humanish 0.111.2 → 0.111.3
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/ARCHITECTURE.md +1 -1
- package/CHANGELOG.md +17 -0
- package/dist/cli/commands/study-route-computer-use.js +3 -8
- package/dist/cli/commands/study-route-preview.js +4 -6
- package/dist/cli/commands/study-route-shared-world.js +6 -7
- package/dist/evidence/redaction.d.ts +2 -2
- package/dist/evidence/redaction.js +3 -3
- package/dist/routes/computer-use/bundle-parts.d.ts +1 -32
- package/dist/routes/computer-use/bundle-parts.js +5 -92
- package/dist/routes/computer-use/e2b-desktop/prepare.js +11 -4
- package/dist/routes/computer-use/fanout-bundle.js +3 -2
- package/dist/routes/computer-use/live-phase.js +3 -7
- package/dist/routes/computer-use/participant-deps.d.ts +46 -0
- package/dist/routes/computer-use/participant-deps.js +48 -0
- package/dist/routes/computer-use/participant-facts.d.ts +5 -4
- package/dist/routes/computer-use/participant-facts.js +37 -2
- package/dist/routes/computer-use/participant-feedback.d.ts +33 -0
- package/dist/routes/computer-use/participant-feedback.js +92 -0
- package/dist/routes/computer-use/participant-model.d.ts +11 -0
- package/dist/routes/computer-use/participant-model.js +15 -0
- package/dist/routes/computer-use/preflight.js +33 -65
- package/dist/routes/computer-use/result.js +13 -52
- package/dist/routes/computer-use/route.js +6 -7
- package/dist/routes/computer-use/setup.d.ts +2 -6
- package/dist/routes/computer-use/setup.js +47 -93
- package/dist/routes/computer-use/single-bundle.js +3 -2
- package/dist/routes/computer-use/types.d.ts +2 -2
- package/dist/routes/preview.js +2 -8
- package/dist/routes/scripted/bundle.d.ts +2 -6
- package/dist/routes/scripted/result.d.ts +2 -1
- package/dist/routes/scripted/result.js +4 -4
- package/dist/routes/scripted/route.js +13 -38
- package/dist/routes/scripted/setup.d.ts +2 -1
- package/dist/routes/scripted/setup.js +21 -28
- package/dist/routes/scripted/subject.d.ts +3 -18
- package/dist/routes/scripted/subject.js +22 -78
- package/dist/routes/scripted/types.d.ts +2 -2
- package/dist/routes/shared-world/bundle.js +36 -1
- package/dist/routes/shared-world/comms.d.ts +2 -1
- package/dist/routes/shared-world/comms.js +3 -7
- package/dist/routes/shared-world/handoff.d.ts +2 -2
- package/dist/routes/shared-world/handoff.js +1 -1
- package/dist/routes/shared-world/participant-specs.d.ts +4 -5
- package/dist/routes/shared-world/participant-specs.js +19 -25
- package/dist/routes/shared-world/plan.js +11 -7
- package/dist/routes/shared-world/provisioned.d.ts +2 -1
- package/dist/routes/shared-world/provisioned.js +37 -80
- package/dist/routes/shared-world/result.js +17 -35
- package/dist/routes/shared-world/route.js +33 -36
- package/dist/routes/shared-world/setup.d.ts +2 -3
- package/dist/routes/shared-world/setup.js +11 -29
- package/dist/routes/shared-world/types.d.ts +6 -15
- package/dist/routes/terminal/dry-run.js +14 -24
- package/dist/routes/terminal/live-finish.d.ts +0 -1
- package/dist/routes/terminal/live-finish.js +2 -2
- package/dist/routes/terminal/route.js +9 -15
- package/dist/routes/terminal/session.d.ts +2 -3
- package/dist/routes/terminal/session.js +28 -40
- package/dist/routes/terminal/trace.d.ts +1 -1
- package/dist/routes/terminal/types.d.ts +2 -2
- package/dist/run/run.d.ts +55 -1
- package/dist/run/run.js +38 -2
- package/dist/run/secrets.d.ts +22 -0
- package/dist/run/secrets.js +33 -0
- package/dist/run/study-result.d.ts +34 -0
- package/dist/run/study-result.js +16 -0
- package/dist/study/requirements.d.ts +60 -5
- package/dist/study/requirements.js +88 -2
- package/dist/study/run-study-events.d.ts +4 -3
- package/dist/study/run-study-events.js +9 -14
- package/dist/study/run-study-options.d.ts +4 -1
- package/dist/study/run-study-options.js +22 -32
- package/dist/substrates/e2b/subject-sandbox.d.ts +58 -0
- package/dist/substrates/e2b/subject-sandbox.js +94 -0
- package/docs/contracts/core.md +5 -4
- package/docs/contracts/feedback.md +12 -5
- package/package.json +1 -1
|
@@ -12,9 +12,9 @@ import { provisionLocalTreeSubject } from "../../subject/local-tree.js";
|
|
|
12
12
|
import { defaultSharedWorldPhaseSink } from "../../subject/steps.js";
|
|
13
13
|
import { phaseEvent } from "../../study/run-study-events.js";
|
|
14
14
|
import { loadE2BDesktopModule } from "../../substrates/e2b/sdk.js";
|
|
15
|
-
import {
|
|
16
|
-
import { acquireE2BDesktopSandbox, readE2BRelease } from "../../substrates/e2b/sandbox.js";
|
|
15
|
+
import { E2BSubjectSandbox, } from "../../substrates/e2b/subject-sandbox.js";
|
|
17
16
|
import { e2bShell } from "../../substrates/e2b/shell.js";
|
|
17
|
+
import { validatePreparedRunArtifactPaths } from "../../run/paths.js";
|
|
18
18
|
import { defaultPackLocalTree } from "../computer-use/local-tree-pack.js";
|
|
19
19
|
import { inboxRecipientFor, participantHasInboxRecipient, } from "../computer-use/participant-desktop.js";
|
|
20
20
|
import { resolveSubjectState } from "../computer-use/subject-projection.js";
|
|
@@ -25,8 +25,6 @@ import { buildConcurrentSharedWorldBundle, judgeSharedWorldRun } from "./bundle.
|
|
|
25
25
|
import { runCheckpointSnapshot } from "./checkpoints.js";
|
|
26
26
|
import { drainSubjectComms } from "./comms.js";
|
|
27
27
|
import { buildSubjectProvenance, hostOriginDigest, isTokenlessHost, servePort, } from "./provenance.js";
|
|
28
|
-
import { SANDBOX_TIMEOUT_BUFFER_MS, SUBJECT_PROVISION_BUDGET_MS, } from "../../substrates/e2b/lifetime.js";
|
|
29
|
-
import { DEFAULT_STATE_STEP_TIMEOUT_MS } from "../../subject/state.js";
|
|
30
28
|
import { resolveActorEntryUrl, participantRunDeps, startParticipantFlush, } from "./participant-specs.js";
|
|
31
29
|
import { CONCURRENT_SHARED_WORLD_PROVIDER_METADATA, } from "./types.js";
|
|
32
30
|
import { addressedRecipients } from "../../study/parse/comms.js";
|
|
@@ -46,17 +44,10 @@ class ObserverGateError extends Error {
|
|
|
46
44
|
class SubjectPlane {
|
|
47
45
|
commsInboxUrl;
|
|
48
46
|
subjectCommit;
|
|
49
|
-
|
|
50
|
-
subjectKilled = false;
|
|
51
|
-
subjectReleaseWarning;
|
|
47
|
+
subject;
|
|
52
48
|
getHostUrl;
|
|
53
|
-
// The subject desktop's host-side span and size, priced in the run's cost estimate.
|
|
54
|
-
subjectCreatedAtMs;
|
|
55
|
-
subjectTornDownAtMs;
|
|
56
|
-
subjectResources;
|
|
57
49
|
ctx;
|
|
58
50
|
setup;
|
|
59
|
-
subjectAllocation;
|
|
60
51
|
subjectDesktop;
|
|
61
52
|
subjectShell;
|
|
62
53
|
// The in-sandbox email catch on the one subject sandbox; drained at teardown. Undefined
|
|
@@ -77,6 +68,11 @@ class SubjectPlane {
|
|
|
77
68
|
constructor(ctx, setup) {
|
|
78
69
|
this.ctx = ctx;
|
|
79
70
|
this.setup = setup;
|
|
71
|
+
this.subject = new E2BSubjectSandbox({
|
|
72
|
+
warnings: ctx.warnings,
|
|
73
|
+
scrub: ctx.scrubKnownValues,
|
|
74
|
+
now: ctx.now,
|
|
75
|
+
});
|
|
80
76
|
this.disposeSignal = new Promise((resolve) => {
|
|
81
77
|
this.releaseDispose = resolve;
|
|
82
78
|
});
|
|
@@ -106,46 +102,35 @@ class SubjectPlane {
|
|
|
106
102
|
// The one subject sandbox: headless service host (no GUI participant). The subject env is
|
|
107
103
|
// provisioned here; the actor sandboxes get none of it. A custom desktop template (image) is
|
|
108
104
|
// honored on both the subject sandbox (here) and every actor sandbox (via runCuaParticipant, which
|
|
109
|
-
// reads the same config); absent keeps the byte-stable Sandbox.create(opts) default.
|
|
110
|
-
|
|
111
|
-
const subject = await acquireE2BDesktopSandbox({
|
|
105
|
+
// reads the same config); absent keeps the byte-stable Sandbox.create(opts) default.
|
|
106
|
+
this.subjectDesktop = await this.subject.acquire({
|
|
112
107
|
module: subjectModule,
|
|
113
|
-
|
|
114
|
-
|
|
115
|
-
|
|
116
|
-
|
|
117
|
-
|
|
118
|
-
|
|
119
|
-
|
|
120
|
-
|
|
121
|
-
|
|
122
|
-
|
|
123
|
-
|
|
124
|
-
|
|
125
|
-
|
|
126
|
-
|
|
127
|
-
|
|
128
|
-
...(subjectEnvNames.length > 0 || Object.keys(commsEnv).length > 0
|
|
129
|
-
? {
|
|
130
|
-
envs: {
|
|
131
|
-
...Object.fromEntries(subjectEnvNames.map((name) => [name, env[name]])),
|
|
132
|
-
...commsEnv,
|
|
133
|
-
},
|
|
134
|
-
}
|
|
135
|
-
: {}),
|
|
136
|
-
dpi: 96,
|
|
137
|
-
lifecycle: { onTimeout: "kill" },
|
|
108
|
+
apiKey: this.ctx.e2bApiKey,
|
|
109
|
+
requestTimeoutMs,
|
|
110
|
+
sessionTimeoutMs: timeoutMs,
|
|
111
|
+
seed: planeStateOf(plan)?.seed ?? [],
|
|
112
|
+
metadata: {
|
|
113
|
+
...CONCURRENT_SHARED_WORLD_PROVIDER_METADATA,
|
|
114
|
+
labId: plan.studyId,
|
|
115
|
+
topology: "shared-world",
|
|
116
|
+
topologyMode: "concurrent",
|
|
117
|
+
kind: "subject",
|
|
118
|
+
participantCount: String(plan.plane.participants.length),
|
|
119
|
+
},
|
|
120
|
+
envs: {
|
|
121
|
+
...Object.fromEntries(subjectEnvNames.map((name) => [name, env[name]])),
|
|
122
|
+
...commsEnv,
|
|
138
123
|
},
|
|
139
124
|
template: plan.residual.execution?.desktop?.template,
|
|
140
|
-
|
|
125
|
+
root: this.ctx.runPaths,
|
|
141
126
|
});
|
|
142
|
-
this.subjectDesktop = subject.sandbox;
|
|
143
127
|
this.subjectShell = e2bShell(this.subjectDesktop);
|
|
144
|
-
this.
|
|
145
|
-
|
|
146
|
-
|
|
147
|
-
|
|
148
|
-
|
|
128
|
+
if (this.ctx.input.prepareDesktop) {
|
|
129
|
+
await this.ctx.input.prepareDesktop(this.subjectDesktop, { kind: "subject" });
|
|
130
|
+
// The caller's hook runs with this process's file access, so the run directory is checked
|
|
131
|
+
// again before more evidence goes into it.
|
|
132
|
+
await validatePreparedRunArtifactPaths(this.ctx.runPaths);
|
|
133
|
+
}
|
|
149
134
|
}
|
|
150
135
|
// Start the in-sandbox email catch before the subject serve, so the app's send-API base URL
|
|
151
136
|
// (injected into its env at create) resolves the moment it boots. Fail closed if the catch can't
|
|
@@ -320,7 +305,6 @@ class SubjectPlane {
|
|
|
320
305
|
* Returns the comms thread's path when the drain wrote one.
|
|
321
306
|
*/
|
|
322
307
|
async teardown() {
|
|
323
|
-
const { warnings, scrubKnownValues } = this.ctx;
|
|
324
308
|
this.proberDisposed = true;
|
|
325
309
|
this.releaseDispose();
|
|
326
310
|
if (this.proberLoop) {
|
|
@@ -339,36 +323,9 @@ class SubjectPlane {
|
|
|
339
323
|
if (commsEmail && this.deployedComms?.ready && this.subjectShell) {
|
|
340
324
|
commsArtifactPath = await drainSubjectComms(this.ctx, commsEmail, this.subjectShell, this.deployedComms);
|
|
341
325
|
}
|
|
342
|
-
|
|
343
|
-
const released = await this.subjectAllocation.close();
|
|
344
|
-
const reading = readE2BRelease(released, {
|
|
345
|
-
label: "Subject sandbox",
|
|
346
|
-
scrub: scrubKnownValues,
|
|
347
|
-
costSpan: true,
|
|
348
|
-
});
|
|
349
|
-
this.subjectKilled = reading.released;
|
|
350
|
-
if (reading.warning)
|
|
351
|
-
warnings.push(reading.warning);
|
|
352
|
-
if (!reading.released)
|
|
353
|
-
this.subjectReleaseWarning = reading.warning ?? "Subject sandbox release is unconfirmed.";
|
|
354
|
-
// Without a kill method, or in E2B debug mode, no kill reached E2B, so there is no teardown
|
|
355
|
-
// time to record.
|
|
356
|
-
if (released.status !== "unconfirmed" || released.reason !== "release_unavailable")
|
|
357
|
-
this.subjectTornDownAtMs = this.ctx.now();
|
|
358
|
-
}
|
|
326
|
+
await this.subject.release();
|
|
359
327
|
return commsArtifactPath;
|
|
360
328
|
}
|
|
361
|
-
desktopUsage() {
|
|
362
|
-
if (this.subjectSandboxId === undefined)
|
|
363
|
-
return undefined;
|
|
364
|
-
return {
|
|
365
|
-
durationMs: this.subjectCreatedAtMs === undefined || this.subjectTornDownAtMs === undefined
|
|
366
|
-
? undefined
|
|
367
|
-
: Math.max(0, this.subjectTornDownAtMs - this.subjectCreatedAtMs),
|
|
368
|
-
observation: this.subjectResources,
|
|
369
|
-
killed: this.subjectKilled,
|
|
370
|
-
};
|
|
371
|
-
}
|
|
372
329
|
}
|
|
373
330
|
/** Writes the in-progress bundle, runs the caller's Observer gate and starts the participants'
|
|
374
331
|
* flush. */
|
|
@@ -476,10 +433,10 @@ export async function runProvisionedPlane(ctx, live, setup) {
|
|
|
476
433
|
actorResults,
|
|
477
434
|
runError,
|
|
478
435
|
subjectCommit: plane.subjectCommit,
|
|
479
|
-
subjectSandboxId: plane.
|
|
480
|
-
subjectKilled: plane.
|
|
481
|
-
subjectReleaseWarning: plane.
|
|
482
|
-
subjectDesktop: plane.desktopUsage(),
|
|
436
|
+
subjectSandboxId: plane.subject.sandboxId,
|
|
437
|
+
subjectKilled: plane.subject.killed,
|
|
438
|
+
subjectReleaseWarning: plane.subject.releaseWarning,
|
|
439
|
+
subjectDesktop: plane.subject.desktopUsage(),
|
|
483
440
|
getHostUrl: plane.getHostUrl,
|
|
484
441
|
commsArtifactPath,
|
|
485
442
|
};
|
|
@@ -3,14 +3,16 @@
|
|
|
3
3
|
import { redactText } from "../../evidence/redaction.js";
|
|
4
4
|
import { adapterScoreFailureMessage, applyBrowserScorer } from "../../study/adapter-extension.js";
|
|
5
5
|
import { attachObserverRuntimeStreamUrls } from "../../observer/render.js";
|
|
6
|
-
import { foldScorerFailures, judgeExecution, sandboxCleanupFailure, OUTCOME_POLICIES,
|
|
7
|
-
import {
|
|
6
|
+
import { foldScorerFailures, judgeExecution, sandboxCleanupFailure, OUTCOME_POLICIES, resultOk, sharedWorldShortfall, } from "../../run/judge.js";
|
|
7
|
+
import { validatePreparedRunArtifactPaths } from "../../run/paths.js";
|
|
8
|
+
import { participantExecutionFailures } from "../computer-use/participant-facts.js";
|
|
9
|
+
import { participantCapWarning } from "../computer-use/participant-model.js";
|
|
8
10
|
import { resolveSubjectState } from "../computer-use/subject-projection.js";
|
|
9
11
|
import { actorRunPassed, buildConcurrentSharedWorldBundle, judgeSharedWorldRun, maxSimultaneousWindows, } from "./bundle.js";
|
|
10
12
|
import { planeStateOf } from "./plan.js";
|
|
11
13
|
import { buildSubjectProvenance, hostOriginDigest } from "./provenance.js";
|
|
12
14
|
import {} from "./types.js";
|
|
13
|
-
import { studyResultIdentity } from "../../run/study-result.js";
|
|
15
|
+
import { refusedResult, studyResultIdentity } from "../../run/study-result.js";
|
|
14
16
|
import { plural } from "../../run/text.js";
|
|
15
17
|
/** The results of a run whose plane did not run (a dry run) or has not reported yet. */
|
|
16
18
|
export function emptyPlaneResults() {
|
|
@@ -143,10 +145,7 @@ function concurrentStudyError(args) {
|
|
|
143
145
|
* and the run id the caller asked for, if any. `actor` is the label when a refusal names none.
|
|
144
146
|
*/
|
|
145
147
|
export function concurrentStudyFailure(envelope) {
|
|
146
|
-
return (code, message, actorLabel) => ({
|
|
147
|
-
...studyResultIdentity("shared-world", envelope.studyId),
|
|
148
|
-
ok: false,
|
|
149
|
-
cwd: envelope.cwd,
|
|
148
|
+
return (code, message, actorLabel) => refusedResult("shared-world", { studyId: envelope.studyId, cwd: envelope.cwd, error: { code, message } }, {
|
|
150
149
|
actor: actorLabel ?? envelope.actor,
|
|
151
150
|
topology: "shared-world",
|
|
152
151
|
topologyMode: "concurrent",
|
|
@@ -155,43 +154,18 @@ export function concurrentStudyFailure(envelope) {
|
|
|
155
154
|
dryRun: envelope.dryRun,
|
|
156
155
|
runId: envelope.runId ?? "not-created",
|
|
157
156
|
roles: [],
|
|
158
|
-
warnings: [],
|
|
159
|
-
error: { code, message },
|
|
160
157
|
});
|
|
161
158
|
}
|
|
162
159
|
/**
|
|
163
|
-
* The run's execution failures before the Observer renders: a run error (the handoff, the plane)
|
|
164
|
-
*
|
|
165
|
-
* or that reported a disallowed item after its last request, and each sandbox whose release is
|
|
160
|
+
* The run's execution failures before the Observer renders: a run error (the handoff, the plane)
|
|
161
|
+
* first, then the participants' failures, then the subject sandbox when its release is
|
|
166
162
|
* unconfirmed. FinishedRun.renderObserver adds an Observer that did not render.
|
|
167
163
|
*/
|
|
168
164
|
function sharedWorldExecutionFailures(args) {
|
|
169
165
|
const { runId, runError, actorResults, subject } = args;
|
|
170
166
|
return [
|
|
171
167
|
...(runError === undefined ? [] : [{ kind: "run", message: runError }]),
|
|
172
|
-
...actorResults
|
|
173
|
-
.filter((result) => participantHarnessFailed(participantFactsOf(result.outcome)))
|
|
174
|
-
.map((result) => ({
|
|
175
|
-
kind: "harness",
|
|
176
|
-
message: `${result.spec.planned.id}: ${result.outcome.sessionError ?? result.outcome.session?.reason ?? "harness error"}`,
|
|
177
|
-
})),
|
|
178
|
-
...actorResults.flatMap((result) => result.outcome.providerCleanupError === undefined
|
|
179
|
-
? []
|
|
180
|
-
: [
|
|
181
|
-
{
|
|
182
|
-
kind: "provider-cleanup",
|
|
183
|
-
message: `${result.spec.planned.id}: ${result.outcome.providerCleanupError}`,
|
|
184
|
-
},
|
|
185
|
-
]),
|
|
186
|
-
...actorResults.flatMap((result) => result.outcome.providerPolicyError === undefined
|
|
187
|
-
? []
|
|
188
|
-
: [
|
|
189
|
-
{
|
|
190
|
-
kind: "provider-policy",
|
|
191
|
-
message: `${result.spec.planned.id}: ${result.outcome.providerPolicyError}`,
|
|
192
|
-
},
|
|
193
|
-
]),
|
|
194
|
-
...unreleasedSandboxFailures(actorResults.map((result) => result.outcome), runId),
|
|
168
|
+
...participantExecutionFailures(actorResults.map((result) => result.outcome), runId),
|
|
195
169
|
...(subject.subjectSandboxId === undefined || subject.subjectKilled
|
|
196
170
|
? []
|
|
197
171
|
: [sandboxCleanupFailure("subject", subject.subjectReleaseWarning, runId)]),
|
|
@@ -230,6 +204,9 @@ export async function finishConcurrentRun(ctx, live, results, plane) {
|
|
|
230
204
|
for (const result of actorResults) {
|
|
231
205
|
warnings.push(...result.outcome.warnings);
|
|
232
206
|
}
|
|
207
|
+
const capWarning = participantCapWarning(plan.caps, participantCount, "each participant's model spend");
|
|
208
|
+
if (capWarning !== undefined)
|
|
209
|
+
warnings.push(capWarning);
|
|
233
210
|
const bundleArgs = {
|
|
234
211
|
run,
|
|
235
212
|
plan,
|
|
@@ -293,6 +270,11 @@ export async function finishConcurrentRun(ctx, live, results, plane) {
|
|
|
293
270
|
policy,
|
|
294
271
|
});
|
|
295
272
|
const observer = await finished.renderObserver();
|
|
273
|
+
// A caller's renderer runs with this process's file access, so after one that reports success
|
|
274
|
+
// the run directory is checked again before the result points anyone at it. A failed render is
|
|
275
|
+
// already an evidence failure, and its error code stays the run's diagnosis.
|
|
276
|
+
if (observer.ok)
|
|
277
|
+
await validatePreparedRunArtifactPaths(finished.paths);
|
|
296
278
|
if (observer.ok && live.observer) {
|
|
297
279
|
attachObserverRuntimeStreamUrls(observer, live.streamUrls);
|
|
298
280
|
}
|
|
@@ -33,15 +33,12 @@
|
|
|
33
33
|
// requires the author attestation subject.exposure: synthetic. This is author-trust + a provenance
|
|
34
34
|
// gate; humanish cannot tell synthetic from real data, so it makes no claim about real data.
|
|
35
35
|
import path from "node:path";
|
|
36
|
-
import {
|
|
37
|
-
import { scrubLiterals } from "../../evidence/redaction.js";
|
|
38
|
-
import { describeMissingKeys } from "../../keys/key-resolution.js";
|
|
36
|
+
import { firstLiveRefusal, keysCheck, localAgentCheck, subjectEnvCheck, } from "../../study/requirements.js";
|
|
39
37
|
import {} from "../../run/run.js";
|
|
38
|
+
import { RunSecrets } from "../../run/secrets.js";
|
|
40
39
|
import { admitRoute, completeRefusalAnalysis, } from "../../run/route-shell.js";
|
|
41
|
-
import { makeCuaRunBudget } from "../computer-use/participant-model.js";
|
|
42
40
|
import { runExternalPublicPlane } from "./external-public.js";
|
|
43
41
|
import { sharedWorldDescriptorOf } from "./plan.js";
|
|
44
|
-
import { localAgentRefusal } from "../../actors/local-agent/readiness.js";
|
|
45
42
|
import { runProvisionedPlane } from "./provisioned.js";
|
|
46
43
|
import { concurrentStudyFailure, finishConcurrentRun } from "./result.js";
|
|
47
44
|
import { prepareConcurrentRun } from "./setup.js";
|
|
@@ -99,29 +96,35 @@ async function admitSharedWorldRun(plan, input) {
|
|
|
99
96
|
if (plan.dryRun)
|
|
100
97
|
return admitted;
|
|
101
98
|
const env = input.env ?? process.env;
|
|
102
|
-
const
|
|
99
|
+
const { requirements } = plan;
|
|
100
|
+
// The plan prices a cap before these, and its refusal wins over theirs.
|
|
101
|
+
const refusal = await firstLiveRefusal([
|
|
102
|
+
// The plan lists OPENAI_API_KEY for an openai brain's participants and the external-public
|
|
103
|
+
// plane's lobby-code reader; a local-agent brain's participants run on the operator's
|
|
104
|
+
// signed-in agent instead. Only on the provisioned plane does a local agent remove the key,
|
|
105
|
+
// so only there does the refusal suggest one.
|
|
106
|
+
() => keysCheck({
|
|
107
|
+
requirements,
|
|
108
|
+
env,
|
|
109
|
+
code: "HUMANISH_SHARED_WORLD_KEYS_MISSING",
|
|
110
|
+
need: (names) => `Live concurrent shared-world studies need ${names} in the environment (values are never persisted).`,
|
|
111
|
+
suggestLocalAgent: plan.plane.kind === "provisioned",
|
|
112
|
+
}),
|
|
113
|
+
() => localAgentCheck({
|
|
114
|
+
brain: plan.brain,
|
|
115
|
+
env,
|
|
116
|
+
caps: plan.caps,
|
|
117
|
+
codes: LOCAL_AGENT_REFUSAL_CODES,
|
|
118
|
+
}),
|
|
119
|
+
() => subjectEnvCheck({ requirements, env, code: "HUMANISH_SHARED_WORLD_SUBJECT_ENV_MISSING" }),
|
|
120
|
+
]);
|
|
121
|
+
if (refusal === undefined)
|
|
122
|
+
return admitted;
|
|
123
|
+
const fail = sharedWorldFailure(plan, input);
|
|
124
|
+
return {
|
|
103
125
|
ok: false,
|
|
104
|
-
result:
|
|
105
|
-
}
|
|
106
|
-
// The plan lists OPENAI_API_KEY for an openai brain's participants and the external-public
|
|
107
|
-
// plane's lobby-code reader; a local-agent brain's participants run on the operator's signed-in
|
|
108
|
-
// agent instead.
|
|
109
|
-
const missing = missingKeys(plan.requirements, env);
|
|
110
|
-
if (missing.length > 0) {
|
|
111
|
-
return fail("HUMANISH_SHARED_WORLD_KEYS_MISSING", `Live concurrent shared-world studies need ${missing.join(" and ")} in the environment (values are never persisted). ${describeMissingKeys(missing, env)}`);
|
|
112
|
-
}
|
|
113
|
-
if (plan.brain.kind === "local-agent") {
|
|
114
|
-
// A missing or signed-out agent found after the participants' desktops are paid for is the same
|
|
115
|
-
// news at the worst moment.
|
|
116
|
-
const refusal = await localAgentRefusal({ agent: plan.brain.agent, env, caps: plan.caps });
|
|
117
|
-
if (refusal)
|
|
118
|
-
return fail(LOCAL_AGENT_REFUSAL_CODES[refusal.kind], refusal.message);
|
|
119
|
-
}
|
|
120
|
-
const unsetSubjectEnv = missingSubjectEnv(plan.requirements, env);
|
|
121
|
-
if (unsetSubjectEnv.length > 0) {
|
|
122
|
-
return fail("HUMANISH_SHARED_WORLD_SUBJECT_ENV_MISSING", `subject.env declares ${unsetSubjectEnv.join(", ")} but the environment does not provide ${unsetSubjectEnv.length === 1 ? "it" : "them"} (pass via --dotenv; values are never persisted).`);
|
|
123
|
-
}
|
|
124
|
-
return admitted;
|
|
126
|
+
result: fail(refusal.code, refusal.message, sharedWorldDescriptorOf(plan.actor).id),
|
|
127
|
+
};
|
|
125
128
|
}
|
|
126
129
|
/** The route's envelope for a run that stops before its bundle. */
|
|
127
130
|
function sharedWorldFailure(plan, input) {
|
|
@@ -136,15 +139,12 @@ function sharedWorldFailure(plan, input) {
|
|
|
136
139
|
});
|
|
137
140
|
}
|
|
138
141
|
async function runPlanInScope(plan, input, config, scope) {
|
|
139
|
-
const { dryRun } = plan;
|
|
140
142
|
const requestedCwd = path.resolve(input.cwd);
|
|
141
143
|
const deps = input.deps ?? {};
|
|
142
144
|
const env = input.env ?? process.env;
|
|
143
145
|
const fail = sharedWorldFailure(plan, input);
|
|
144
146
|
const descriptor = sharedWorldDescriptorOf(plan.actor);
|
|
145
147
|
const planeClass = plan.plane.kind === "external-public" ? "external-public" : "provisioned-getHost";
|
|
146
|
-
const { maxTotalUsd } = plan.caps;
|
|
147
|
-
const runBudget = !dryRun && maxTotalUsd !== undefined ? makeCuaRunBudget(maxTotalUsd) : undefined;
|
|
148
148
|
// The provisioned plane's subject; the external-public plane has none.
|
|
149
149
|
const subject = plan.plane.kind === "provisioned" ? plan.plane.subject : undefined;
|
|
150
150
|
const serve = subject?.serve;
|
|
@@ -155,13 +155,12 @@ async function runPlanInScope(plan, input, config, scope) {
|
|
|
155
155
|
const runSession = deps.runSession ?? descriptor.runSession;
|
|
156
156
|
const openaiApiKey = env.OPENAI_API_KEY?.trim() ?? "";
|
|
157
157
|
const e2bApiKey = env.E2B_API_KEY?.trim() ?? "";
|
|
158
|
-
const
|
|
158
|
+
const secrets = new RunSecrets([
|
|
159
159
|
openaiApiKey,
|
|
160
160
|
e2bApiKey,
|
|
161
161
|
...subjectEnvNames.map((name) => env[name] ?? ""),
|
|
162
162
|
...checkpoints.flatMap((probe) => probe.redact ?? []),
|
|
163
|
-
]
|
|
164
|
-
const scrubKnownValues = scrubLiterals(knownSecretValues);
|
|
163
|
+
]);
|
|
165
164
|
const redactRepoLabel = plan.residual.policies?.redactRepos ?? subjectEnvNames.includes("GITHUB_TOKEN");
|
|
166
165
|
const publicRepo = redactRepoLabel ? "repo-01" : subjectRepo;
|
|
167
166
|
const hasGithubToken = subjectEnvNames.includes("GITHUB_TOKEN");
|
|
@@ -174,7 +173,6 @@ async function runPlanInScope(plan, input, config, scope) {
|
|
|
174
173
|
env,
|
|
175
174
|
descriptor,
|
|
176
175
|
planeClass,
|
|
177
|
-
runBudget,
|
|
178
176
|
runSession,
|
|
179
177
|
serve,
|
|
180
178
|
localTreeRoute,
|
|
@@ -183,8 +181,7 @@ async function runPlanInScope(plan, input, config, scope) {
|
|
|
183
181
|
checkpoints,
|
|
184
182
|
openaiApiKey,
|
|
185
183
|
e2bApiKey,
|
|
186
|
-
|
|
187
|
-
scrubKnownValues,
|
|
184
|
+
secrets,
|
|
188
185
|
publicRepo,
|
|
189
186
|
hasGithubToken,
|
|
190
187
|
fail,
|
|
@@ -1,6 +1,7 @@
|
|
|
1
1
|
import type { CuaActorDescriptor } from "../../actors/registry.js";
|
|
2
2
|
import type { StudyConfig, StudySubjectServe, StudySubjectStateCheckpoint } from "../../study/types.js";
|
|
3
3
|
import type { RunScope } from "../../run/run.js";
|
|
4
|
+
import type { RunSecrets } from "../../run/secrets.js";
|
|
4
5
|
import type { ConcurrentSharedWorldStudyErrorCode, ConcurrentSharedWorldStudyResult, ConcurrentSharedWorldPlaneClass, FinishFacts, LiveParticipants, PlaneContext, PlaneResults, PlaneSelection, SharedWorldRunInput } from "./types.js";
|
|
5
6
|
import type { StudyDeps } from "../../study/study-deps.js";
|
|
6
7
|
import type { SharedWorldPlan } from "../../study/plan-types.js";
|
|
@@ -16,7 +17,6 @@ interface AdmittedStudy {
|
|
|
16
17
|
env: Record<string, string | undefined>;
|
|
17
18
|
descriptor: CuaActorDescriptor;
|
|
18
19
|
planeClass: ConcurrentSharedWorldPlaneClass;
|
|
19
|
-
runBudget: PlaneContext["runBudget"];
|
|
20
20
|
runSession: PlaneContext["runSession"];
|
|
21
21
|
serve: StudySubjectServe | undefined;
|
|
22
22
|
localTreeRoute: boolean;
|
|
@@ -25,8 +25,7 @@ interface AdmittedStudy {
|
|
|
25
25
|
checkpoints: StudySubjectStateCheckpoint[];
|
|
26
26
|
openaiApiKey: string;
|
|
27
27
|
e2bApiKey: string;
|
|
28
|
-
|
|
29
|
-
scrubKnownValues: (text: string) => string;
|
|
28
|
+
secrets: RunSecrets;
|
|
30
29
|
publicRepo: string;
|
|
31
30
|
hasGithubToken: boolean;
|
|
32
31
|
fail: (code: ConcurrentSharedWorldStudyErrorCode, message: string, actorLabel?: string) => ConcurrentSharedWorldStudyResult;
|
|
@@ -1,8 +1,7 @@
|
|
|
1
1
|
// Starts a concurrent shared-world run and prepares what its plane reads: the physical project,
|
|
2
2
|
// the run, the participant specs, comms, the packed tree and email receiving.
|
|
3
|
-
import { randomBytes } from "node:crypto";
|
|
4
3
|
import { realpath } from "node:fs/promises";
|
|
5
|
-
import {
|
|
4
|
+
import {} from "../../run/bundle.js";
|
|
6
5
|
import { prepareSelectedOutputDirectory } from "../../run/contained-output.js";
|
|
7
6
|
import { renderConcurrentReviewMarkdown } from "./bundle.js";
|
|
8
7
|
import { seedRecipeDigest } from "./checkpoints.js";
|
|
@@ -15,10 +14,6 @@ import { planeStateOf } from "./plan.js";
|
|
|
15
14
|
import path from "node:path";
|
|
16
15
|
import { e2bRequestTimeoutMs } from "../../substrates/e2b/lifetime.js";
|
|
17
16
|
const DEFAULT_PROBER_CADENCE_MS = 1000;
|
|
18
|
-
function makeRunId() {
|
|
19
|
-
const stamp = new Date().toISOString().replace(/[:.]/g, "-");
|
|
20
|
-
return `concurrent-shared-world-${stamp}-${randomBytes(4).toString("hex")}`;
|
|
21
|
-
}
|
|
22
17
|
/** The provisioned plane's setup, or undefined when the study declares no `subject.serve`. */
|
|
23
18
|
function provisionedSetup(study, prepared) {
|
|
24
19
|
if (!study.serve)
|
|
@@ -49,16 +44,11 @@ async function bindPhysicalProject(requestedCwd) {
|
|
|
49
44
|
.physicalPath;
|
|
50
45
|
}
|
|
51
46
|
function startConcurrentRun(study, cwd, scope) {
|
|
52
|
-
|
|
53
|
-
return scope.startRun({
|
|
47
|
+
return scope.startRun(study.plan, study.input, {
|
|
54
48
|
cwd,
|
|
55
|
-
|
|
56
|
-
mintRunId: makeRunId,
|
|
57
|
-
mode: plan.dryRun ? "dry-run" : "live",
|
|
58
|
-
study: plan.study,
|
|
59
|
-
warnings: plan.warnings,
|
|
49
|
+
prefix: "concurrent-shared-world",
|
|
60
50
|
renderReview: renderConcurrentReviewMarkdown,
|
|
61
|
-
|
|
51
|
+
secrets: study.secrets,
|
|
62
52
|
});
|
|
63
53
|
}
|
|
64
54
|
/**
|
|
@@ -67,8 +57,8 @@ function startConcurrentRun(study, cwd, scope) {
|
|
|
67
57
|
*/
|
|
68
58
|
export async function prepareConcurrentRun(study, scope) {
|
|
69
59
|
const { plan, input, config, requestedCwd, deps, env, descriptor, planeClass, fail } = study;
|
|
70
|
-
const {
|
|
71
|
-
const { openaiApiKey, e2bApiKey,
|
|
60
|
+
const { runSession, localTreeRoute, subjectEnvNames, publicRepo } = study;
|
|
61
|
+
const { openaiApiKey, e2bApiKey, secrets } = study;
|
|
72
62
|
const { dryRun, concurrency } = plan;
|
|
73
63
|
const cwd = await bindPhysicalProject(requestedCwd);
|
|
74
64
|
const warnings = [];
|
|
@@ -81,7 +71,7 @@ export async function prepareConcurrentRun(study, scope) {
|
|
|
81
71
|
// Pack the working tree once per run, on the host, before any sandbox exists: a packing
|
|
82
72
|
// failure fails the run closed without sandbox cost. Dry-run packs nothing.
|
|
83
73
|
const packed = localTreeRoute && !dryRun
|
|
84
|
-
? await packSubjectTree(cwd, plan.residual, deps,
|
|
74
|
+
? await packSubjectTree(cwd, plan.residual, deps, secrets.scrub)
|
|
85
75
|
: { ok: true, archive: undefined, buffer: undefined };
|
|
86
76
|
if (!packed.ok) {
|
|
87
77
|
return {
|
|
@@ -95,7 +85,7 @@ export async function prepareConcurrentRun(study, scope) {
|
|
|
95
85
|
if (!started.ok)
|
|
96
86
|
return { ok: false, result: fail(started.code, started.message, descriptor.id) };
|
|
97
87
|
const { run } = started;
|
|
98
|
-
const { runId, createdAt, paths: runPaths } = run;
|
|
88
|
+
const { runId, createdAt, paths: runPaths, source } = run;
|
|
99
89
|
const artifactRoot = runPaths.absoluteRunRoot;
|
|
100
90
|
const timeoutMs = plan.sessionTimeoutMs ?? defaultSessionTimeoutMs(plan);
|
|
101
91
|
const requestTimeoutMs = e2bRequestTimeoutMs(env);
|
|
@@ -104,15 +94,9 @@ export async function prepareConcurrentRun(study, scope) {
|
|
|
104
94
|
const now = deps.now ?? Date.now;
|
|
105
95
|
const proberCadenceMs = deps.proberCadenceMs ?? DEFAULT_PROBER_CADENCE_MS;
|
|
106
96
|
const seedDigest = seedRecipeDigest(planeStateOf(plan));
|
|
107
|
-
const source = await buildRunSource({
|
|
108
|
-
capturedAt: createdAt,
|
|
109
|
-
cwd,
|
|
110
|
-
humanishSource: "present",
|
|
111
|
-
packageName: "humanish",
|
|
112
|
-
});
|
|
113
97
|
const stateStepRecords = [];
|
|
114
98
|
const stateSnapshots = [];
|
|
115
|
-
const actorSpecs = await buildParticipantSpecs(plan.plane.participants, cwd,
|
|
99
|
+
const actorSpecs = await buildParticipantSpecs(plan.plane.participants, cwd, run.secrets.scrub);
|
|
116
100
|
const results = emptyPlaneResults();
|
|
117
101
|
const live = { streamUrls: [] };
|
|
118
102
|
const subjectComms = subjectCommsOf(plan.residual, planeClass);
|
|
@@ -124,7 +108,7 @@ export async function prepareConcurrentRun(study, scope) {
|
|
|
124
108
|
env,
|
|
125
109
|
participants: actorSpecs.map((spec) => spec.planned.id),
|
|
126
110
|
runPaths,
|
|
127
|
-
|
|
111
|
+
secrets: run.secrets,
|
|
128
112
|
dryRun,
|
|
129
113
|
});
|
|
130
114
|
if (!email.ok) {
|
|
@@ -148,12 +132,10 @@ export async function prepareConcurrentRun(study, scope) {
|
|
|
148
132
|
deps,
|
|
149
133
|
env,
|
|
150
134
|
concurrency,
|
|
151
|
-
runBudget,
|
|
152
135
|
runSession,
|
|
153
136
|
openaiApiKey,
|
|
154
137
|
e2bApiKey,
|
|
155
|
-
scrubKnownValues,
|
|
156
|
-
knownSecretValues,
|
|
138
|
+
scrubKnownValues: run.secrets.scrub,
|
|
157
139
|
cwd,
|
|
158
140
|
run,
|
|
159
141
|
runId,
|
|
@@ -7,7 +7,7 @@ import type { AutomaticAnalysisResult } from "../../analysis/automatic-completio
|
|
|
7
7
|
import type { CuaActorDescriptor } from "../../actors/registry.js";
|
|
8
8
|
import type { CommsReceivingRun } from "../../comms/receiving.js";
|
|
9
9
|
import type { StudyCommsEmail, StudyCommsExternal, StudyConfig } from "../../study/types.js";
|
|
10
|
-
import type {
|
|
10
|
+
import type { SubjectDesktopUsage } from "../../substrates/e2b/subject-sandbox.js";
|
|
11
11
|
import type { ObserverResult } from "../../observer/render.js";
|
|
12
12
|
import type { ObserverRuntimeStreamUrl } from "../../observer/run-routes.js";
|
|
13
13
|
import type { BundleRun, RunBundle, RunScorerProvenance, RunSubjectProvenance, RunSubjectStateStepRecord } from "../../run/bundle.js";
|
|
@@ -15,10 +15,10 @@ import type { RunScope } from "../../run/run.js";
|
|
|
15
15
|
import type { SharedWorldStateSnapshot } from "../../run/shared-world-evidence.js";
|
|
16
16
|
import type { LocalTreeArchive } from "../../subject/local-tree-archive.js";
|
|
17
17
|
import type { LiveTraceFlush } from "../computer-use/live-flush.js";
|
|
18
|
-
import type { CuaParticipantDeps, DesktopParticipantRun,
|
|
18
|
+
import type { CuaParticipantDeps, DesktopParticipantRun, ParticipantRunOutcome } from "../computer-use/types.js";
|
|
19
19
|
import type { ProvisionedPlaneSetup } from "./provisioned.js";
|
|
20
20
|
import type { SharedWorldPlan } from "../../study/plan-types.js";
|
|
21
|
-
import { type StudyResultIdentity } from "../../run/study-result.js";
|
|
21
|
+
import { type StudyResultErrorCode, type StudyResultIdentity } from "../../run/study-result.js";
|
|
22
22
|
export declare const CONCURRENT_SHARED_WORLD_PROVIDER_METADATA: {
|
|
23
23
|
readonly mode: "concurrent-shared-world-lab";
|
|
24
24
|
readonly tool: "humanish";
|
|
@@ -54,13 +54,13 @@ export interface SharedWorldRunInput {
|
|
|
54
54
|
* core-stamped onto the bundle as evidence. Absent for library callers. */
|
|
55
55
|
scorerProvenance?: RunScorerProvenance;
|
|
56
56
|
}
|
|
57
|
-
export type ConcurrentSharedWorldStudyErrorCode =
|
|
57
|
+
export type ConcurrentSharedWorldStudyErrorCode = StudyResultErrorCode | "HUMANISH_SHARED_WORLD_FAILED" | "HUMANISH_SHARED_WORLD_ACTOR_UNSUPPORTED" | "HUMANISH_SHARED_WORLD_INVALID" | "HUMANISH_SHARED_WORLD_KEYS_MISSING"
|
|
58
58
|
/** A local-agent brain's CLI is not on `PATH`. */
|
|
59
59
|
| "HUMANISH_SHARED_WORLD_AGENT_MISSING"
|
|
60
60
|
/** A local-agent brain's CLI is signed out, or could not report its sign-in status. */
|
|
61
61
|
| "HUMANISH_SHARED_WORLD_AGENT_SIGNIN_REQUIRED"
|
|
62
62
|
/** A dollar cap that cannot be priced: an unpriced model, or a ChatGPT-account Codex brain. */
|
|
63
|
-
| "HUMANISH_SHARED_WORLD_UNPRICED_CAP" | "HUMANISH_SHARED_WORLD_SUBJECT_ENV_MISSING" | "HUMANISH_SHARED_WORLD_GETHOST_UNAVAILABLE" | "HUMANISH_SHARED_WORLD_HANDOFF_TIMEOUT"
|
|
63
|
+
| "HUMANISH_SHARED_WORLD_UNPRICED_CAP" | "HUMANISH_SHARED_WORLD_SUBJECT_ENV_MISSING" | "HUMANISH_SHARED_WORLD_GETHOST_UNAVAILABLE" | "HUMANISH_SHARED_WORLD_HANDOFF_TIMEOUT"
|
|
64
64
|
/** A declared adopter-hosted comms catch did not answer as a humanish catch. Fail closed
|
|
65
65
|
* before any actor spend, since the funnel would silently collect nothing. */
|
|
66
66
|
| "HUMANISH_SHARED_WORLD_COMMS_CATCH_UNREACHABLE"
|
|
@@ -134,12 +134,6 @@ export interface ConcurrentSharedWorldStudyResult extends AutomaticAnalysisResul
|
|
|
134
134
|
message: string;
|
|
135
135
|
};
|
|
136
136
|
}
|
|
137
|
-
/** The provisioned plane's own desktop, for the run's cost estimate. */
|
|
138
|
-
export interface SubjectDesktopUsage {
|
|
139
|
-
durationMs: number | undefined;
|
|
140
|
-
observation: DesktopResourceObservation | undefined;
|
|
141
|
-
killed: boolean;
|
|
142
|
-
}
|
|
143
137
|
/** One actor's measured run (internal). */
|
|
144
138
|
export interface ActorRunResult {
|
|
145
139
|
spec: DesktopParticipantRun;
|
|
@@ -163,14 +157,11 @@ export interface PlaneContext {
|
|
|
163
157
|
deps: StudyDeps;
|
|
164
158
|
env: Record<string, string | undefined>;
|
|
165
159
|
concurrency: number;
|
|
166
|
-
runBudget: CuaRunBudget | undefined;
|
|
167
160
|
runSession: CuaParticipantDeps["runSession"];
|
|
168
161
|
openaiApiKey: string;
|
|
169
162
|
e2bApiKey: string;
|
|
163
|
+
/** `run.secrets.scrub`, which every plane step that writes text reads. */
|
|
170
164
|
scrubKnownValues: (text: string) => string;
|
|
171
|
-
/** The values scrubKnownValues removes. Real email receiving adds its secrets before participants
|
|
172
|
-
* start. */
|
|
173
|
-
knownSecretValues: readonly string[];
|
|
174
165
|
cwd: string;
|
|
175
166
|
run: StartedRun;
|
|
176
167
|
runId: string;
|