.js';
+import { AcpSession } from './session/acp.js';
+import { RoleControlServer } from './session/control.js';
+import { TmuxSession } from './session/tmux.js';
 const defaultDeps = () => ({
     tmux: new Tmux(),
     exec: realExec,
@@ -182,10 +185,17 @@ export async function runOnce(name, opts = {}, partialDeps = {}) {
         writeFileSync(bootedFile, '');
     const runCwd = role.cwd && existsSync(role.cwd) ? role.cwd : dir;
     const prep = await adapter.prepareSession(role, { stateDir: dir, runCwd });
-    const launch = adapter.buildLaunch(role, mode, { sessionId }, prep);
+    const sessionBackend = role.session ?? 'tmux';
+    const launch = sessionBackend === 'acp'
+        ? (() => {
+            if (!adapter.buildAcpLaunch)
+                throw new Error(`harness '${role.harness}' does not support the ACP session backend`);
+            return adapter.buildAcpLaunch(role, prep);
+        })()
+        : adapter.buildLaunch(role, mode, { sessionId }, prep);
     // Isolation is additive: only roles that declare `isolation:` are wrapped. The
     // env prefix + exit capture in buildPaneCommand stay host-side (see §5.3).
-    let paneArgv = launch.argv;
+    let wrappedArgv = launch.argv;
     if (role.isolation) {
         const addDirs = role.harness === 'codex'
             ? (role.harness_options?.add_dirs ?? [])
@@ -204,14 +214,14 @@ export async function runOnce(name, opts = {}, partialDeps = {}) {
             deps.log(`[${name}] isolation: ${sel.backend.id} (net=${policy.network}) ${sel.detail}`);
             rmSync(degradedMarker, { force: true });
         }
-        paneArgv = sel.backend.wrap(launch.argv, policy, ctx);
+        wrappedArgv = sel.backend.wrap(launch.argv, policy, ctx);
         // Resource caps wrap the sandbox from OUTSIDE, at the pane's own cgroup scope
         // (§5.4). Applies even when the sandbox degraded to none.
         const { argv: rprefix, warnings } = resourceArgs(policy.resources, deps.cpuDelegated());
         for (const w of warnings)
             deps.log(`[${name}] WARNING ${w}`);
         if (rprefix.length)
-            paneArgv = [...rprefix, ...paneArgv];
+            wrappedArgv = [...rprefix, ...wrappedArgv];
     }
     // Start-stagger: space this launch at least start_stagger_ms after the previous
     // agent launch across the whole host, so a burst of boots (systemd starts every
@@ -232,34 +242,79 @@ export async function runOnce(name, opts = {}, partialDeps = {}) {
     // (backlog before the tip is the SessionStart hook's job). Disabled roles keep
     // the legacy in-session watch. Temp snapshots predating `monitor:` are treated
     // as disabled (monitor may be undefined on an old role.yaml).
+    const resolvedMonitorDeps = monitorDeps(deps, role.env);
     const monitor = role.monitor?.enabled ? deps.createMonitor({
-        name, agentDir: dir, cfg: role.monitor,
-        deps: monitorDeps(deps, role.env),
+        name, identity: role.identity, agentDir: dir, cfg: role.monitor,
+        deps: resolvedMonitorDeps,
     }) : null;
     if (monitor)
         await monitor.prime();
     rmSync(exitFile, { force: true });
-    await deps.tmux.kill(name);
-    await deps.tmux.newSession(name, runCwd, buildPaneCommand(launch, role.env, exitFile, paneArgv));
-    let pid = null;
-    for (let i = 0; i < 40 && pid === null; i++) {
-        pid = await deps.tmux.panePid(name);
-        if (pid === null)
-            await deps.sleep(250);
+    let pid;
+    let sessionHandle;
+    let acpSession;
+    let control;
+    if (sessionBackend === 'acp') {
+        acpSession = await AcpSession.start({
+            name,
+            argv: wrappedArgv,
+            cwd: runCwd,
+            env: { ...launch.env, ...(role.env ?? {}) },
+            stateDir: dir,
+            mode,
+            permissions: role.permissions ?? resolvePermissions(undefined, undefined),
+            log: deps.log,
+        });
+        pid = acpSession.pid;
+        sessionHandle = acpSession;
+        control = new RoleControlServer(dir, acpSession, deps.log);
+        await control.start();
+        resolvedMonitorDeps.delivery = {
+            submit: async (text) => {
+                const result = await acpSession.submitPrompt(text);
+                return { accepted: result.accepted, detail: result.detail };
+            },
+        };
+        const firstPrompt = mode === 'fresh'
+            ? `Read and follow ${join(dir, 'briefing.md')} now.`
+            : adapter.vocabulary.restartPrompt(role.identity, join(dir, 'WORKLOG.md'), role);
+        const started = await acpSession.submitPrompt(firstPrompt);
+        if (!started.accepted) {
+            monitor?.stop();
+            await control.close();
+            await acpSession.close();
+            throw new Error(`[${name}] ACP session rejected startup prompt: ${started.detail ?? started.outcome}`);
+        }
+    }
+    else {
+        await deps.tmux.kill(name);
+        await deps.tmux.newSession(name, runCwd, buildPaneCommand(launch, role.env, exitFile, wrappedArgv));
+        let panePid = null;
+        for (let i = 0; i < 40 && panePid === null; i++) {
+            panePid = await deps.tmux.panePid(name);
+            if (panePid === null)
+                await deps.sleep(250);
+        }
+        if (panePid === null)
+            throw new Error(`[${name}] could not resolve tmux pane pid`);
+        pid = panePid;
+        sessionHandle = new TmuxSession(name, pid, deps.tmux, deps.isAlive);
     }
-    if (pid === null)
-        throw new Error(`[${name}] could not resolve tmux pane pid`);
-    deps.log(`[${name}] up; pid=${pid} cwd=${runCwd} harness=${role.harness} mode=${mode}`);
+    deps.log(`[${name}] up; pid=${pid} cwd=${runCwd} harness=${role.harness} session=${sessionBackend} mode=${mode}`);
     // The monitor loop lives exactly as long as the session: it starts once the
     // pane pid is known and is stopped when that pid dies (task dies with runner).
     const monitorLoop = monitor?.run(pid);
     const start = deps.now();
-    while (deps.isAlive(pid))
+    while (sessionHandle.isAlive())
         await deps.sleep(2000);
     if (monitor) {
         monitor.stop();
         await monitorLoop;
     }
+    if (control)
+        await control.close();
+    if (acpSession)
+        await acpSession.close();
     const elapsed = (deps.now() - start) / 1000;
     const code = existsSync(exitFile) ? readFileSync(exitFile, 'utf8').trim() : 'crash';
     const rotate = (why) => {

package/dist/session/acp.d.ts

@@ -0,0 +1,49 @@
+import type { CommonPermissions } from '../config.js';
+import type { SessionEvent, SessionHandle, SessionSnapshot, TurnResult } from './types.js';
+export interface AcpSessionOptions {
+    name: string;
+    argv: string[];
+    cwd: string;
+    env: Record<string, string>;
+    stateDir: string;
+    mode: 'fresh' | 'resume';
+    permissions: CommonPermissions;
+    log(line: string): void;
+}
+/**
+ * Persistent ACP v1 client. It is the sole owner of the agent's stdio; all
+ * human/automation attachment happens through the fleet role-control protocol.
+ */
+export declare class AcpSession implements SessionHandle {
+    private readonly options;
+    readonly backend: "acp";
+    readonly pid: number;
+    private readonly child;
+    private readonly events;
+    private readonly sessionFile;
+    private readonly pendingPermissions;
+    private connection;
+    private sessionId?;
+    private readiness;
+    private lastError?;
+    private promptTail;
+    private capabilities?;
+    private controllerCount;
+    private constructor();
+    static start(options: AcpSessionOptions): Promise<AcpSession>;
+    isAlive(): boolean;
+    snapshot(): SessionSnapshot;
+    submitPrompt(text: string): Promise<TurnResult>;
+    interrupt(): Promise<void>;
+    respondPermission(permissionId: string, optionId: string): boolean;
+    eventsSince(seq: number): SessionEvent[];
+    subscribe(listener: (event: SessionEvent) => void): () => void;
+    setControllerAttached(attached: boolean): void;
+    close(): Promise<void>;
+    private initialize;
+    private runPrompt;
+    private requestPermission;
+    private withinAutomaticBoundary;
+    private recordUpdate;
+    private fail;
+}