@ours.network/fleet 0.9.5 → 0.9.7
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +101 -0
- package/dist/atomic-file.d.ts +30 -0
- package/dist/atomic-file.js +86 -0
- package/dist/briefing.d.ts +6 -0
- package/dist/briefing.js +41 -11
- package/dist/cli.js +95 -21
- package/dist/config.d.ts +15 -1
- package/dist/config.js +47 -2
- package/dist/creation.d.ts +179 -0
- package/dist/creation.js +254 -0
- package/dist/docs.d.ts +28 -1
- package/dist/docs.js +132 -0
- package/dist/doctor.js +74 -16
- package/dist/harness/claude-code.d.ts +39 -3
- package/dist/harness/claude-code.js +126 -24
- package/dist/harness/codex.d.ts +7 -1
- package/dist/harness/codex.js +57 -10
- package/dist/harness/registry.d.ts +2 -0
- package/dist/harness/registry.js +19 -0
- package/dist/harness/types.d.ts +50 -3
- package/dist/isolation/bubblewrap.js +7 -1
- package/dist/isolation/policy.d.ts +34 -5
- package/dist/isolation/policy.js +114 -7
- package/dist/isolation/resources.d.ts +6 -3
- package/dist/isolation/resources.js +6 -3
- package/dist/isolation/types.d.ts +19 -1
- package/dist/monitor.d.ts +30 -3
- package/dist/monitor.js +63 -25
- package/dist/ops.d.ts +15 -2
- package/dist/ops.js +32 -9
- package/dist/permissions.d.ts +70 -0
- package/dist/permissions.js +97 -0
- package/dist/runner.d.ts +65 -2
- package/dist/runner.js +239 -19
- package/dist/session/acp.d.ts +22 -1
- package/dist/session/acp.js +110 -26
- package/dist/session/control.d.ts +49 -1
- package/dist/session/control.js +116 -12
- package/dist/session/tmux.d.ts +8 -1
- package/dist/session/tmux.js +34 -4
- package/dist/session/types.d.ts +92 -1
- package/dist/session/types.js +42 -1
- package/dist/spawn.d.ts +27 -2
- package/dist/spawn.js +153 -15
- package/dist/supervisor/launchd.d.ts +50 -0
- package/dist/supervisor/launchd.js +121 -4
- package/dist/supervisor/none.js +22 -4
- package/dist/supervisor/systemd.d.ts +8 -1
- package/dist/supervisor/systemd.js +94 -4
- package/dist/supervisor/types.d.ts +36 -3
- package/dist/tmux.d.ts +34 -2
- package/dist/tmux.js +48 -11
- package/package.json +1 -1
package/dist/runner.js
CHANGED
|
@@ -2,8 +2,8 @@ import { existsSync, readFileSync, writeFileSync, rmSync, mkdirSync } from 'node
|
|
|
2
2
|
import { join } from 'node:path';
|
|
3
3
|
import { randomUUID } from 'node:crypto';
|
|
4
4
|
import { parse } from 'yaml';
|
|
5
|
-
import { agentDir,
|
|
6
|
-
import { loadConfig, findRole, resolvePermissions } from './config.js';
|
|
5
|
+
import { agentDir, stateRoot } from './paths.js';
|
|
6
|
+
import { loadConfig, findRole, isolationContextFor, resolvePermissions, } from './config.js';
|
|
7
7
|
import { getAdapter } from './harness/registry.js';
|
|
8
8
|
import { Tmux } from './tmux.js';
|
|
9
9
|
import { createMonitor } from './monitor.js';
|
|
@@ -14,6 +14,7 @@ import { resourceArgs, cpuControllerDelegated } from './isolation/resources.js';
|
|
|
14
14
|
import { AcpSession } from './session/acp.js';
|
|
15
15
|
import { RoleControlServer } from './session/control.js';
|
|
16
16
|
import { TmuxSession } from './session/tmux.js';
|
|
17
|
+
import { classifyShellStatus } from './session/types.js';
|
|
17
18
|
const defaultDeps = () => ({
|
|
18
19
|
tmux: new Tmux(),
|
|
19
20
|
exec: realExec,
|
|
@@ -42,7 +43,12 @@ export function buildPaneCommand(launch, roleEnv, exitStatusPath, paneArgv = lau
|
|
|
42
43
|
const env = { PATH: process.env.PATH ?? '', ...launch.env, ...(roleEnv ?? {}) };
|
|
43
44
|
const envPfx = 'env ' + Object.entries(env).map(([k, v]) => `${k}=${shq(v)}`).join(' ');
|
|
44
45
|
const cmd = paneArgv.map(shq).join(' ');
|
|
45
|
-
|
|
46
|
+
// Write a structured record, not a bare number: the wait status alone cannot
|
|
47
|
+
// say whether the file is missing because the program never exited or because
|
|
48
|
+
// nothing ever wrote it. `printf` is POSIX; no shell branching is needed
|
|
49
|
+
// because classification happens in one place, in TypeScript.
|
|
50
|
+
const record = `'{"version":1,"backend":"tmux","status":'"$__ofs"'}'`;
|
|
51
|
+
return `${envPfx} ${cmd}; __ofs=$?; printf %s ${record} > ${shq(exitStatusPath)}`;
|
|
46
52
|
}
|
|
47
53
|
/** Adapt runner deps and the role's daemon-profile overrides for the monitor. */
|
|
48
54
|
function monitorDeps(deps, roleEnv) {
|
|
@@ -59,6 +65,89 @@ function monitorDeps(deps, roleEnv) {
|
|
|
59
65
|
timers: { set: (fn, ms) => setTimeout(fn, ms), clear: t => clearTimeout(t) },
|
|
60
66
|
};
|
|
61
67
|
}
|
|
68
|
+
/**
|
|
69
|
+
* Read the pane's `.exit-status`. Three shapes are accepted: the structured
|
|
70
|
+
* record written above, a bare number left by a pre-upgrade pane (so an
|
|
71
|
+
* in-place upgrade does not misread a real exit), and anything else — which is
|
|
72
|
+
* `unknown`, never an invented failure. A missing file returns null so the
|
|
73
|
+
* caller can distinguish "no record" from "a record saying unknown".
|
|
74
|
+
*/
|
|
75
|
+
export function readExitRecord(path) {
|
|
76
|
+
if (!existsSync(path))
|
|
77
|
+
return null;
|
|
78
|
+
const raw = readFileSync(path, 'utf8').trim();
|
|
79
|
+
if (!raw)
|
|
80
|
+
return { version: 1, class: 'unknown', detail: 'the pane left an empty exit record' };
|
|
81
|
+
if (/^-?\d+$/.test(raw))
|
|
82
|
+
return classifyShellStatus(Number(raw)); // legacy `echo $?`
|
|
83
|
+
try {
|
|
84
|
+
const parsed = JSON.parse(raw);
|
|
85
|
+
if (typeof parsed.status === 'number')
|
|
86
|
+
return classifyShellStatus(parsed.status);
|
|
87
|
+
}
|
|
88
|
+
catch { /* fall through to unknown */ }
|
|
89
|
+
return { version: 1, class: 'unknown', detail: `unreadable exit record: ${raw.slice(0, 120)}` };
|
|
90
|
+
}
|
|
91
|
+
// ─── Restart-loop containment (3.2) ──────────────────────────────────────────
|
|
92
|
+
//
|
|
93
|
+
// The child-session restart loop used to BE the service manager: systemd's
|
|
94
|
+
// `Restart=always RestartSec=2` and launchd's `KeepAlive`. Neither can count,
|
|
95
|
+
// so a program that dies instantly was relaunched every two seconds forever,
|
|
96
|
+
// and each relaunch was a fresh process with no memory of the previous one.
|
|
97
|
+
// The count now lives with the role, in its state directory, so it survives the
|
|
98
|
+
// runner being restarted and is consistent across both service managers.
|
|
99
|
+
export const RESTART_LEDGER_FILE = '.restart-ledger.json';
|
|
100
|
+
/** Consecutive immediate failures tolerated before the agent is held down. */
|
|
101
|
+
export const RESTART_FAIL_THRESHOLD = 5;
|
|
102
|
+
const RESTART_BACKOFF_BASE_MS = 2_000;
|
|
103
|
+
const RESTART_BACKOFF_MAX_MS = 60_000;
|
|
104
|
+
/** How often a held-down runner re-reads its ledger, so `up` can release it. */
|
|
105
|
+
const HELD_DOWN_POLL_MS = 5_000;
|
|
106
|
+
const emptyLedger = () => ({
|
|
107
|
+
version: 1,
|
|
108
|
+
consecutiveImmediateFailures: 0,
|
|
109
|
+
lastReason: '',
|
|
110
|
+
nextDelayMs: 0,
|
|
111
|
+
resumeDiscarded: false,
|
|
112
|
+
circuit: 'closed',
|
|
113
|
+
updatedAt: new Date(0).toISOString(),
|
|
114
|
+
});
|
|
115
|
+
/** Bounded exponential backoff for the nth consecutive immediate failure. */
|
|
116
|
+
export function backoffFor(consecutiveFailures) {
|
|
117
|
+
if (consecutiveFailures <= 0)
|
|
118
|
+
return 0;
|
|
119
|
+
return Math.min(RESTART_BACKOFF_BASE_MS * 2 ** (consecutiveFailures - 1), RESTART_BACKOFF_MAX_MS);
|
|
120
|
+
}
|
|
121
|
+
/** Read a role's restart ledger; a missing or corrupt one starts clean. */
|
|
122
|
+
export function readRestartLedger(dir) {
|
|
123
|
+
try {
|
|
124
|
+
const raw = JSON.parse(readFileSync(join(dir, RESTART_LEDGER_FILE), 'utf8'));
|
|
125
|
+
if (raw.version !== 1)
|
|
126
|
+
return emptyLedger();
|
|
127
|
+
return { ...emptyLedger(), ...raw, version: 1 };
|
|
128
|
+
}
|
|
129
|
+
catch {
|
|
130
|
+
return emptyLedger();
|
|
131
|
+
}
|
|
132
|
+
}
|
|
133
|
+
export function writeRestartLedger(dir, ledger) {
|
|
134
|
+
try {
|
|
135
|
+
mkdirSync(dir, { recursive: true });
|
|
136
|
+
writeFileSync(join(dir, RESTART_LEDGER_FILE), JSON.stringify(ledger, null, 2) + '\n');
|
|
137
|
+
}
|
|
138
|
+
catch { /* diagnostics must never take the role down */ }
|
|
139
|
+
}
|
|
140
|
+
/**
|
|
141
|
+
* Close the circuit and forget the failure streak. Called by an explicit
|
|
142
|
+
* operator `up`/`restart`, which is the only thing that may release a held-down
|
|
143
|
+
* role — a held-down runner polls this file, so a role can be released without
|
|
144
|
+
* bouncing its unit.
|
|
145
|
+
*/
|
|
146
|
+
export function resetRestartLedger(dir) {
|
|
147
|
+
if (!existsSync(dir))
|
|
148
|
+
return;
|
|
149
|
+
writeRestartLedger(dir, { ...emptyLedger(), updatedAt: new Date().toISOString() });
|
|
150
|
+
}
|
|
62
151
|
/** Filename spawnTemp writes into a temp agent dir to carry the fleet start-stagger. */
|
|
63
152
|
export const START_STAGGER_FILE = '.start-stagger-ms';
|
|
64
153
|
/** Read the start-stagger a temp agent was spawned with (0 if none / unreadable). */
|
|
@@ -151,7 +240,7 @@ function resolveConfigPath(dir, explicit) {
|
|
|
151
240
|
return undefined;
|
|
152
241
|
return readFileSync(marker, 'utf8').trim() || undefined;
|
|
153
242
|
}
|
|
154
|
-
/** One
|
|
243
|
+
/** One session lifecycle. `runSupervised` (or a one-shot caller) drives it. */
|
|
155
244
|
export async function runOnce(name, opts = {}, partialDeps = {}) {
|
|
156
245
|
const deps = { ...defaultDeps(), ...partialDeps };
|
|
157
246
|
const temp = opts.temp === true;
|
|
@@ -197,12 +286,10 @@ export async function runOnce(name, opts = {}, partialDeps = {}) {
|
|
|
197
286
|
// env prefix + exit capture in buildPaneCommand stay host-side (see §5.3).
|
|
198
287
|
let wrappedArgv = launch.argv;
|
|
199
288
|
if (role.isolation) {
|
|
200
|
-
|
|
201
|
-
|
|
202
|
-
|
|
203
|
-
const ctx = {
|
|
204
|
-
stateDir: dir, runCwd, home: home(), harness: role.harness, additionalWriteDirs: addDirs,
|
|
205
|
-
};
|
|
289
|
+
// The SAME context config validation and doctor judged (5.2): a policy
|
|
290
|
+
// checked against a different mount set than the one that launches is not a
|
|
291
|
+
// check at all.
|
|
292
|
+
const ctx = { ...isolationContextFor(role), stateDir: dir, runCwd };
|
|
206
293
|
const policy = resolveIsolation(role.isolation, ctx);
|
|
207
294
|
const sel = await selectIsolationBackend(policy, deps.exec); // throws on strict + unavailable
|
|
208
295
|
const degradedMarker = join(dir, '.isolation-degraded');
|
|
@@ -255,6 +342,13 @@ export async function runOnce(name, opts = {}, partialDeps = {}) {
|
|
|
255
342
|
let acpSession;
|
|
256
343
|
let control;
|
|
257
344
|
if (sessionBackend === 'acp') {
|
|
345
|
+
const perms = role.permissions ?? resolvePermissions(undefined, undefined);
|
|
346
|
+
// Say once, at startup, that this role will decide permission requests by
|
|
347
|
+
// itself. Without it the only trace of an auto-denied tool call is a turn
|
|
348
|
+
// that quietly did less than it was asked to.
|
|
349
|
+
if (perms.unattended === 'deny')
|
|
350
|
+
deps.log(`[${name}] permission policy: unattended=deny — with no console attached, ` +
|
|
351
|
+
`permission requests are automatically denied once each (reject_once) and the turn continues`);
|
|
258
352
|
acpSession = await AcpSession.start({
|
|
259
353
|
name,
|
|
260
354
|
argv: wrappedArgv,
|
|
@@ -262,7 +356,7 @@ export async function runOnce(name, opts = {}, partialDeps = {}) {
|
|
|
262
356
|
env: { ...launch.env, ...(role.env ?? {}) },
|
|
263
357
|
stateDir: dir,
|
|
264
358
|
mode,
|
|
265
|
-
permissions:
|
|
359
|
+
permissions: perms,
|
|
266
360
|
log: deps.log,
|
|
267
361
|
});
|
|
268
362
|
pid = acpSession.pid;
|
|
@@ -270,20 +364,27 @@ export async function runOnce(name, opts = {}, partialDeps = {}) {
|
|
|
270
364
|
control = new RoleControlServer(dir, acpSession, deps.log);
|
|
271
365
|
await control.start();
|
|
272
366
|
resolvedMonitorDeps.delivery = {
|
|
367
|
+
// A wake is only delivered when its turn TERMINATES successfully. A
|
|
368
|
+
// refusal or a cancellation reached the agent and was not acted on, so
|
|
369
|
+
// the monitor must keep its cursor and try again.
|
|
273
370
|
submit: async (text) => {
|
|
274
371
|
const result = await acpSession.submitPrompt(text);
|
|
275
|
-
return {
|
|
372
|
+
return { succeeded: result.succeeded, outcome: result.outcome, detail: result.detail };
|
|
276
373
|
},
|
|
277
374
|
};
|
|
278
375
|
const firstPrompt = mode === 'fresh'
|
|
279
376
|
? `Read and follow ${join(dir, 'briefing.md')} now.`
|
|
280
377
|
: adapter.vocabulary.restartPrompt(role.identity, join(dir, 'WORKLOG.md'), role);
|
|
378
|
+
// Wait for the first turn's TERMINAL result. An agent that accepts the
|
|
379
|
+
// startup prompt and then refuses it has not started; logging the role as
|
|
380
|
+
// up would hide a role that never read its briefing.
|
|
281
381
|
const started = await acpSession.submitPrompt(firstPrompt);
|
|
282
|
-
if (!started.
|
|
382
|
+
if (!started.succeeded) {
|
|
283
383
|
monitor?.stop();
|
|
284
384
|
await control.close();
|
|
285
385
|
await acpSession.close();
|
|
286
|
-
throw new Error(`[${name}] ACP
|
|
386
|
+
throw new Error(`[${name}] ACP startup prompt ${started.outcome}` +
|
|
387
|
+
`${started.detail ? `: ${started.detail}` : ''}`);
|
|
287
388
|
}
|
|
288
389
|
}
|
|
289
390
|
else {
|
|
@@ -316,18 +417,137 @@ export async function runOnce(name, opts = {}, partialDeps = {}) {
|
|
|
316
417
|
if (acpSession)
|
|
317
418
|
await acpSession.close();
|
|
318
419
|
const elapsed = (deps.now() - start) / 1000;
|
|
319
|
-
|
|
420
|
+
// Establish what actually happened before deciding anything. Absence of a
|
|
421
|
+
// record is `unknown` — except when the console itself is gone, which is a
|
|
422
|
+
// different event with a different consequence.
|
|
423
|
+
const exitRecord = acpSession
|
|
424
|
+
? acpSession.exitResult()
|
|
425
|
+
?? { version: 1, class: 'unknown', detail: 'the ACP agent stopped without reporting an exit' }
|
|
426
|
+
: readExitRecord(exitFile)
|
|
427
|
+
?? (await deps.tmux.has(name)
|
|
428
|
+
? { version: 1, class: 'unknown', detail: 'the pane process ended without writing an exit record' }
|
|
429
|
+
: { version: 1, class: 'session-destroyed', detail: `the tmux session '${name}' no longer exists` });
|
|
430
|
+
writeFileSync(exitFile, JSON.stringify({
|
|
431
|
+
...exitRecord, at: new Date(deps.now()).toISOString(), elapsedSecs: Number(elapsed.toFixed(1)),
|
|
432
|
+
}) + '\n');
|
|
433
|
+
let rotated = false;
|
|
320
434
|
const rotate = (why) => {
|
|
321
435
|
writeFileSync(sidFile, randomUUID() + '\n');
|
|
322
436
|
rmSync(bootedFile, { force: true });
|
|
437
|
+
rotated = true;
|
|
323
438
|
deps.log(`[${name}] ${why} -> rotated session-id; next start is FRESH`);
|
|
324
439
|
};
|
|
325
|
-
if (
|
|
440
|
+
if (exitRecord.class === 'clean' && adapter.exitPolicy.cleanExitIsFresh)
|
|
326
441
|
rotate(`clean exit (code 0)`);
|
|
327
|
-
else if (
|
|
328
|
-
|
|
442
|
+
else if (exitRecord.class === 'session-destroyed')
|
|
443
|
+
// Someone tore the console down; the agent did not fail. Rotating here
|
|
444
|
+
// would discard a live conversation for an operator action.
|
|
445
|
+
deps.log(`[${name}] ${exitRecord.detail} (${elapsed.toFixed(0)}s) -> next start RESUMES context`);
|
|
446
|
+
else if (mode === 'resume' && elapsed < adapter.exitPolicy.fastFailSecs) {
|
|
447
|
+
// Self-heal a poisoned resume — but only once per failure sequence. Rotating
|
|
448
|
+
// on every attempt would discard the conversation again and again while the
|
|
449
|
+
// real cause (a broken command, a missing binary) went unaddressed.
|
|
450
|
+
if (opts.allowResumeRotation === false)
|
|
451
|
+
deps.log(`[${name}] resume failed fast again (${elapsed.toFixed(0)}s, ${exitRecord.detail}) ` +
|
|
452
|
+
`-> resume state was already discarded once; keeping it`);
|
|
453
|
+
else
|
|
454
|
+
rotate(`resume failed fast (${elapsed.toFixed(0)}s, ${exitRecord.detail})`);
|
|
455
|
+
}
|
|
329
456
|
else
|
|
330
|
-
deps.log(`[${name}]
|
|
457
|
+
deps.log(`[${name}] ${exitRecord.detail} (${elapsed.toFixed(0)}s) -> next start RESUMES context`);
|
|
458
|
+
return { elapsedSecs: elapsed, exit: exitRecord, rotated, mode };
|
|
459
|
+
}
|
|
460
|
+
/**
|
|
461
|
+
* The persistent supervisor for one permanent role: run child sessions in a
|
|
462
|
+
* loop, count consecutive immediate failures across them, back off between
|
|
463
|
+
* attempts, and after `RESTART_FAIL_THRESHOLD` hold the agent down while
|
|
464
|
+
* staying alive — so the service manager has nothing to restart and cannot
|
|
465
|
+
* resume the two-second loop behind our back.
|
|
466
|
+
*
|
|
467
|
+
* `attempt` is injectable so the policy can be tested against a fake clock and
|
|
468
|
+
* fake child instead of real sessions.
|
|
469
|
+
*/
|
|
470
|
+
export async function runSupervised(name, opts = {}, partialDeps = {}, attempt = runOnce) {
|
|
471
|
+
const deps = { ...defaultDeps(), ...partialDeps };
|
|
472
|
+
const dir = agentDir(name);
|
|
473
|
+
mkdirSync(dir, { recursive: true });
|
|
474
|
+
const shouldStop = deps.shouldStop ?? (() => false);
|
|
475
|
+
const stamp = () => new Date(deps.now()).toISOString();
|
|
476
|
+
while (!shouldStop()) {
|
|
477
|
+
let ledger = readRestartLedger(dir);
|
|
478
|
+
if (ledger.circuit === 'open') {
|
|
479
|
+
// Held down. Stay alive — exiting would hand the role straight back to
|
|
480
|
+
// the service manager — and watch for an operator reset.
|
|
481
|
+
await deps.sleep(HELD_DOWN_POLL_MS);
|
|
482
|
+
continue;
|
|
483
|
+
}
|
|
484
|
+
let result;
|
|
485
|
+
try {
|
|
486
|
+
result = await attempt(name, { configPath: opts.configPath, allowResumeRotation: !ledger.resumeDiscarded }, deps);
|
|
487
|
+
}
|
|
488
|
+
catch (e) {
|
|
489
|
+
// A session that could not even start is an immediate failure like any
|
|
490
|
+
// other; it must count, or an unstartable role loops forever.
|
|
491
|
+
result = {
|
|
492
|
+
elapsedSecs: 0,
|
|
493
|
+
exit: { version: 1, class: 'unknown', detail: e instanceof Error ? e.message : String(e) },
|
|
494
|
+
rotated: false,
|
|
495
|
+
mode: 'fresh',
|
|
496
|
+
};
|
|
497
|
+
}
|
|
498
|
+
// Re-read: the attempt itself may have taken minutes, and an operator may
|
|
499
|
+
// have reset the ledger meanwhile.
|
|
500
|
+
ledger = readRestartLedger(dir);
|
|
501
|
+
const fastFailSecs = fastFailSecsFor(name, opts.configPath);
|
|
502
|
+
const immediate = result.elapsedSecs < fastFailSecs;
|
|
503
|
+
if (!immediate) {
|
|
504
|
+
// A session that ran for a while is not a restart loop, whatever ended it.
|
|
505
|
+
writeRestartLedger(dir, {
|
|
506
|
+
...emptyLedger(),
|
|
507
|
+
lastReason: result.exit.detail,
|
|
508
|
+
updatedAt: stamp(),
|
|
509
|
+
});
|
|
510
|
+
continue;
|
|
511
|
+
}
|
|
512
|
+
const failures = ledger.consecutiveImmediateFailures + 1;
|
|
513
|
+
const reason = `${result.exit.detail} after ${result.elapsedSecs.toFixed(1)}s`;
|
|
514
|
+
const next = {
|
|
515
|
+
version: 1,
|
|
516
|
+
consecutiveImmediateFailures: failures,
|
|
517
|
+
lastReason: reason,
|
|
518
|
+
nextDelayMs: backoffFor(failures),
|
|
519
|
+
resumeDiscarded: ledger.resumeDiscarded || result.rotated,
|
|
520
|
+
circuit: failures >= RESTART_FAIL_THRESHOLD ? 'open' : 'closed',
|
|
521
|
+
updatedAt: stamp(),
|
|
522
|
+
};
|
|
523
|
+
if (next.circuit === 'open') {
|
|
524
|
+
next.openedAt = stamp();
|
|
525
|
+
next.nextDelayMs = 0;
|
|
526
|
+
writeRestartLedger(dir, next);
|
|
527
|
+
deps.log(`[${name}] HELD DOWN after ${failures} immediate failures at ${next.openedAt} — ` +
|
|
528
|
+
`${reason}; the agent will not be restarted until: ours-fleet restart ${name}`);
|
|
529
|
+
continue;
|
|
530
|
+
}
|
|
531
|
+
writeRestartLedger(dir, next);
|
|
532
|
+
deps.log(`[${name}] immediate failure ${failures}/${RESTART_FAIL_THRESHOLD} (${reason}) ` +
|
|
533
|
+
`-> backing off ${next.nextDelayMs}ms`);
|
|
534
|
+
await deps.sleep(next.nextDelayMs);
|
|
535
|
+
}
|
|
536
|
+
return readRestartLedger(dir);
|
|
537
|
+
}
|
|
538
|
+
/**
|
|
539
|
+
* How short an attempt has to be to count as immediate. The role's harness
|
|
540
|
+
* decides; an unreadable config falls back to the common 20s so a broken config
|
|
541
|
+
* cannot disable the breaker.
|
|
542
|
+
*/
|
|
543
|
+
function fastFailSecsFor(name, configPath) {
|
|
544
|
+
try {
|
|
545
|
+
const role = findRole(loadConfig(configPath), name);
|
|
546
|
+
return getAdapter(role.harness).exitPolicy.fastFailSecs;
|
|
547
|
+
}
|
|
548
|
+
catch {
|
|
549
|
+
return 20;
|
|
550
|
+
}
|
|
331
551
|
}
|
|
332
552
|
/** Temp-agent entrypoint: run one session, then remove the temp dir. */
|
|
333
553
|
export async function runTemp(name, deps = {}) {
|
package/dist/session/acp.d.ts
CHANGED
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
import type { CommonPermissions } from '../config.js';
|
|
2
|
-
import type { SessionEvent, SessionHandle, SessionSnapshot, TurnResult } from './types.js';
|
|
2
|
+
import type { ExitRecord, QueuedPrompt, SessionEvent, SessionHandle, SessionSnapshot, TurnOutcome, TurnResult } from './types.js';
|
|
3
3
|
export interface AcpSessionOptions {
|
|
4
4
|
name: string;
|
|
5
5
|
argv: string[];
|
|
@@ -10,6 +10,12 @@ export interface AcpSessionOptions {
|
|
|
10
10
|
permissions: CommonPermissions;
|
|
11
11
|
log(line: string): void;
|
|
12
12
|
}
|
|
13
|
+
/**
|
|
14
|
+
* Classify an ACP `stopReason` into a terminal outcome. A refusal and a
|
|
15
|
+
* cancellation are the two ways a delivered prompt ends without being carried
|
|
16
|
+
* out; every other stop reason ran the turn to an end the agent chose.
|
|
17
|
+
*/
|
|
18
|
+
export declare function classifyStopReason(stopReason: string | undefined): TurnOutcome;
|
|
13
19
|
/**
|
|
14
20
|
* Persistent ACP v1 client. It is the sole owner of the agent's stdio; all
|
|
15
21
|
* human/automation attachment happens through the fleet role-control protocol.
|
|
@@ -27,22 +33,37 @@ export declare class AcpSession implements SessionHandle {
|
|
|
27
33
|
private readiness;
|
|
28
34
|
private lastError?;
|
|
29
35
|
private promptTail;
|
|
36
|
+
private queueDepth;
|
|
37
|
+
private exit;
|
|
30
38
|
private capabilities?;
|
|
31
39
|
private controllerCount;
|
|
32
40
|
private constructor();
|
|
33
41
|
static start(options: AcpSessionOptions): Promise<AcpSession>;
|
|
34
42
|
isAlive(): boolean;
|
|
35
43
|
snapshot(): SessionSnapshot;
|
|
44
|
+
/**
|
|
45
|
+
* Accept responsibility for a prompt, then return. The turn itself may run
|
|
46
|
+
* for minutes behind other queued turns; making an interactive caller wait
|
|
47
|
+
* for it is what turned a busy agent into a timeout and then into "dead".
|
|
48
|
+
*/
|
|
49
|
+
queuePrompt(text: string): Promise<QueuedPrompt>;
|
|
36
50
|
submitPrompt(text: string): Promise<TurnResult>;
|
|
37
51
|
interrupt(): Promise<void>;
|
|
38
52
|
respondPermission(permissionId: string, optionId: string): boolean;
|
|
39
53
|
eventsSince(seq: number): SessionEvent[];
|
|
40
54
|
subscribe(listener: (event: SessionEvent) => void): () => void;
|
|
41
55
|
setControllerAttached(attached: boolean): void;
|
|
56
|
+
exitResult(): ExitRecord | null;
|
|
42
57
|
close(): Promise<void>;
|
|
43
58
|
private initialize;
|
|
44
59
|
private runPrompt;
|
|
45
60
|
private requestPermission;
|
|
61
|
+
/**
|
|
62
|
+
* Resolve a permission request from policy alone and leave a record of it.
|
|
63
|
+
* Nothing else in the system can observe an automatic decision, so an
|
|
64
|
+
* unrecorded one is indistinguishable from a request that was never made.
|
|
65
|
+
*/
|
|
66
|
+
private settleAutomatically;
|
|
46
67
|
private withinAutomaticBoundary;
|
|
47
68
|
private recordUpdate;
|
|
48
69
|
private fail;
|
package/dist/session/acp.js
CHANGED
|
@@ -5,6 +5,19 @@ import { isAbsolute, join, relative, resolve } from 'node:path';
|
|
|
5
5
|
import { Readable, Writable } from 'node:stream';
|
|
6
6
|
import * as acp from '@agentclientprotocol/sdk';
|
|
7
7
|
import { SessionEvents } from './events.js';
|
|
8
|
+
import { SessionControlError, classifyChildExit, turnResult } from './types.js';
|
|
9
|
+
/**
|
|
10
|
+
* Classify an ACP `stopReason` into a terminal outcome. A refusal and a
|
|
11
|
+
* cancellation are the two ways a delivered prompt ends without being carried
|
|
12
|
+
* out; every other stop reason ran the turn to an end the agent chose.
|
|
13
|
+
*/
|
|
14
|
+
export function classifyStopReason(stopReason) {
|
|
15
|
+
switch (stopReason) {
|
|
16
|
+
case 'refusal': return 'refused';
|
|
17
|
+
case 'cancelled': return 'cancelled';
|
|
18
|
+
default: return 'completed';
|
|
19
|
+
}
|
|
20
|
+
}
|
|
8
21
|
/**
|
|
9
22
|
* Persistent ACP v1 client. It is the sole owner of the agent's stdio; all
|
|
10
23
|
* human/automation attachment happens through the fleet role-control protocol.
|
|
@@ -22,6 +35,8 @@ export class AcpSession {
|
|
|
22
35
|
readiness = 'starting';
|
|
23
36
|
lastError;
|
|
24
37
|
promptTail = Promise.resolve();
|
|
38
|
+
queueDepth = 0;
|
|
39
|
+
exit = null;
|
|
25
40
|
capabilities;
|
|
26
41
|
controllerCount = 0;
|
|
27
42
|
constructor(options, child, connection) {
|
|
@@ -33,9 +48,12 @@ export class AcpSession {
|
|
|
33
48
|
this.sessionFile = join(options.stateDir, '.acp-session-id');
|
|
34
49
|
child.stderr.on('data', chunk => options.log(`[${options.name}] acp: ${String(chunk).trimEnd()}`));
|
|
35
50
|
child.once('exit', (code, signal) => {
|
|
51
|
+
// Record the child's real exit code/signal. The tmux path can only see a
|
|
52
|
+
// shell's `$?`; here the truth is available, so keep it.
|
|
53
|
+
this.exit = classifyChildExit(code, signal);
|
|
36
54
|
if (this.readiness !== 'failed') {
|
|
37
55
|
this.readiness = 'failed';
|
|
38
|
-
this.lastError = `ACP agent
|
|
56
|
+
this.lastError = `ACP agent ${this.exit.detail}`;
|
|
39
57
|
}
|
|
40
58
|
this.events.emit('state', { status: 'failed', text: this.lastError });
|
|
41
59
|
});
|
|
@@ -88,10 +106,33 @@ export class AcpSession {
|
|
|
88
106
|
pendingPermissionId: this.pendingPermissions.keys().next().value,
|
|
89
107
|
};
|
|
90
108
|
}
|
|
91
|
-
|
|
92
|
-
|
|
93
|
-
|
|
94
|
-
|
|
109
|
+
/**
|
|
110
|
+
* Accept responsibility for a prompt, then return. The turn itself may run
|
|
111
|
+
* for minutes behind other queued turns; making an interactive caller wait
|
|
112
|
+
* for it is what turned a busy agent into a timeout and then into "dead".
|
|
113
|
+
*/
|
|
114
|
+
async queuePrompt(text) {
|
|
115
|
+
if (!this.sessionId || !this.isAlive())
|
|
116
|
+
throw new SessionControlError('offline', this.lastError ?? 'ACP session is offline');
|
|
117
|
+
const promptId = randomUUID();
|
|
118
|
+
const queuedBehind = this.queueDepth++;
|
|
119
|
+
const run = this.promptTail.then(() => this.runPrompt(text, promptId));
|
|
120
|
+
this.promptTail = run.then(() => undefined, () => undefined);
|
|
121
|
+
const completion = run.then(result => { this.queueDepth = Math.max(0, this.queueDepth - 1); return result; }, error => {
|
|
122
|
+
this.queueDepth = Math.max(0, this.queueDepth - 1);
|
|
123
|
+
return turnResult(false, 'failed', error?.message ?? String(error));
|
|
124
|
+
});
|
|
125
|
+
return { promptId, queuedBehind, completion };
|
|
126
|
+
}
|
|
127
|
+
async submitPrompt(text) {
|
|
128
|
+
try {
|
|
129
|
+
return await (await this.queuePrompt(text)).completion;
|
|
130
|
+
}
|
|
131
|
+
catch (error) {
|
|
132
|
+
if (error instanceof SessionControlError)
|
|
133
|
+
return turnResult(false, 'failed', error.message);
|
|
134
|
+
throw error;
|
|
135
|
+
}
|
|
95
136
|
}
|
|
96
137
|
async interrupt() {
|
|
97
138
|
if (!this.sessionId)
|
|
@@ -100,10 +141,19 @@ export class AcpSession {
|
|
|
100
141
|
}
|
|
101
142
|
respondPermission(permissionId, optionId) {
|
|
102
143
|
const pending = this.pendingPermissions.get(permissionId);
|
|
103
|
-
|
|
144
|
+
const chosen = pending?.options.find(option => option.optionId === optionId);
|
|
145
|
+
if (!pending || !chosen)
|
|
104
146
|
return false;
|
|
105
147
|
this.pendingPermissions.delete(permissionId);
|
|
106
148
|
pending.resolve({ outcome: { outcome: 'selected', optionId } });
|
|
149
|
+
this.events.emit('permission', {
|
|
150
|
+
permissionId,
|
|
151
|
+
status: 'completed',
|
|
152
|
+
decision: chosen.kind.startsWith('reject') ? 'denied' : 'allowed',
|
|
153
|
+
decisionSource: 'manual',
|
|
154
|
+
reason: `answered from an attached controller (${chosen.kind})`,
|
|
155
|
+
optionId,
|
|
156
|
+
});
|
|
107
157
|
this.readiness = 'running';
|
|
108
158
|
return true;
|
|
109
159
|
}
|
|
@@ -116,6 +166,9 @@ export class AcpSession {
|
|
|
116
166
|
setControllerAttached(attached) {
|
|
117
167
|
this.controllerCount = Math.max(0, this.controllerCount + (attached ? 1 : -1));
|
|
118
168
|
}
|
|
169
|
+
exitResult() {
|
|
170
|
+
return this.exit;
|
|
171
|
+
}
|
|
119
172
|
async close() {
|
|
120
173
|
for (const pending of this.pendingPermissions.values())
|
|
121
174
|
pending.resolve({ outcome: { outcome: 'cancelled' } });
|
|
@@ -166,11 +219,10 @@ export class AcpSession {
|
|
|
166
219
|
this.readiness = 'idle';
|
|
167
220
|
this.events.emit('state', { status: 'idle', text: `ACP session ${this.sessionId}` });
|
|
168
221
|
}
|
|
169
|
-
async runPrompt(text) {
|
|
222
|
+
async runPrompt(text, turnId = randomUUID()) {
|
|
170
223
|
if (!this.sessionId || !this.isAlive())
|
|
171
|
-
return
|
|
224
|
+
return turnResult(false, 'failed', this.lastError ?? 'ACP session is offline');
|
|
172
225
|
this.readiness = 'running';
|
|
173
|
-
const turnId = randomUUID();
|
|
174
226
|
this.events.emit('state', { turnId, status: 'running' });
|
|
175
227
|
try {
|
|
176
228
|
const response = await this.connection.agent.request(acp.methods.agent.session.prompt, {
|
|
@@ -180,12 +232,9 @@ export class AcpSession {
|
|
|
180
232
|
this.readiness = 'idle';
|
|
181
233
|
this.events.emit('turn_stop', { turnId, stopReason: response.stopReason });
|
|
182
234
|
this.events.emit('state', { status: 'idle' });
|
|
183
|
-
|
|
184
|
-
|
|
185
|
-
|
|
186
|
-
? 'refused'
|
|
187
|
-
: 'completed';
|
|
188
|
-
return { accepted: true, outcome, detail: response.stopReason };
|
|
235
|
+
// The prompt was accepted either way — the agent answered. Whether the
|
|
236
|
+
// turn SUCCEEDED is a separate question, and only `stopReason` answers it.
|
|
237
|
+
return turnResult(true, classifyStopReason(response.stopReason), response.stopReason);
|
|
189
238
|
}
|
|
190
239
|
catch (error) {
|
|
191
240
|
this.lastError = error?.message ?? String(error);
|
|
@@ -193,23 +242,35 @@ export class AcpSession {
|
|
|
193
242
|
this.events.emit('error', { turnId, text: this.lastError });
|
|
194
243
|
if (this.isAlive())
|
|
195
244
|
this.events.emit('state', { status: 'idle' });
|
|
196
|
-
return
|
|
245
|
+
return turnResult(false, 'failed', this.lastError);
|
|
197
246
|
}
|
|
198
247
|
}
|
|
199
248
|
requestPermission(params) {
|
|
200
|
-
|
|
249
|
+
// `kinds` is a PRIORITY order. Scanning the agent's option array instead
|
|
250
|
+
// (`options.find(o => kinds.includes(o.kind))`) hands the choice to whatever
|
|
251
|
+
// order the agent happened to list, which is exactly how an automatic denial
|
|
252
|
+
// could land on `reject_always`.
|
|
253
|
+
const choose = (kinds) => {
|
|
254
|
+
for (const kind of kinds) {
|
|
255
|
+
const option = params.options.find(o => o.kind === kind);
|
|
256
|
+
if (option)
|
|
257
|
+
return option;
|
|
258
|
+
}
|
|
259
|
+
return undefined;
|
|
260
|
+
};
|
|
201
261
|
if (this.options.permissions.approval === 'allow' && this.withinAutomaticBoundary(params)) {
|
|
202
262
|
const option = choose(['allow_always', 'allow_once']);
|
|
203
|
-
return Promise.resolve(option
|
|
204
|
-
? { outcome: { outcome: 'selected', optionId: option.optionId } }
|
|
205
|
-
: { outcome: { outcome: 'cancelled' } });
|
|
263
|
+
return Promise.resolve(this.settleAutomatically(params, option, 'allowed', 'permissions.approval=allow', `the request is inside the ${this.options.permissions.filesystem} boundary`));
|
|
206
264
|
}
|
|
207
|
-
|
|
208
|
-
|
|
209
|
-
|
|
210
|
-
|
|
211
|
-
|
|
212
|
-
|
|
265
|
+
const unattended = this.controllerCount === 0 && this.options.permissions.unattended === 'deny';
|
|
266
|
+
if (this.options.permissions.approval === 'deny' || unattended) {
|
|
267
|
+
// reject_once FIRST: `reject_always` teaches the agent a standing rule from
|
|
268
|
+
// a decision no human made, so one unattended denial would silently disable
|
|
269
|
+
// the tool for the rest of the session.
|
|
270
|
+
const option = choose(['reject_once', 'reject_always']);
|
|
271
|
+
return Promise.resolve(this.settleAutomatically(params, option, 'denied', unattended ? 'permissions.unattended=deny' : 'permissions.approval=deny', unattended
|
|
272
|
+
? 'no controller is attached, so the request cannot be shown to anyone'
|
|
273
|
+
: 'the role denies every permission request by policy'));
|
|
213
274
|
}
|
|
214
275
|
const permissionId = randomUUID();
|
|
215
276
|
this.readiness = 'awaiting_permission';
|
|
@@ -226,6 +287,29 @@ export class AcpSession {
|
|
|
226
287
|
this.pendingPermissions.set(permissionId, { options: params.options, resolve });
|
|
227
288
|
});
|
|
228
289
|
}
|
|
290
|
+
/**
|
|
291
|
+
* Resolve a permission request from policy alone and leave a record of it.
|
|
292
|
+
* Nothing else in the system can observe an automatic decision, so an
|
|
293
|
+
* unrecorded one is indistinguishable from a request that was never made.
|
|
294
|
+
*/
|
|
295
|
+
settleAutomatically(params, option, decision, policy, reason) {
|
|
296
|
+
const settled = option ? decision : 'cancelled';
|
|
297
|
+
this.events.emit('permission', {
|
|
298
|
+
permissionId: randomUUID(),
|
|
299
|
+
toolCallId: params.toolCall.toolCallId,
|
|
300
|
+
title: params.toolCall.title ?? 'Permission requested',
|
|
301
|
+
status: 'completed',
|
|
302
|
+
decision: settled,
|
|
303
|
+
decisionSource: 'automatic',
|
|
304
|
+
policy,
|
|
305
|
+
reason: option ? reason : `${reason}, but the agent offered no matching option`,
|
|
306
|
+
optionId: option?.optionId,
|
|
307
|
+
options: params.options.map(o => ({ optionId: o.optionId, name: o.name, kind: o.kind })),
|
|
308
|
+
});
|
|
309
|
+
return option
|
|
310
|
+
? { outcome: { outcome: 'selected', optionId: option.optionId } }
|
|
311
|
+
: { outcome: { outcome: 'cancelled' } };
|
|
312
|
+
}
|
|
229
313
|
withinAutomaticBoundary(params) {
|
|
230
314
|
const filesystem = this.options.permissions.filesystem;
|
|
231
315
|
if (filesystem === 'unrestricted')
|