pi-oracle 0.2.1 → 0.3.0
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/CHANGELOG.md +22 -0
- package/README.md +1 -0
- package/docs/ORACLE_DESIGN.md +5 -7
- package/extensions/oracle/lib/config.ts +55 -19
- package/extensions/oracle/lib/jobs.ts +4 -14
- package/extensions/oracle/lib/tools.ts +18 -51
- package/extensions/oracle/worker/artifact-heuristics.d.mts +19 -3
- package/extensions/oracle/worker/artifact-heuristics.mjs +89 -18
- package/extensions/oracle/worker/run-job.mjs +134 -59
- package/package.json +1 -1
- package/prompts/oracle.md +7 -2
package/CHANGELOG.md
CHANGED
|
@@ -1,5 +1,27 @@
|
|
|
1
1
|
# Changelog
|
|
2
2
|
|
|
3
|
+
## 0.3.0 - 2026-04-08
|
|
4
|
+
|
|
5
|
+
### Changed
|
|
6
|
+
- breaking: `oracle_submit` and oracle config defaults now use preset-only model selection; legacy `modelFamily` / `effort` / `autoSwitchToThinking` submit inputs and default config fields were removed in favor of canonical preset ids
|
|
7
|
+
- oracle jobs now persist a resolved `selection` snapshot and the worker configures ChatGPT from that persisted selection instead of re-deriving model settings from legacy job fields
|
|
8
|
+
- oracle model preset definitions now come from a single canonical registry in `extensions/oracle/lib/config.ts`
|
|
9
|
+
|
|
10
|
+
### Fixed
|
|
11
|
+
- removed duplicate hand-maintained preset-id examples from agent-facing prompt and design docs so callers are directed to the tool schema / canonical registry instead of stale inline lists
|
|
12
|
+
- oracle sanity coverage now validates the preset-only contract from the registered tool schema and canonical registry instead of brittle prose-only assertions
|
|
13
|
+
- worker model configuration now consistently uses the explicit `configureModel(job)` parameter instead of hidden coupling through the module-global current job
|
|
14
|
+
|
|
15
|
+
## 0.2.2 - 2026-04-07
|
|
16
|
+
|
|
17
|
+
### Fixed
|
|
18
|
+
- missed ChatGPT file artifacts now map generic download controls onto nearby filenames and download from live DOM selectors instead of relying only on filename-labeled snapshot refs
|
|
19
|
+
- oracle jobs no longer report a false-clean completion when response-local artifact signals are present but capture fails or remains inconclusive
|
|
20
|
+
- artifact label extraction now collapses paths and mixed response text down to real filenames so suspicious artifact fallback logic does not emit bogus labels
|
|
21
|
+
|
|
22
|
+
### Added
|
|
23
|
+
- regression coverage for artifact label extraction edge cases and ambiguous download-control artifact detection
|
|
24
|
+
|
|
3
25
|
## 0.2.1 - 2026-04-07
|
|
4
26
|
|
|
5
27
|
### Fixed
|
package/README.md
CHANGED
package/docs/ORACLE_DESIGN.md
CHANGED
|
@@ -133,7 +133,9 @@ The authenticated seed profile remains the source of truth for future oracle run
|
|
|
133
133
|
|
|
134
134
|
### `oracle_submit`
|
|
135
135
|
|
|
136
|
-
|
|
136
|
+
Agent-facing submissions use **`preset`**; the canonical registry is `ORACLE_SUBMIT_PRESETS` in `extensions/oracle/lib/config.ts`. **`preset` is the only model-selection parameter** on `oracle_submit`. There are no `modelFamily`, `effort`, or `autoSwitchToThinking` fields.
|
|
137
|
+
|
|
138
|
+
1. resolve the preset (submit-time or config default) into an execution snapshot
|
|
137
139
|
2. resolve optional `followUpJobId` into a prior `chatUrl` and `conversationId`
|
|
138
140
|
3. build the archive first into a temporary path
|
|
139
141
|
4. allocate a unique runtime:
|
|
@@ -216,9 +218,7 @@ Browser/auth settings are global-only because they control local privileged brow
|
|
|
216
218
|
```json
|
|
217
219
|
{
|
|
218
220
|
"defaults": {
|
|
219
|
-
"
|
|
220
|
-
"effort": "extended",
|
|
221
|
-
"autoSwitchToThinking": false
|
|
221
|
+
"preset": "<preset id from ORACLE_SUBMIT_PRESETS>"
|
|
222
222
|
},
|
|
223
223
|
"browser": {
|
|
224
224
|
"sessionPrefix": "oracle",
|
|
@@ -317,9 +317,7 @@ Important fields include:
|
|
|
317
317
|
- `sessionId`
|
|
318
318
|
- `originSessionFile`
|
|
319
319
|
- `requestSource`
|
|
320
|
-
- `
|
|
321
|
-
- `effort`
|
|
322
|
-
- `autoSwitchToThinking`
|
|
320
|
+
- `selection`: resolved execution snapshot with `{ preset, modelFamily, effort?, autoSwitchToThinking }`
|
|
323
321
|
- `followUpToJobId`
|
|
324
322
|
- `chatUrl`
|
|
325
323
|
- `conversationId`
|
|
@@ -10,13 +10,61 @@ export type OracleModelFamily = (typeof MODEL_FAMILIES)[number];
|
|
|
10
10
|
export const EFFORTS = ["light", "standard", "extended", "heavy"] as const;
|
|
11
11
|
export type OracleEffort = (typeof EFFORTS)[number];
|
|
12
12
|
|
|
13
|
+
/**
|
|
14
|
+
* Canonical preset registry for `oracle_submit` preset selection.
|
|
15
|
+
* This is the single authored source of truth — all derived lists come from `Object.keys(...)`.
|
|
16
|
+
*/
|
|
17
|
+
export const ORACLE_SUBMIT_PRESETS = {
|
|
18
|
+
pro_standard: { label: "Pro - Standard", modelFamily: "pro" as const, effort: "standard" as const, autoSwitchToThinking: false },
|
|
19
|
+
pro_extended: { label: "Pro - Extended", modelFamily: "pro" as const, effort: "extended" as const, autoSwitchToThinking: false },
|
|
20
|
+
thinking_light: { label: "Thinking - Light", modelFamily: "thinking" as const, effort: "light" as const, autoSwitchToThinking: false },
|
|
21
|
+
thinking_standard: { label: "Thinking - Standard", modelFamily: "thinking" as const, effort: "standard" as const, autoSwitchToThinking: false },
|
|
22
|
+
thinking_extended: { label: "Thinking - Extended", modelFamily: "thinking" as const, effort: "extended" as const, autoSwitchToThinking: false },
|
|
23
|
+
thinking_heavy: { label: "Thinking - Heavy", modelFamily: "thinking" as const, effort: "heavy" as const, autoSwitchToThinking: false },
|
|
24
|
+
instant: { label: "Instant", modelFamily: "instant" as const, autoSwitchToThinking: false },
|
|
25
|
+
instant_auto_switch: { label: "Instant - Auto-switch to Thinking Enabled", modelFamily: "instant" as const, autoSwitchToThinking: true },
|
|
26
|
+
} as const;
|
|
27
|
+
|
|
28
|
+
export type OracleSubmitPresetId = keyof typeof ORACLE_SUBMIT_PRESETS;
|
|
29
|
+
|
|
30
|
+
export type OracleSubmitPreset = typeof ORACLE_SUBMIT_PRESETS[OracleSubmitPresetId];
|
|
31
|
+
|
|
32
|
+
export function getOracleSubmitPresetById(id: OracleSubmitPresetId): OracleSubmitPreset {
|
|
33
|
+
const found = ORACLE_SUBMIT_PRESETS[id];
|
|
34
|
+
if (!found) {
|
|
35
|
+
throw new Error(`Unknown oracle_submit preset: ${id}`);
|
|
36
|
+
}
|
|
37
|
+
return found;
|
|
38
|
+
}
|
|
39
|
+
|
|
40
|
+
/** Resolved execution snapshot generated from a preset at submit time. */
|
|
41
|
+
export type OracleResolvedSelection = {
|
|
42
|
+
preset: OracleSubmitPresetId;
|
|
43
|
+
modelFamily: OracleModelFamily;
|
|
44
|
+
effort?: OracleEffort;
|
|
45
|
+
autoSwitchToThinking: boolean;
|
|
46
|
+
};
|
|
47
|
+
|
|
48
|
+
/**
|
|
49
|
+
* Resolve a preset id into the execution snapshot that gets persisted on the job.
|
|
50
|
+
* @throws if the preset id is unknown.
|
|
51
|
+
*/
|
|
52
|
+
export function resolveOracleSubmitPreset(presetId: OracleSubmitPresetId): OracleResolvedSelection {
|
|
53
|
+
const def = getOracleSubmitPresetById(presetId);
|
|
54
|
+
return {
|
|
55
|
+
preset: presetId,
|
|
56
|
+
modelFamily: def.modelFamily,
|
|
57
|
+
effort: def.modelFamily === "instant" ? undefined : def.effort,
|
|
58
|
+
autoSwitchToThinking: def.modelFamily === "instant" ? def.autoSwitchToThinking : false,
|
|
59
|
+
};
|
|
60
|
+
}
|
|
61
|
+
|
|
13
62
|
export const BROWSER_RUN_MODES = ["headless", "headed"] as const;
|
|
14
63
|
export type OracleBrowserRunMode = (typeof BROWSER_RUN_MODES)[number];
|
|
15
64
|
|
|
16
65
|
export const CLONE_STRATEGIES = ["apfs-clone", "copy"] as const;
|
|
17
66
|
export type OracleCloneStrategy = (typeof CLONE_STRATEGIES)[number];
|
|
18
67
|
|
|
19
|
-
const PRO_EFFORTS = ["standard", "extended"] as const satisfies readonly OracleEffort[];
|
|
20
68
|
const ALLOWED_CHATGPT_ORIGINS = new Set(["https://chatgpt.com", "https://chat.openai.com"]);
|
|
21
69
|
const PROJECT_OVERRIDE_KEYS = new Set(["defaults", "worker", "poller", "artifacts", "cleanup"]);
|
|
22
70
|
const DEFAULT_MAC_CHROME_EXECUTABLE = "/Applications/Google Chrome.app/Contents/MacOS/Google Chrome";
|
|
@@ -24,9 +72,7 @@ const DEFAULT_MAC_CHROME_USER_DATA_DIR = join(homedir(), "Library", "Application
|
|
|
24
72
|
|
|
25
73
|
export interface OracleConfig {
|
|
26
74
|
defaults: {
|
|
27
|
-
|
|
28
|
-
effort: OracleEffort;
|
|
29
|
-
autoSwitchToThinking: boolean;
|
|
75
|
+
preset: OracleSubmitPresetId;
|
|
30
76
|
};
|
|
31
77
|
browser: {
|
|
32
78
|
sessionPrefix: string;
|
|
@@ -98,9 +144,7 @@ const detectedChromeProfileName = detectDefaultChromeProfileName();
|
|
|
98
144
|
|
|
99
145
|
export const DEFAULT_CONFIG: OracleConfig = {
|
|
100
146
|
defaults: {
|
|
101
|
-
|
|
102
|
-
effort: "extended",
|
|
103
|
-
autoSwitchToThinking: false,
|
|
147
|
+
preset: "pro_extended",
|
|
104
148
|
},
|
|
105
149
|
browser: {
|
|
106
150
|
sessionPrefix: "oracle",
|
|
@@ -292,19 +336,13 @@ function normalizeLegacyBrowserConfig(root: Record<string, unknown>): Record<str
|
|
|
292
336
|
return root;
|
|
293
337
|
}
|
|
294
338
|
|
|
339
|
+
const PRESET_IDS = Object.keys(ORACLE_SUBMIT_PRESETS) as unknown as readonly OracleSubmitPresetId[];
|
|
340
|
+
|
|
295
341
|
function validateOracleConfig(value: unknown): OracleConfig {
|
|
296
342
|
const root = normalizeLegacyBrowserConfig(expectObject(value, "root"));
|
|
297
343
|
|
|
298
344
|
const defaults = expectObject(root.defaults, "defaults");
|
|
299
|
-
const
|
|
300
|
-
const effort = expectEnum(defaults.effort, "defaults.effort", EFFORTS);
|
|
301
|
-
const autoSwitchToThinking = expectBoolean(defaults.autoSwitchToThinking, "defaults.autoSwitchToThinking");
|
|
302
|
-
if (modelFamily === "pro" && effort !== "standard" && effort !== "extended") {
|
|
303
|
-
throw new Error(`Invalid oracle config: defaults.effort must be one of ${PRO_EFFORTS.join(", ")} for pro`);
|
|
304
|
-
}
|
|
305
|
-
if (modelFamily !== "instant" && autoSwitchToThinking) {
|
|
306
|
-
throw new Error("Invalid oracle config: defaults.autoSwitchToThinking is only valid for instant");
|
|
307
|
-
}
|
|
345
|
+
const preset = expectEnum(defaults.preset, "defaults.preset", PRESET_IDS);
|
|
308
346
|
|
|
309
347
|
const browser = expectObject(root.browser, "browser");
|
|
310
348
|
const auth = expectObject(root.auth, "auth");
|
|
@@ -321,9 +359,7 @@ function validateOracleConfig(value: unknown): OracleConfig {
|
|
|
321
359
|
|
|
322
360
|
return {
|
|
323
361
|
defaults: {
|
|
324
|
-
|
|
325
|
-
effort,
|
|
326
|
-
autoSwitchToThinking,
|
|
362
|
+
preset,
|
|
327
363
|
},
|
|
328
364
|
browser: {
|
|
329
365
|
sessionPrefix: expectString(browser.sessionPrefix, "browser.sessionPrefix"),
|
|
@@ -4,7 +4,7 @@ import { existsSync, readdirSync, readFileSync } from "node:fs";
|
|
|
4
4
|
import { chmod, mkdir, readFile, rename, rm, writeFile } from "node:fs/promises";
|
|
5
5
|
import { join, resolve } from "node:path";
|
|
6
6
|
import type { ExtensionContext } from "@mariozechner/pi-coding-agent";
|
|
7
|
-
import type { OracleConfig,
|
|
7
|
+
import type { OracleConfig, OracleResolvedSelection } from "./config.js";
|
|
8
8
|
import { withJobLock, withLock } from "./locks.js";
|
|
9
9
|
import { cleanupRuntimeArtifacts, getProjectId, getSessionId, parseConversationId, requirePersistedSessionFile, type OracleCleanupReport } from "./runtime.js";
|
|
10
10
|
|
|
@@ -133,9 +133,7 @@ export interface OracleJob {
|
|
|
133
133
|
sessionId: string;
|
|
134
134
|
originSessionFile?: string;
|
|
135
135
|
requestSource: "command" | "tool";
|
|
136
|
-
|
|
137
|
-
effort?: OracleEffort;
|
|
138
|
-
autoSwitchToThinking?: boolean;
|
|
136
|
+
selection: OracleResolvedSelection;
|
|
139
137
|
followUpToJobId?: string;
|
|
140
138
|
chatUrl?: string;
|
|
141
139
|
conversationId?: string;
|
|
@@ -185,9 +183,7 @@ export interface OracleJob {
|
|
|
185
183
|
export interface OracleSubmitInput {
|
|
186
184
|
prompt: string;
|
|
187
185
|
files: string[];
|
|
188
|
-
|
|
189
|
-
effort?: OracleEffort;
|
|
190
|
-
autoSwitchToThinking?: boolean;
|
|
186
|
+
selection: OracleResolvedSelection;
|
|
191
187
|
followUpToJobId?: string;
|
|
192
188
|
chatUrl?: string;
|
|
193
189
|
requestSource: "command" | "tool";
|
|
@@ -971,10 +967,6 @@ export async function createJob(
|
|
|
971
967
|
|
|
972
968
|
const createdAt = options?.createdAt ?? new Date().toISOString();
|
|
973
969
|
const initialState = options?.initialState ?? "submitted";
|
|
974
|
-
const normalizedEffort = input.modelFamily === "instant" ? undefined : (input.effort ?? config.defaults.effort);
|
|
975
|
-
const normalizedAutoSwitchToThinking = input.modelFamily === "instant"
|
|
976
|
-
? (input.autoSwitchToThinking ?? config.defaults.autoSwitchToThinking)
|
|
977
|
-
: false;
|
|
978
970
|
const job: OracleJob = {
|
|
979
971
|
id,
|
|
980
972
|
status: initialState,
|
|
@@ -988,9 +980,7 @@ export async function createJob(
|
|
|
988
980
|
sessionId,
|
|
989
981
|
originSessionFile: sessionFile,
|
|
990
982
|
requestSource: input.requestSource,
|
|
991
|
-
|
|
992
|
-
effort: normalizedEffort,
|
|
993
|
-
autoSwitchToThinking: normalizedAutoSwitchToThinking,
|
|
983
|
+
selection: input.selection,
|
|
994
984
|
followUpToJobId: input.followUpToJobId,
|
|
995
985
|
chatUrl: input.followUpToJobId ? input.chatUrl : undefined,
|
|
996
986
|
conversationId,
|
|
@@ -5,7 +5,12 @@ import { basename, join, posix } from "node:path";
|
|
|
5
5
|
import type { ExtensionAPI } from "@mariozechner/pi-coding-agent";
|
|
6
6
|
import { Type } from "@sinclair/typebox";
|
|
7
7
|
import { isLockTimeoutError, withGlobalReconcileLock, withLock } from "./locks.js";
|
|
8
|
-
import {
|
|
8
|
+
import {
|
|
9
|
+
loadOracleConfig,
|
|
10
|
+
ORACLE_SUBMIT_PRESETS,
|
|
11
|
+
resolveOracleSubmitPreset,
|
|
12
|
+
type OracleSubmitPresetId,
|
|
13
|
+
} from "./config.js";
|
|
9
14
|
import {
|
|
10
15
|
appendCleanupWarnings,
|
|
11
16
|
cancelOracleJob,
|
|
@@ -53,10 +58,11 @@ const ORACLE_SUBMIT_PARAMS = Type.Object({
|
|
|
53
58
|
description: "Exact project-relative files/directories to include in the oracle archive.",
|
|
54
59
|
minItems: 1,
|
|
55
60
|
}),
|
|
56
|
-
|
|
57
|
-
|
|
58
|
-
|
|
59
|
-
|
|
61
|
+
preset: Type.Optional(
|
|
62
|
+
stringEnum(
|
|
63
|
+
[...Object.keys(ORACLE_SUBMIT_PRESETS)] as const,
|
|
64
|
+
"ChatGPT model preset. Omit to use the configured default preset.",
|
|
65
|
+
),
|
|
60
66
|
),
|
|
61
67
|
followUpJobId: Type.Optional(Type.String({ description: "Earlier oracle job id whose chat thread should be continued." })),
|
|
62
68
|
});
|
|
@@ -69,12 +75,6 @@ const ORACLE_CANCEL_PARAMS = Type.Object({
|
|
|
69
75
|
jobId: Type.String({ description: "Oracle job id." }),
|
|
70
76
|
});
|
|
71
77
|
|
|
72
|
-
const VALID_EFFORTS: Record<OracleModelFamily, readonly OracleEffort[]> = {
|
|
73
|
-
instant: [],
|
|
74
|
-
thinking: ["light", "standard", "extended", "heavy"],
|
|
75
|
-
pro: ["standard", "extended"],
|
|
76
|
-
};
|
|
77
|
-
|
|
78
78
|
const MAX_ARCHIVE_BYTES = 250 * 1024 * 1024;
|
|
79
79
|
const MAX_QUEUED_JOBS_PER_ACTIVE_RUNTIME = 1;
|
|
80
80
|
const MAX_QUEUED_ARCHIVE_BYTES_PER_ACTIVE_RUNTIME = MAX_ARCHIVE_BYTES;
|
|
@@ -491,29 +491,6 @@ export function getQueueAdmissionFailure(args: {
|
|
|
491
491
|
return undefined;
|
|
492
492
|
}
|
|
493
493
|
|
|
494
|
-
function validateSubmissionOptions(
|
|
495
|
-
params: { effort?: OracleEffort; autoSwitchToThinking?: boolean },
|
|
496
|
-
modelFamily: OracleModelFamily,
|
|
497
|
-
effort: OracleEffort | undefined,
|
|
498
|
-
autoSwitchToThinking: boolean,
|
|
499
|
-
): void {
|
|
500
|
-
if (modelFamily === "instant" && params.effort !== undefined) {
|
|
501
|
-
throw new Error("Instant model family does not support effort selection");
|
|
502
|
-
}
|
|
503
|
-
|
|
504
|
-
if (effort && !VALID_EFFORTS[modelFamily].includes(effort)) {
|
|
505
|
-
throw new Error(`Invalid effort for ${modelFamily}: ${effort}`);
|
|
506
|
-
}
|
|
507
|
-
|
|
508
|
-
if (modelFamily !== "instant" && params.autoSwitchToThinking === true) {
|
|
509
|
-
throw new Error("autoSwitchToThinking is only valid for the instant model family");
|
|
510
|
-
}
|
|
511
|
-
|
|
512
|
-
if (modelFamily !== "instant" && autoSwitchToThinking) {
|
|
513
|
-
throw new Error(`autoSwitchToThinking cannot be enabled for ${modelFamily}`);
|
|
514
|
-
}
|
|
515
|
-
}
|
|
516
|
-
|
|
517
494
|
function resolveFollowUp(previousJobId: string | undefined, cwd: string): {
|
|
518
495
|
followUpToJobId?: string;
|
|
519
496
|
chatUrl?: string;
|
|
@@ -601,7 +578,8 @@ export function registerOracleTools(pi: ExtensionAPI, workerPath: string): void
|
|
|
601
578
|
name: "oracle_submit",
|
|
602
579
|
label: "Oracle Submit",
|
|
603
580
|
description:
|
|
604
|
-
"Dispatch a background ChatGPT web oracle job after gathering context. Always pass a prompt and exact project-relative archive inputs."
|
|
581
|
+
"Dispatch a background ChatGPT web oracle job after gathering context. Always pass a prompt and exact project-relative archive inputs. " +
|
|
582
|
+
"Optional ChatGPT model: set parameter `preset`, or omit it for configured defaults (see `preset` field for allowed ids).",
|
|
605
583
|
promptSnippet: "Dispatch a background ChatGPT web oracle job after gathering repo context.",
|
|
606
584
|
promptGuidelines: [
|
|
607
585
|
"Gather context before calling oracle_submit.",
|
|
@@ -613,7 +591,7 @@ export function registerOracleTools(pi: ExtensionAPI, workerPath: string): void
|
|
|
613
591
|
"If oracle_submit itself fails because the local archive still exceeds the upload limit after default exclusions and automatic generic generated-output-dir pruning, or for any other submit-time error, stop and report the error instead of retrying automatically.",
|
|
614
592
|
"If oracle_submit returns a queued job instead of an immediately dispatched one, treat that as success and stop exactly the same way.",
|
|
615
593
|
"Stop after dispatching oracle_submit; do not continue the task while the oracle job is running.",
|
|
616
|
-
"
|
|
594
|
+
"Use `preset` as the only model-selection parameter on oracle_submit. Allowed values come from the tool schema enum. Omit preset to use the configured default.",
|
|
617
595
|
],
|
|
618
596
|
parameters: ORACLE_SUBMIT_PARAMS,
|
|
619
597
|
async execute(_toolCallId, params, _signal, _onUpdate, ctx) {
|
|
@@ -621,16 +599,9 @@ export function registerOracleTools(pi: ExtensionAPI, workerPath: string): void
|
|
|
621
599
|
const originSessionFile = requirePersistedSessionFile(getSessionFile(ctx), "submit oracle jobs");
|
|
622
600
|
const projectId = getProjectId(ctx.cwd);
|
|
623
601
|
const sessionId = getSessionId(originSessionFile, projectId);
|
|
624
|
-
const
|
|
625
|
-
const
|
|
626
|
-
const modelFamily: OracleModelFamily = submittedModelFamily ?? config.defaults.modelFamily;
|
|
627
|
-
const requestedEffort: OracleEffort = submittedEffort ?? config.defaults.effort;
|
|
628
|
-
const effort: OracleEffort | undefined = modelFamily === "instant" ? undefined : requestedEffort;
|
|
629
|
-
const rawAutoSwitchToThinking = params.autoSwitchToThinking ?? config.defaults.autoSwitchToThinking;
|
|
630
|
-
const autoSwitchToThinking = modelFamily === "instant" ? rawAutoSwitchToThinking : false;
|
|
602
|
+
const presetId = (params.preset as OracleSubmitPresetId | undefined) ?? config.defaults.preset;
|
|
603
|
+
const selection = resolveOracleSubmitPreset(presetId);
|
|
631
604
|
const followUp = resolveFollowUp(params.followUpJobId, ctx.cwd);
|
|
632
|
-
|
|
633
|
-
validateSubmissionOptions({ effort: submittedEffort, autoSwitchToThinking: params.autoSwitchToThinking }, modelFamily, effort, autoSwitchToThinking);
|
|
634
605
|
try {
|
|
635
606
|
await withGlobalReconcileLock({ processPid: process.pid, source: "oracle_submit", cwd: ctx.cwd }, async () => {
|
|
636
607
|
await reconcileStaleOracleJobs();
|
|
@@ -691,9 +662,7 @@ export function registerOracleTools(pi: ExtensionAPI, workerPath: string): void
|
|
|
691
662
|
{
|
|
692
663
|
prompt: params.prompt,
|
|
693
664
|
files: params.files,
|
|
694
|
-
|
|
695
|
-
effort,
|
|
696
|
-
autoSwitchToThinking,
|
|
665
|
+
selection,
|
|
697
666
|
followUpToJobId: followUp.followUpToJobId,
|
|
698
667
|
chatUrl: followUp.chatUrl,
|
|
699
668
|
requestSource: "tool",
|
|
@@ -736,9 +705,7 @@ export function registerOracleTools(pi: ExtensionAPI, workerPath: string): void
|
|
|
736
705
|
{
|
|
737
706
|
prompt: params.prompt,
|
|
738
707
|
files: params.files,
|
|
739
|
-
|
|
740
|
-
effort,
|
|
741
|
-
autoSwitchToThinking,
|
|
708
|
+
selection,
|
|
742
709
|
followUpToJobId: followUp.followUpToJobId,
|
|
743
710
|
chatUrl: followUp.chatUrl,
|
|
744
711
|
requestSource: "tool",
|
|
@@ -10,20 +10,36 @@ export interface SnapshotEntry {
|
|
|
10
10
|
|
|
11
11
|
export interface StructuralArtifactCandidateInput {
|
|
12
12
|
label?: string;
|
|
13
|
+
selector?: string;
|
|
14
|
+
controlLabel?: string;
|
|
13
15
|
paragraphText?: string;
|
|
14
16
|
listItemText?: string;
|
|
15
|
-
|
|
17
|
+
paragraphInteractiveCount?: number;
|
|
18
|
+
paragraphArtifactLabelCount?: number;
|
|
16
19
|
paragraphOtherTextLength?: number;
|
|
17
|
-
|
|
18
|
-
|
|
20
|
+
listItemInteractiveCount?: number;
|
|
21
|
+
listItemArtifactLabelCount?: number;
|
|
22
|
+
focusableInteractiveCount?: number;
|
|
23
|
+
focusableArtifactLabelCount?: number;
|
|
19
24
|
focusableOtherTextLength?: number;
|
|
20
25
|
}
|
|
21
26
|
|
|
22
27
|
export interface StructuralArtifactCandidate {
|
|
23
28
|
label: string;
|
|
29
|
+
selector?: string;
|
|
30
|
+
controlLabel?: string;
|
|
31
|
+
}
|
|
32
|
+
|
|
33
|
+
export interface StructuralArtifactCandidatePartition {
|
|
34
|
+
confirmed: StructuralArtifactCandidate[];
|
|
35
|
+
suspicious: StructuralArtifactCandidate[];
|
|
24
36
|
}
|
|
25
37
|
|
|
26
38
|
export function parseSnapshotEntries(snapshot: string): SnapshotEntry[];
|
|
39
|
+
export function extractArtifactLabels(value: string): string[];
|
|
27
40
|
export function filterStructuralArtifactCandidates(
|
|
28
41
|
candidates: StructuralArtifactCandidateInput[],
|
|
29
42
|
): StructuralArtifactCandidate[];
|
|
43
|
+
export function partitionStructuralArtifactCandidates(
|
|
44
|
+
candidates: StructuralArtifactCandidateInput[],
|
|
45
|
+
): StructuralArtifactCandidatePartition;
|
|
@@ -1,7 +1,8 @@
|
|
|
1
|
-
export const FILE_LABEL_PATTERN_SOURCE = String.raw`(?:^|[
|
|
2
|
-
const FILE_LABEL_PATTERN = new RegExp(FILE_LABEL_PATTERN_SOURCE);
|
|
1
|
+
export const FILE_LABEL_PATTERN_SOURCE = String.raw`(?:^|[^A-Za-z0-9._~/-])((?:(?:[A-Za-z]:)?[\\/]|[.~][\\/])?(?:[^\\/\s"'<>|]+[\\/])*[^\\/\s"'<>|]+\.[A-Za-z0-9]{1,12})(?=$|[^A-Za-z0-9._~/-])`;
|
|
2
|
+
const FILE_LABEL_PATTERN = new RegExp(FILE_LABEL_PATTERN_SOURCE, "g");
|
|
3
3
|
export const GENERIC_ARTIFACT_LABELS = ["ATTACHED", "DONE"];
|
|
4
4
|
const GENERIC_ARTIFACT_LABEL_SET = new Set(GENERIC_ARTIFACT_LABELS);
|
|
5
|
+
const GENERIC_DOWNLOAD_CONTROL_PATTERN = /(?:^|\b)(?:download|save)(?:\b|$)/i;
|
|
5
6
|
|
|
6
7
|
export function parseSnapshotEntries(snapshot) {
|
|
7
8
|
return String(snapshot || "")
|
|
@@ -29,11 +30,61 @@ function normalizeText(value) {
|
|
|
29
30
|
return String(value || "").replace(/\s+/g, " ").trim();
|
|
30
31
|
}
|
|
31
32
|
|
|
33
|
+
function sanitizeArtifactLabel(value) {
|
|
34
|
+
const normalized = normalizeText(value).replace(/^[^A-Za-z0-9._~/-]+|[^A-Za-z0-9._~/-]+$/g, "");
|
|
35
|
+
if (!normalized) return "";
|
|
36
|
+
const basename = normalized.split(/[\\/]/).filter(Boolean).at(-1) || "";
|
|
37
|
+
return basename.replace(/^[^A-Za-z0-9._-]+|[^A-Za-z0-9._-]+$/g, "");
|
|
38
|
+
}
|
|
39
|
+
|
|
40
|
+
export function extractArtifactLabels(value) {
|
|
41
|
+
const seen = new Set();
|
|
42
|
+
const labels = [];
|
|
43
|
+
for (const match of String(value || "").matchAll(FILE_LABEL_PATTERN)) {
|
|
44
|
+
const normalized = sanitizeArtifactLabel(match[1] || match[0] || "");
|
|
45
|
+
if (!normalized || seen.has(normalized)) continue;
|
|
46
|
+
seen.add(normalized);
|
|
47
|
+
labels.push(normalized);
|
|
48
|
+
}
|
|
49
|
+
return labels;
|
|
50
|
+
}
|
|
51
|
+
|
|
32
52
|
export function isLikelyArtifactLabel(label) {
|
|
33
53
|
const normalized = normalizeText(label);
|
|
34
54
|
if (!normalized) return false;
|
|
35
55
|
if (GENERIC_ARTIFACT_LABEL_SET.has(normalized.toUpperCase())) return true;
|
|
36
|
-
return
|
|
56
|
+
return extractArtifactLabels(normalized).length > 0;
|
|
57
|
+
}
|
|
58
|
+
|
|
59
|
+
function hasGenericDownloadControl(controlLabel) {
|
|
60
|
+
return GENERIC_DOWNLOAD_CONTROL_PATTERN.test(normalizeText(controlLabel));
|
|
61
|
+
}
|
|
62
|
+
|
|
63
|
+
function normalizeCandidate(candidate) {
|
|
64
|
+
const label = normalizeText(candidate?.label);
|
|
65
|
+
return label ? { ...candidate, label } : undefined;
|
|
66
|
+
}
|
|
67
|
+
|
|
68
|
+
function hasArtifactSignal(candidate) {
|
|
69
|
+
const label = normalizeText(candidate?.label);
|
|
70
|
+
if (!isLikelyArtifactLabel(label)) return false;
|
|
71
|
+
|
|
72
|
+
const paragraphInteractiveCount = Number(candidate?.paragraphInteractiveCount || 0);
|
|
73
|
+
const listItemInteractiveCount = Number(candidate?.listItemInteractiveCount || 0);
|
|
74
|
+
const focusableInteractiveCount = Number(candidate?.focusableInteractiveCount || 0);
|
|
75
|
+
const paragraphArtifactLabelCount = Number(candidate?.paragraphArtifactLabelCount || 0);
|
|
76
|
+
const listItemArtifactLabelCount = Number(candidate?.listItemArtifactLabelCount || 0);
|
|
77
|
+
const focusableArtifactLabelCount = Number(candidate?.focusableArtifactLabelCount || 0);
|
|
78
|
+
|
|
79
|
+
return (
|
|
80
|
+
hasGenericDownloadControl(candidate?.controlLabel) ||
|
|
81
|
+
paragraphInteractiveCount > 0 ||
|
|
82
|
+
listItemInteractiveCount > 0 ||
|
|
83
|
+
focusableInteractiveCount > 0 ||
|
|
84
|
+
paragraphArtifactLabelCount > 0 ||
|
|
85
|
+
listItemArtifactLabelCount > 0 ||
|
|
86
|
+
focusableArtifactLabelCount > 0
|
|
87
|
+
);
|
|
37
88
|
}
|
|
38
89
|
|
|
39
90
|
export function isStructuralArtifactCandidate(candidate) {
|
|
@@ -41,36 +92,56 @@ export function isStructuralArtifactCandidate(candidate) {
|
|
|
41
92
|
if (!isLikelyArtifactLabel(label)) return false;
|
|
42
93
|
|
|
43
94
|
const listItemText = normalizeText(candidate?.listItemText);
|
|
44
|
-
const
|
|
45
|
-
const
|
|
95
|
+
const listItemInteractiveCount = Number(candidate?.listItemInteractiveCount || 0);
|
|
96
|
+
const listItemArtifactLabelCount = Number(candidate?.listItemArtifactLabelCount || 0);
|
|
97
|
+
const paragraphInteractiveCount = Number(candidate?.paragraphInteractiveCount || 0);
|
|
98
|
+
const paragraphArtifactLabelCount = Number(candidate?.paragraphArtifactLabelCount || 0);
|
|
46
99
|
const paragraphOtherTextLength = Number(candidate?.paragraphOtherTextLength ?? Number.POSITIVE_INFINITY);
|
|
47
|
-
const
|
|
100
|
+
const focusableInteractiveCount = Number(candidate?.focusableInteractiveCount || 0);
|
|
101
|
+
const focusableArtifactLabelCount = Number(candidate?.focusableArtifactLabelCount || 0);
|
|
48
102
|
const focusableOtherTextLength = Number(candidate?.focusableOtherTextLength ?? Number.POSITIVE_INFINITY);
|
|
49
103
|
|
|
50
|
-
if (listItemText === label &&
|
|
104
|
+
if (listItemText === label && listItemInteractiveCount === 1 && listItemArtifactLabelCount === 1) {
|
|
51
105
|
return true;
|
|
52
106
|
}
|
|
53
107
|
|
|
54
|
-
if (
|
|
108
|
+
if (paragraphArtifactLabelCount === 1 && paragraphInteractiveCount === 1 && paragraphOtherTextLength <= 32) {
|
|
55
109
|
return true;
|
|
56
110
|
}
|
|
57
111
|
|
|
58
|
-
if (
|
|
112
|
+
if (focusableArtifactLabelCount >= 1 && focusableInteractiveCount >= 1 && focusableOtherTextLength <= 64) {
|
|
59
113
|
return true;
|
|
60
114
|
}
|
|
61
115
|
|
|
62
116
|
return false;
|
|
63
117
|
}
|
|
64
118
|
|
|
65
|
-
export function
|
|
66
|
-
const
|
|
67
|
-
const
|
|
119
|
+
export function partitionStructuralArtifactCandidates(candidates) {
|
|
120
|
+
const confirmedSeen = new Set();
|
|
121
|
+
const suspiciousSeen = new Set();
|
|
122
|
+
const confirmed = [];
|
|
123
|
+
const suspicious = [];
|
|
124
|
+
|
|
68
125
|
for (const candidate of candidates || []) {
|
|
69
|
-
const
|
|
70
|
-
if (!
|
|
71
|
-
if (!
|
|
72
|
-
|
|
73
|
-
|
|
126
|
+
const normalized = normalizeCandidate(candidate);
|
|
127
|
+
if (!normalized) continue;
|
|
128
|
+
if (!hasArtifactSignal(normalized)) continue;
|
|
129
|
+
|
|
130
|
+
if (isStructuralArtifactCandidate(normalized)) {
|
|
131
|
+
if (confirmedSeen.has(normalized.label)) continue;
|
|
132
|
+
confirmedSeen.add(normalized.label);
|
|
133
|
+
confirmed.push(normalized);
|
|
134
|
+
continue;
|
|
135
|
+
}
|
|
136
|
+
|
|
137
|
+
if (suspiciousSeen.has(normalized.label)) continue;
|
|
138
|
+
suspiciousSeen.add(normalized.label);
|
|
139
|
+
suspicious.push(normalized);
|
|
74
140
|
}
|
|
75
|
-
|
|
141
|
+
|
|
142
|
+
return { confirmed, suspicious: suspicious.filter((candidate) => !confirmedSeen.has(candidate.label)) };
|
|
143
|
+
}
|
|
144
|
+
|
|
145
|
+
export function filterStructuralArtifactCandidates(candidates) {
|
|
146
|
+
return partitionStructuralArtifactCandidates(candidates).confirmed;
|
|
76
147
|
}
|
|
@@ -4,7 +4,7 @@ import { appendFile, chmod, mkdir, readFile, rename, rm, stat, writeFile } from
|
|
|
4
4
|
import { basename, dirname, join } from "node:path";
|
|
5
5
|
import { fileURLToPath } from "node:url";
|
|
6
6
|
import { spawn, execFileSync } from "node:child_process";
|
|
7
|
-
import {
|
|
7
|
+
import { extractArtifactLabels, FILE_LABEL_PATTERN_SOURCE, GENERIC_ARTIFACT_LABELS, parseSnapshotEntries, partitionStructuralArtifactCandidates } from "./artifact-heuristics.mjs";
|
|
8
8
|
import { createLease, listLeaseMetadata, readLeaseMetadata, releaseLease, withLock } from "./state-locks.mjs";
|
|
9
9
|
|
|
10
10
|
const jobId = process.argv[2];
|
|
@@ -807,7 +807,7 @@ function matchesModelFamilyButton(candidate, family) {
|
|
|
807
807
|
}
|
|
808
808
|
|
|
809
809
|
function requestedEffortLabel(job) {
|
|
810
|
-
return job.effort ? titleCase(job.effort) : undefined;
|
|
810
|
+
return job.selection?.effort ? titleCase(job.selection.effort) : undefined;
|
|
811
811
|
}
|
|
812
812
|
|
|
813
813
|
function effortSelectionVisible(snapshot, effortLabel) {
|
|
@@ -852,12 +852,12 @@ function snapshotHasModelConfigurationUi(snapshot) {
|
|
|
852
852
|
|
|
853
853
|
function snapshotStronglyMatchesRequestedModel(snapshot, job) {
|
|
854
854
|
const entries = parseSnapshotEntries(snapshot);
|
|
855
|
-
const familyMatched = entries.some((entry) => matchesModelFamilyButton(entry, job.
|
|
855
|
+
const familyMatched = entries.some((entry) => matchesModelFamilyButton(entry, job.selection.modelFamily));
|
|
856
856
|
const effortLabel = requestedEffortLabel(job);
|
|
857
|
-
if (job.
|
|
857
|
+
if (job.selection.modelFamily === "thinking") {
|
|
858
858
|
return familyMatched || effortSelectionVisible(snapshot, effortLabel);
|
|
859
859
|
}
|
|
860
|
-
if (job.
|
|
860
|
+
if (job.selection.modelFamily === "pro") {
|
|
861
861
|
return effortLabel ? familyMatched && effortSelectionVisible(snapshot, effortLabel) : familyMatched;
|
|
862
862
|
}
|
|
863
863
|
return familyMatched;
|
|
@@ -880,13 +880,13 @@ function composerControlsVisible(snapshot) {
|
|
|
880
880
|
}
|
|
881
881
|
|
|
882
882
|
function snapshotWeaklyMatchesRequestedModel(snapshot, job) {
|
|
883
|
-
if (job.
|
|
883
|
+
if (job.selection.modelFamily === "thinking") {
|
|
884
884
|
return effortSelectionVisible(snapshot, requestedEffortLabel(job)) || thinkingSelectionVisible(snapshot);
|
|
885
885
|
}
|
|
886
|
-
if (job.
|
|
886
|
+
if (job.selection.modelFamily === "pro") {
|
|
887
887
|
return !thinkingChipVisible(snapshot);
|
|
888
888
|
}
|
|
889
|
-
if (job.
|
|
889
|
+
if (job.selection.modelFamily === "instant") {
|
|
890
890
|
return !thinkingChipVisible(snapshot);
|
|
891
891
|
}
|
|
892
892
|
return false;
|
|
@@ -1214,7 +1214,7 @@ async function waitForModelConfigurationToSettle(job, options = {}) {
|
|
|
1214
1214
|
if (options.stronglyVerified) {
|
|
1215
1215
|
if (!fallbackLogged) {
|
|
1216
1216
|
fallbackLogged = true;
|
|
1217
|
-
await log(`Model configuration closed after strong in-dialog verification for family=${job.
|
|
1217
|
+
await log(`Model configuration closed after strong in-dialog verification for family=${job.selection.modelFamily} effort=${job.selection?.effort || "(none)"}`);
|
|
1218
1218
|
}
|
|
1219
1219
|
return;
|
|
1220
1220
|
}
|
|
@@ -1223,7 +1223,7 @@ async function waitForModelConfigurationToSettle(job, options = {}) {
|
|
|
1223
1223
|
if (!configurationUiVisible && composerControlsVisible(snapshot) && options.stronglyVerified) {
|
|
1224
1224
|
if (!fallbackLogged) {
|
|
1225
1225
|
fallbackLogged = true;
|
|
1226
|
-
await log(`Composer became usable after strong in-dialog verification for family=${job.
|
|
1226
|
+
await log(`Composer became usable after strong in-dialog verification for family=${job.selection.modelFamily} effort=${job.selection?.effort || "(none)"}`);
|
|
1227
1227
|
}
|
|
1228
1228
|
return;
|
|
1229
1229
|
}
|
|
@@ -1239,30 +1239,30 @@ async function waitForModelConfigurationToSettle(job, options = {}) {
|
|
|
1239
1239
|
}
|
|
1240
1240
|
|
|
1241
1241
|
if (options.stronglyVerified && lastSnapshot && !snapshotHasModelConfigurationUi(lastSnapshot)) {
|
|
1242
|
-
await log(`Model configuration closed only after settle-timeout for family=${job.
|
|
1242
|
+
await log(`Model configuration closed only after settle-timeout for family=${job.selection.modelFamily} effort=${job.selection?.effort || "(none)"}`);
|
|
1243
1243
|
return;
|
|
1244
1244
|
}
|
|
1245
1245
|
|
|
1246
|
-
throw new Error(`Could not verify requested model settings after configuration for ${job.
|
|
1246
|
+
throw new Error(`Could not verify requested model settings after configuration for ${job.selection.modelFamily}`);
|
|
1247
1247
|
}
|
|
1248
1248
|
|
|
1249
1249
|
async function configureModel(job) {
|
|
1250
1250
|
const initialSnapshot = await snapshotText(job);
|
|
1251
1251
|
if (snapshotStronglyMatchesRequestedModel(initialSnapshot, job)) {
|
|
1252
|
-
await log(`Model already appears configured for family=${job.
|
|
1252
|
+
await log(`Model already appears configured for family=${job.selection.modelFamily} effort=${job.selection?.effort || "(none)"}; skipping reconfiguration`);
|
|
1253
1253
|
return;
|
|
1254
1254
|
}
|
|
1255
1255
|
|
|
1256
|
-
await log(`Configuring model family=${job.
|
|
1256
|
+
await log(`Configuring model family=${job.selection.modelFamily} effort=${job.selection?.effort || "(none)"}`);
|
|
1257
1257
|
let familySnapshot = await openModelConfiguration(job);
|
|
1258
1258
|
let verificationSnapshot = familySnapshot;
|
|
1259
1259
|
|
|
1260
|
-
let familyEntry = findEntry(familySnapshot, (candidate) => matchesModelFamilyButton(candidate, job.
|
|
1260
|
+
let familyEntry = findEntry(familySnapshot, (candidate) => matchesModelFamilyButton(candidate, job.selection.modelFamily));
|
|
1261
1261
|
if (!familyEntry && snapshotStronglyMatchesRequestedModel(familySnapshot, job)) {
|
|
1262
1262
|
await log("Model configuration UI opened with requested settings already selected");
|
|
1263
1263
|
}
|
|
1264
1264
|
if (!familyEntry && !snapshotStronglyMatchesRequestedModel(familySnapshot, job)) {
|
|
1265
|
-
throw new Error(`Could not find model family button for ${job.
|
|
1265
|
+
throw new Error(`Could not find model family button for ${job.selection.modelFamily}`);
|
|
1266
1266
|
}
|
|
1267
1267
|
|
|
1268
1268
|
if (familyEntry) {
|
|
@@ -1272,7 +1272,7 @@ async function configureModel(job) {
|
|
|
1272
1272
|
verificationSnapshot = familySnapshot;
|
|
1273
1273
|
}
|
|
1274
1274
|
|
|
1275
|
-
if (job.
|
|
1275
|
+
if (job.selection.modelFamily === "thinking" || job.selection.modelFamily === "pro") {
|
|
1276
1276
|
const effortLabel = requestedEffortLabel(job);
|
|
1277
1277
|
if (effortLabel && !effortSelectionVisible(familySnapshot, effortLabel)) {
|
|
1278
1278
|
const opened = await openEffortDropdown(job);
|
|
@@ -1294,14 +1294,14 @@ async function configureModel(job) {
|
|
|
1294
1294
|
}
|
|
1295
1295
|
}
|
|
1296
1296
|
|
|
1297
|
-
if (job.
|
|
1297
|
+
if (job.selection.modelFamily === "instant" && job.selection.autoSwitchToThinking) {
|
|
1298
1298
|
await maybeClickLabeledEntry(job, CHATGPT_LABELS.autoSwitchToThinking);
|
|
1299
1299
|
verificationSnapshot = await snapshotText(job);
|
|
1300
1300
|
}
|
|
1301
1301
|
|
|
1302
1302
|
const stronglyVerified = snapshotStronglyMatchesRequestedModel(verificationSnapshot, job);
|
|
1303
1303
|
if (!stronglyVerified) {
|
|
1304
|
-
throw new Error(`Could not verify requested model settings in configuration UI for ${job.
|
|
1304
|
+
throw new Error(`Could not verify requested model settings in configuration UI for ${job.selection.modelFamily}`);
|
|
1305
1305
|
}
|
|
1306
1306
|
|
|
1307
1307
|
if (!(await maybeClickLabeledEntry(job, CHATGPT_LABELS.close, { kind: "button" }))) {
|
|
@@ -1497,28 +1497,52 @@ function preferredArtifactName(label, index) {
|
|
|
1497
1497
|
async function collectArtifactCandidates(job, responseIndex, responseText = "") {
|
|
1498
1498
|
const snapshot = await snapshotText(job);
|
|
1499
1499
|
const targetSlice = assistantSnapshotSlice(snapshot, responseIndex);
|
|
1500
|
-
if (!targetSlice) return { snapshot, targetSlice, candidates: [] };
|
|
1500
|
+
if (!targetSlice) return { snapshot, targetSlice, candidates: [], suspiciousLabels: [] };
|
|
1501
1501
|
|
|
1502
1502
|
const structural = await evalPage(
|
|
1503
1503
|
job,
|
|
1504
1504
|
toJsonScript(`
|
|
1505
1505
|
const normalize = (value) => String(value || '').replace(/\s+/g, ' ').trim();
|
|
1506
1506
|
const genericArtifactLabels = new Set(${JSON.stringify(GENERIC_ARTIFACT_LABELS)});
|
|
1507
|
-
const fileLabelPattern = new RegExp(${JSON.stringify(FILE_LABEL_PATTERN_SOURCE)});
|
|
1507
|
+
const fileLabelPattern = new RegExp(${JSON.stringify(FILE_LABEL_PATTERN_SOURCE)}, 'g');
|
|
1508
|
+
const downloadControlPattern = /(?:^|\\b)(?:download|save)(?:\\b|$)/i;
|
|
1509
|
+
const artifactMarkerAttr = 'data-pi-oracle-artifact-candidate';
|
|
1510
|
+
const artifactPrefix = 'pi-oracle-artifact-${jobId}-${responseIndex}-';
|
|
1511
|
+
const sanitize = (value) => normalize(value).replace(/^[^A-Za-z0-9._~/-]+|[^A-Za-z0-9._~/-]+$/g, '');
|
|
1512
|
+
const sanitizeArtifactLabel = (value) => {
|
|
1513
|
+
const normalized = sanitize(value);
|
|
1514
|
+
if (!normalized) return '';
|
|
1515
|
+
const basename = normalized.split(/[\\/]/).filter(Boolean).at(-1) || '';
|
|
1516
|
+
return basename.replace(/^[^A-Za-z0-9._-]+|[^A-Za-z0-9._-]+$/g, '');
|
|
1517
|
+
};
|
|
1518
|
+
const extractArtifactLabels = (value) => {
|
|
1519
|
+
const seen = new Set();
|
|
1520
|
+
const labels = [];
|
|
1521
|
+
for (const match of String(value || '').matchAll(fileLabelPattern)) {
|
|
1522
|
+
const label = sanitizeArtifactLabel(match[1] || match[0] || '');
|
|
1523
|
+
if (!label || seen.has(label)) continue;
|
|
1524
|
+
seen.add(label);
|
|
1525
|
+
labels.push(label);
|
|
1526
|
+
}
|
|
1527
|
+
return labels;
|
|
1528
|
+
};
|
|
1508
1529
|
const isFileLabel = (value) => {
|
|
1509
1530
|
const normalized = normalize(value);
|
|
1510
1531
|
if (!normalized) return false;
|
|
1511
1532
|
if (genericArtifactLabels.has(normalized.toUpperCase())) return true;
|
|
1512
|
-
return
|
|
1533
|
+
return extractArtifactLabels(normalized).length > 0;
|
|
1513
1534
|
};
|
|
1535
|
+
const isDownloadControl = (value) => downloadControlPattern.test(normalize(value));
|
|
1514
1536
|
const headings = Array.from(document.querySelectorAll('h1,h2,h3,h4,h5,h6,[role="heading"]'))
|
|
1515
1537
|
.filter((el) => normalize(el.textContent) === 'ChatGPT said:');
|
|
1516
1538
|
const host = headings[${responseIndex}]?.nextElementSibling;
|
|
1517
1539
|
if (!host) return { candidates: [] };
|
|
1518
1540
|
|
|
1519
|
-
const
|
|
1520
|
-
|
|
1521
|
-
|
|
1541
|
+
const interactiveElements = (node) => node ? Array.from(node.querySelectorAll('button, a')) : [];
|
|
1542
|
+
const interactiveLabels = (node) => interactiveElements(node)
|
|
1543
|
+
.map((candidate) => normalize(candidate.textContent || candidate.getAttribute('aria-label') || candidate.getAttribute('title')))
|
|
1544
|
+
.filter(Boolean);
|
|
1545
|
+
const artifactLabelsForNode = (node) => extractArtifactLabels(node?.textContent || '');
|
|
1522
1546
|
const otherTextLength = (text, labels) => {
|
|
1523
1547
|
let remaining = normalize(text);
|
|
1524
1548
|
for (const label of labels || []) {
|
|
@@ -1528,25 +1552,43 @@ async function collectArtifactCandidates(job, responseIndex, responseText = "")
|
|
|
1528
1552
|
return remaining.length;
|
|
1529
1553
|
};
|
|
1530
1554
|
const focusableFor = (node) => node?.closest('[tabindex]');
|
|
1555
|
+
const uniqueLabel = (...groups) => {
|
|
1556
|
+
for (const group of groups) {
|
|
1557
|
+
const labels = Array.from(new Set((group || []).map(sanitizeArtifactLabel).filter(Boolean)));
|
|
1558
|
+
if (labels.length === 1) return labels[0];
|
|
1559
|
+
}
|
|
1560
|
+
return undefined;
|
|
1561
|
+
};
|
|
1531
1562
|
|
|
1532
|
-
const candidates =
|
|
1533
|
-
.map((button) => {
|
|
1534
|
-
const
|
|
1535
|
-
if (!isFileLabel(label)) return null;
|
|
1563
|
+
const candidates = interactiveElements(host)
|
|
1564
|
+
.map((button, index) => {
|
|
1565
|
+
const controlLabel = normalize(button.textContent || button.getAttribute('aria-label') || button.getAttribute('title'));
|
|
1536
1566
|
const paragraph = button.closest('p');
|
|
1537
1567
|
const listItem = button.closest('li');
|
|
1538
1568
|
const focusable = focusableFor(button);
|
|
1539
|
-
const
|
|
1540
|
-
const
|
|
1569
|
+
const ownArtifactLabels = extractArtifactLabels(controlLabel);
|
|
1570
|
+
const paragraphArtifactLabels = artifactLabelsForNode(paragraph);
|
|
1571
|
+
const listItemArtifactLabels = artifactLabelsForNode(listItem);
|
|
1572
|
+
const focusableArtifactLabels = artifactLabelsForNode(focusable);
|
|
1573
|
+
const label = uniqueLabel(ownArtifactLabels, listItemArtifactLabels, paragraphArtifactLabels, focusableArtifactLabels);
|
|
1574
|
+
if (!label && !isFileLabel(controlLabel) && !isDownloadControl(controlLabel)) return null;
|
|
1575
|
+
if (!label) return null;
|
|
1576
|
+
const marker = artifactPrefix + index;
|
|
1577
|
+
button.setAttribute(artifactMarkerAttr, marker);
|
|
1541
1578
|
return {
|
|
1542
1579
|
label,
|
|
1580
|
+
selector: '[' + artifactMarkerAttr + '="' + marker + '"]',
|
|
1581
|
+
controlLabel,
|
|
1543
1582
|
paragraphText: normalize(paragraph?.textContent),
|
|
1544
1583
|
listItemText: normalize(listItem?.textContent),
|
|
1545
|
-
|
|
1546
|
-
|
|
1547
|
-
|
|
1548
|
-
|
|
1549
|
-
|
|
1584
|
+
paragraphInteractiveCount: interactiveElements(paragraph).length,
|
|
1585
|
+
paragraphArtifactLabelCount: Array.from(new Set(paragraphArtifactLabels)).length,
|
|
1586
|
+
paragraphOtherTextLength: otherTextLength(paragraph?.textContent, [...paragraphArtifactLabels, ...interactiveLabels(paragraph)]),
|
|
1587
|
+
listItemInteractiveCount: interactiveElements(listItem).length,
|
|
1588
|
+
listItemArtifactLabelCount: Array.from(new Set(listItemArtifactLabels)).length,
|
|
1589
|
+
focusableInteractiveCount: interactiveElements(focusable).length,
|
|
1590
|
+
focusableArtifactLabelCount: Array.from(new Set(focusableArtifactLabels)).length,
|
|
1591
|
+
focusableOtherTextLength: otherTextLength(focusable?.textContent, [...focusableArtifactLabels, ...interactiveLabels(focusable)]),
|
|
1550
1592
|
};
|
|
1551
1593
|
})
|
|
1552
1594
|
.filter(Boolean);
|
|
@@ -1555,10 +1597,26 @@ async function collectArtifactCandidates(job, responseIndex, responseText = "")
|
|
|
1555
1597
|
`),
|
|
1556
1598
|
);
|
|
1557
1599
|
|
|
1600
|
+
const partitioned = partitionStructuralArtifactCandidates(structural?.candidates || []);
|
|
1601
|
+
const snapshotEntries = parseSnapshotEntries(targetSlice);
|
|
1602
|
+
const hasGenericArtifactControl = snapshotEntries.some(
|
|
1603
|
+
(entry) =>
|
|
1604
|
+
(entry.kind === "button" || entry.kind === "link") &&
|
|
1605
|
+
!entry.disabled &&
|
|
1606
|
+
/(?:^|\b)(?:download|save)(?:\b|$)/i.test(`${entry.label || ""} ${entry.value || ""}`),
|
|
1607
|
+
);
|
|
1608
|
+
const suspiciousFromText = hasGenericArtifactControl
|
|
1609
|
+
? extractArtifactLabels(responseText)
|
|
1610
|
+
.filter((label) => !partitioned.confirmed.some((candidate) => candidate.label === label) && !partitioned.suspicious.some((candidate) => candidate.label === label))
|
|
1611
|
+
.map((label) => ({ label }))
|
|
1612
|
+
: [];
|
|
1613
|
+
|
|
1558
1614
|
return {
|
|
1559
1615
|
snapshot,
|
|
1560
1616
|
targetSlice,
|
|
1561
|
-
candidates:
|
|
1617
|
+
candidates: partitioned.confirmed,
|
|
1618
|
+
suspiciousLabels: [...partitioned.suspicious.map((candidate) => candidate.label), ...suspiciousFromText.map((candidate) => candidate.label)]
|
|
1619
|
+
.filter((label, index, labels) => labels.indexOf(label) === index),
|
|
1562
1620
|
};
|
|
1563
1621
|
}
|
|
1564
1622
|
|
|
@@ -1566,11 +1624,14 @@ async function waitForStableArtifactCandidates(job, responseIndex, responseText
|
|
|
1566
1624
|
const deadline = Date.now() + ARTIFACT_CANDIDATE_STABILITY_TIMEOUT_MS;
|
|
1567
1625
|
let lastSignature;
|
|
1568
1626
|
let stablePolls = 0;
|
|
1569
|
-
let latest = { snapshot: "", targetSlice: undefined, candidates: [] };
|
|
1627
|
+
let latest = { snapshot: "", targetSlice: undefined, candidates: [], suspiciousLabels: [] };
|
|
1570
1628
|
|
|
1571
1629
|
while (Date.now() < deadline) {
|
|
1572
1630
|
latest = await collectArtifactCandidates(job, responseIndex, responseText);
|
|
1573
|
-
const signature =
|
|
1631
|
+
const signature = JSON.stringify({
|
|
1632
|
+
candidates: latest.candidates.map((candidate) => candidate.label),
|
|
1633
|
+
suspiciousLabels: latest.suspiciousLabels,
|
|
1634
|
+
});
|
|
1574
1635
|
if (signature === lastSignature) stablePolls += 1;
|
|
1575
1636
|
else {
|
|
1576
1637
|
lastSignature = signature;
|
|
@@ -1628,7 +1689,7 @@ async function downloadArtifacts(job, responseIndex, responseText = "") {
|
|
|
1628
1689
|
return [];
|
|
1629
1690
|
}
|
|
1630
1691
|
|
|
1631
|
-
|
|
1692
|
+
let { targetSlice, candidates, suspiciousLabels } = await reopenConversationForArtifacts(job, responseIndex, responseText, "initial");
|
|
1632
1693
|
if (!targetSlice) {
|
|
1633
1694
|
await log(`No assistant response found in snapshot for response index ${responseIndex}`);
|
|
1634
1695
|
await secureWriteText(`${jobDir}/artifacts.json`, "[]\n");
|
|
@@ -1637,33 +1698,32 @@ async function downloadArtifacts(job, responseIndex, responseText = "") {
|
|
|
1637
1698
|
}
|
|
1638
1699
|
|
|
1639
1700
|
await log(`Artifact candidates: ${candidates.map((candidate) => candidate.label).join(", ") || "(none)"}`);
|
|
1701
|
+
if (suspiciousLabels.length > 0) {
|
|
1702
|
+
await log(`Suspicious artifact signals: ${suspiciousLabels.join(", ")}`);
|
|
1703
|
+
}
|
|
1640
1704
|
|
|
1641
1705
|
const artifactsDir = `${jobDir}/artifacts`;
|
|
1642
1706
|
await ensurePrivateDir(artifactsDir);
|
|
1643
1707
|
const artifacts = [];
|
|
1644
1708
|
await flushArtifactsState(artifacts);
|
|
1645
1709
|
|
|
1646
|
-
for (const [index,
|
|
1710
|
+
for (const [index, originalCandidate] of candidates.entries()) {
|
|
1647
1711
|
let downloaded = false;
|
|
1712
|
+
let activeCandidate = originalCandidate;
|
|
1648
1713
|
for (let attempt = 1; attempt <= ARTIFACT_DOWNLOAD_MAX_ATTEMPTS && !downloaded; attempt += 1) {
|
|
1649
|
-
|
|
1650
|
-
|
|
1651
|
-
|
|
1652
|
-
|
|
1653
|
-
const entry = freshEntries.find(
|
|
1654
|
-
(artifactEntry) => artifactEntry.label === candidate.label && (artifactEntry.kind === "button" || artifactEntry.kind === "link") && !artifactEntry.disabled,
|
|
1655
|
-
);
|
|
1656
|
-
if (!entry) {
|
|
1657
|
-
await log(`Artifact "${candidate.label}" not found in fresh snapshot, skipping`);
|
|
1714
|
+
if (!activeCandidate?.selector) {
|
|
1715
|
+
await log(`Artifact "${originalCandidate.label}" has no live selector, marking unconfirmed`);
|
|
1716
|
+
artifacts.push({ displayName: originalCandidate.label, unconfirmed: true, error: "Artifact candidate lost its live selector before download." });
|
|
1717
|
+
await flushArtifactsState(artifacts);
|
|
1658
1718
|
break;
|
|
1659
1719
|
}
|
|
1660
1720
|
|
|
1661
|
-
const destinationPath = join(artifactsDir, preferredArtifactName(
|
|
1721
|
+
const destinationPath = join(artifactsDir, preferredArtifactName(originalCandidate.label, index));
|
|
1662
1722
|
await rm(destinationPath, { force: true }).catch(() => undefined);
|
|
1663
1723
|
try {
|
|
1664
|
-
await log(`Artifact "${
|
|
1724
|
+
await log(`Artifact "${originalCandidate.label}" download attempt ${attempt}/${ARTIFACT_DOWNLOAD_MAX_ATTEMPTS} using selector ${activeCandidate.selector}`);
|
|
1665
1725
|
await withHeartbeatWhile(() =>
|
|
1666
|
-
agentBrowser(job, "download",
|
|
1726
|
+
agentBrowser(job, "download", activeCandidate.selector, destinationPath, {
|
|
1667
1727
|
timeoutMs: ARTIFACT_DOWNLOAD_TIMEOUT_MS,
|
|
1668
1728
|
}),
|
|
1669
1729
|
);
|
|
@@ -1675,7 +1735,7 @@ async function downloadArtifacts(job, responseIndex, responseText = "") {
|
|
|
1675
1735
|
detectType(destinationPath),
|
|
1676
1736
|
]);
|
|
1677
1737
|
artifacts.push({
|
|
1678
|
-
displayName:
|
|
1738
|
+
displayName: originalCandidate.label,
|
|
1679
1739
|
fileName: basename(destinationPath),
|
|
1680
1740
|
copiedPath: destinationPath,
|
|
1681
1741
|
size,
|
|
@@ -1686,11 +1746,15 @@ async function downloadArtifacts(job, responseIndex, responseText = "") {
|
|
|
1686
1746
|
} catch (error) {
|
|
1687
1747
|
const message = error instanceof Error ? error.message : String(error);
|
|
1688
1748
|
await rm(destinationPath, { force: true }).catch(() => undefined);
|
|
1689
|
-
await log(`Artifact "${
|
|
1749
|
+
await log(`Artifact "${originalCandidate.label}" download failed on attempt ${attempt}/${ARTIFACT_DOWNLOAD_MAX_ATTEMPTS}: ${message}`);
|
|
1690
1750
|
if (attempt >= ARTIFACT_DOWNLOAD_MAX_ATTEMPTS) {
|
|
1691
|
-
artifacts.push({ displayName:
|
|
1751
|
+
artifacts.push({ displayName: originalCandidate.label, unconfirmed: true, error: message });
|
|
1692
1752
|
} else {
|
|
1693
|
-
await reopenConversationForArtifacts(job, responseIndex, responseText, `retry ${attempt + 1} for ${
|
|
1753
|
+
const refreshed = await reopenConversationForArtifacts(job, responseIndex, responseText, `retry ${attempt + 1} for ${originalCandidate.label}`);
|
|
1754
|
+
targetSlice = refreshed.targetSlice;
|
|
1755
|
+
candidates = refreshed.candidates;
|
|
1756
|
+
suspiciousLabels = refreshed.suspiciousLabels;
|
|
1757
|
+
activeCandidate = candidates.find((candidate) => candidate.label === originalCandidate.label);
|
|
1694
1758
|
await sleep(1_000);
|
|
1695
1759
|
}
|
|
1696
1760
|
} finally {
|
|
@@ -1699,6 +1763,16 @@ async function downloadArtifacts(job, responseIndex, responseText = "") {
|
|
|
1699
1763
|
}
|
|
1700
1764
|
}
|
|
1701
1765
|
|
|
1766
|
+
const capturedArtifactLabels = new Set(artifacts.map((artifact) => artifact.displayName).filter(Boolean));
|
|
1767
|
+
const missedArtifactLabels = suspiciousLabels.filter((label) => !capturedArtifactLabels.has(label));
|
|
1768
|
+
if (missedArtifactLabels.length > 0) {
|
|
1769
|
+
await log(`Marking missed artifact signals as unconfirmed: ${missedArtifactLabels.join(", ")}`);
|
|
1770
|
+
for (const label of missedArtifactLabels) {
|
|
1771
|
+
artifacts.push({ displayName: label, unconfirmed: true, error: "Response-local artifact signal was present, but no downloadable artifact was captured." });
|
|
1772
|
+
}
|
|
1773
|
+
await flushArtifactsState(artifacts);
|
|
1774
|
+
}
|
|
1775
|
+
|
|
1702
1776
|
return artifacts;
|
|
1703
1777
|
}
|
|
1704
1778
|
|
|
@@ -1759,9 +1833,10 @@ async function run() {
|
|
|
1759
1833
|
currentJob = await mutateJob((job) => ({ ...job, ...phasePatch("downloading_artifacts", { heartbeatAt: new Date().toISOString() }) }));
|
|
1760
1834
|
const artifacts = await downloadArtifacts(currentJob, completion.responseIndex, completion.responseText);
|
|
1761
1835
|
const artifactFailureCount = artifacts.filter((artifact) => artifact.unconfirmed || artifact.error).length;
|
|
1836
|
+
const finalPhase = artifactFailureCount > 0 ? "complete_with_artifact_errors" : "complete";
|
|
1762
1837
|
|
|
1763
1838
|
await heartbeat(
|
|
1764
|
-
phasePatch(
|
|
1839
|
+
phasePatch(finalPhase, {
|
|
1765
1840
|
status: "complete",
|
|
1766
1841
|
completedAt: new Date().toISOString(),
|
|
1767
1842
|
responsePath: currentJob.responsePath,
|
|
@@ -1773,7 +1848,7 @@ async function run() {
|
|
|
1773
1848
|
);
|
|
1774
1849
|
const persistedJob = await readJob().catch(() => undefined);
|
|
1775
1850
|
await log(`Persisted final status after completion write: ${persistedJob?.status || "unknown"}`);
|
|
1776
|
-
await log(`Job ${currentJob.id} complete`);
|
|
1851
|
+
await log(`Job ${currentJob.id} complete (${finalPhase}, artifact failures=${artifactFailureCount})`);
|
|
1777
1852
|
} catch (error) {
|
|
1778
1853
|
if (!shuttingDown) {
|
|
1779
1854
|
const message = error instanceof Error ? error.message : String(error);
|
package/package.json
CHANGED
package/prompts/oracle.md
CHANGED
|
@@ -13,6 +13,12 @@ Required workflow:
|
|
|
13
13
|
5. Call oracle_submit with the prompt and exact archive inputs.
|
|
14
14
|
6. Stop immediately after dispatching the oracle job.
|
|
15
15
|
|
|
16
|
+
Oracle model (`oracle_submit`):
|
|
17
|
+
- To choose a specific ChatGPT model, pass **`preset`** with one of the allowed ids from the tool schema enum / canonical preset registry.
|
|
18
|
+
- **Or** omit **`preset`** entirely to use the configured default model (from oracle config).
|
|
19
|
+
- **`preset`** is the only model-selection parameter on `oracle_submit`. Do not pass `modelFamily`, `effort`, or `autoSwitchToThinking`.
|
|
20
|
+
- If unsure which preset fits the task, ask the user.
|
|
21
|
+
|
|
16
22
|
Rules:
|
|
17
23
|
- Always include an archive. Do not submit without context files.
|
|
18
24
|
- By default, include the whole repository by passing `.`. Default archive exclusions apply automatically, including common bulky outputs and obvious credentials/private data like `.env` files, key material, credential dotfiles, local database files, and root `secrets/` directories.
|
|
@@ -21,8 +27,7 @@ Rules:
|
|
|
21
27
|
- If the request depends on git state or pending changes (for example code review, ship readiness, or release approval), create a tracked diff bundle file inside the repo (for example under `.pi/`) containing `git status` plus `git diff` output, include that file in the archive, and tell the oracle to use it because the `.git` directory is not included in oracle exports.
|
|
22
28
|
- When `files=["."]` and the post-exclusion archive is still too large, submit automatically prunes the largest nested directories matching generic generated-output names like `build/`, `dist/`, `out/`, `coverage/`, and `tmp/` outside obvious source roots like `src/` and `lib/` until the archive fits or no candidate remains. Successful submissions report what was pruned.
|
|
23
29
|
- If a submitted oracle job later fails because upload is rejected, retry with a smaller archive in this order: (1) remove the largest obviously irrelevant/generated content, (2) if still too large, include modified files plus adjacent files plus directly relevant subtrees, (3) if still too large, explain the cut or ask the user.
|
|
24
|
-
- Prefer the configured default
|
|
25
|
-
- Only use autoSwitchToThinking with the instant model family.
|
|
30
|
+
- Prefer the configured default (omit **`preset`**) unless the task clearly needs a different model; then choose a **`preset`** id from the tool schema enum.
|
|
26
31
|
- If `oracle_submit` itself fails because the local archive still exceeds the upload limit after default exclusions and automatic generic generated-output-dir pruning, or for any other submit-time error, stop and report the error. Do not retry automatically.
|
|
27
32
|
- If `oracle_submit` returns a queued job instead of an immediately dispatched one, treat that as success and end your turn exactly the same way.
|
|
28
33
|
- After oracle_submit returns, end your turn. Do not keep working while the oracle runs.
|