open-agents-ai 0.187.523 → 0.187.524
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/dist/index.js +119 -64
- package/npm-shrinkwrap.json +2 -2
- package/package.json +1 -1
package/dist/index.js
CHANGED
|
@@ -526383,6 +526383,122 @@ Your hypotheses MUST address this specific error, not generic causes.
|
|
|
526383
526383
|
return null;
|
|
526384
526384
|
}
|
|
526385
526385
|
}
|
|
526386
|
+
/**
|
|
526387
|
+
* REG-61 sliding-window first-edit / sustained-edit nudge.
|
|
526388
|
+
*
|
|
526389
|
+
* Fires when:
|
|
526390
|
+
* • turn ≥ 4 (give the agent a few turns to orient)
|
|
526391
|
+
* • EITHER this run has zero creative edits OR the last creative edit
|
|
526392
|
+
* was ≥ 5 turns ago (re-fires when agent slips back into read mode)
|
|
526393
|
+
* • there have been ≥ 3 read-class tool calls in the trailing window
|
|
526394
|
+
* (since the last creative edit)
|
|
526395
|
+
* • not in the 5-turn cooldown after a previous firing
|
|
526396
|
+
*
|
|
526397
|
+
* Why earlier than REG-44/REG-58: by turn 12 the read habit is locked in
|
|
526398
|
+
* and prompts get ignored. Re-fire-with-cooldown handles the batch524 stax
|
|
526399
|
+
* case where one early file_write satisfied the initial gate, then the
|
|
526400
|
+
* agent spent 22 turns re-reading source. Directive text names the 4 valid
|
|
526401
|
+
* creative-edit tools EXPLICITLY and lists todo_write as NOT counting
|
|
526402
|
+
* (batch524 osm responded to the original directive with todo_write
|
|
526403
|
+
* thinking it qualified). Generic across ecosystems — no manifest/file/
|
|
526404
|
+
* language keywords.
|
|
526405
|
+
*
|
|
526406
|
+
* BFC-61 (root-cause from batch527-midi-solo): refactored from inline
|
|
526407
|
+
* block in the main turn loop into this method so the brute-force
|
|
526408
|
+
* re-engagement inner loop can also invoke it. Without BFC-61, REG-61
|
|
526409
|
+
* stops checking once the runner enters brute-force cycles, allowing
|
|
526410
|
+
* 30+ minutes of read-paralysis (batch527 hit this — 804 consecutive
|
|
526411
|
+
* non-edit calls in brute-force, REG-61 fired 0× in that phase).
|
|
526412
|
+
*
|
|
526413
|
+
* Side effects when firing:
|
|
526414
|
+
* - bumps `this._reg61CooldownUntilTurn`
|
|
526415
|
+
* - pushes a system message with the directive into `messages`
|
|
526416
|
+
* - emits a `status` event with telemetry (turn, reads_in_window, gap)
|
|
526417
|
+
*
|
|
526418
|
+
* State (instance fields, persist across cycles — desired):
|
|
526419
|
+
* - `_lastFileWriteTurn`: bumped by tool dispatch on creative edits
|
|
526420
|
+
* - `_reg61CooldownUntilTurn`: bumped by this method when firing
|
|
526421
|
+
*
|
|
526422
|
+
* @param turn - Current turn counter from the enclosing loop. In
|
|
526423
|
+
* brute-force cycles this resets to 0 each cycle, but the
|
|
526424
|
+
* cooldown / lastFileWriteTurn state is retained across
|
|
526425
|
+
* cycles via instance fields, so behavior is consistent.
|
|
526426
|
+
* @param toolCallLog - Append-only ordered tool-call history.
|
|
526427
|
+
* @param messages - Conversation history to inject the system directive into.
|
|
526428
|
+
* @param cycleLabel - Optional label to disambiguate brute-force fires
|
|
526429
|
+
* from main-loop fires in the status emit (e.g. "bf-cycle 2").
|
|
526430
|
+
*/
|
|
526431
|
+
_runReg61Check(turn, toolCallLog, messages2, cycleLabel) {
|
|
526432
|
+
const REG61_TURN_FLOOR = 4;
|
|
526433
|
+
const REG61_MIN_READS = 3;
|
|
526434
|
+
const REG61_NO_WRITE_GAP = 5;
|
|
526435
|
+
const REG61_COOLDOWN = 5;
|
|
526436
|
+
if (process.env["OA_DISABLE_REG61"] === "1")
|
|
526437
|
+
return;
|
|
526438
|
+
if (turn < REG61_TURN_FLOOR)
|
|
526439
|
+
return;
|
|
526440
|
+
if (turn <= this._reg61CooldownUntilTurn)
|
|
526441
|
+
return;
|
|
526442
|
+
const _writeGate = this._lastFileWriteTurn < 0 || turn - this._lastFileWriteTurn >= REG61_NO_WRITE_GAP;
|
|
526443
|
+
if (!_writeGate)
|
|
526444
|
+
return;
|
|
526445
|
+
const _readClassNames = /* @__PURE__ */ new Set([
|
|
526446
|
+
"file_read",
|
|
526447
|
+
"file_explore",
|
|
526448
|
+
"list_directory",
|
|
526449
|
+
"grep_search",
|
|
526450
|
+
"glob_find",
|
|
526451
|
+
"find_files",
|
|
526452
|
+
"code_neighbors",
|
|
526453
|
+
"repo_map",
|
|
526454
|
+
"codebase_map",
|
|
526455
|
+
"semantic_map",
|
|
526456
|
+
"symbol_search",
|
|
526457
|
+
"todo_read",
|
|
526458
|
+
"memory_read",
|
|
526459
|
+
"memory_search"
|
|
526460
|
+
]);
|
|
526461
|
+
const _editClassNames = /* @__PURE__ */ new Set([
|
|
526462
|
+
"file_write",
|
|
526463
|
+
"file_edit",
|
|
526464
|
+
"batch_edit",
|
|
526465
|
+
"file_patch"
|
|
526466
|
+
]);
|
|
526467
|
+
let _readsInWindow = 0;
|
|
526468
|
+
for (let k = toolCallLog.length - 1; k >= 0; k--) {
|
|
526469
|
+
const _name = toolCallLog[k].name;
|
|
526470
|
+
if (_editClassNames.has(_name))
|
|
526471
|
+
break;
|
|
526472
|
+
if (_readClassNames.has(_name))
|
|
526473
|
+
_readsInWindow++;
|
|
526474
|
+
}
|
|
526475
|
+
if (_readsInWindow < REG61_MIN_READS)
|
|
526476
|
+
return;
|
|
526477
|
+
this._reg61CooldownUntilTurn = turn + REG61_COOLDOWN;
|
|
526478
|
+
const _gapDesc = this._lastFileWriteTurn < 0 ? `no creative edits yet this run` : `${turn - this._lastFileWriteTurn} turns since last creative edit (turn ${this._lastFileWriteTurn})`;
|
|
526479
|
+
const reg61Msg = `[FIRST-EDIT NUDGE — REG-61]
|
|
526480
|
+
You have made ${_readsInWindow} read/exploration calls in the trailing window — ${_gapDesc}. Reading is preparation; writing is progress. Runs that stay in pure-read mode produce zero deliverables.
|
|
526481
|
+
|
|
526482
|
+
Your NEXT tool call MUST be EXACTLY ONE of:
|
|
526483
|
+
• file_write — create a new file
|
|
526484
|
+
• file_edit — modify an existing file (find/replace)
|
|
526485
|
+
• batch_edit — multiple find/replace edits in one call
|
|
526486
|
+
• file_patch — apply a unified diff
|
|
526487
|
+
|
|
526488
|
+
These are the ONLY four tools that count as creative edits. The following do NOT count and will NOT satisfy this directive:
|
|
526489
|
+
• todo_write (only updates the todo list, not the filesystem)
|
|
526490
|
+
• memory_write (writes to memory store, not the project)
|
|
526491
|
+
• list_directory / file_read / file_explore / grep_search / shell
|
|
526492
|
+
|
|
526493
|
+
Pick the SMALLEST concrete deliverable from the spec — typically the project entry point or the file most other modules depend on. Write a stub or skeleton if the full implementation is too large; you can iterate later. Do NOT issue another read or todo update before producing the next file_write/file_edit/batch_edit/file_patch.`;
|
|
526494
|
+
messages2.push({ role: "system", content: reg61Msg });
|
|
526495
|
+
const _cyclePart = cycleLabel ? ` (${cycleLabel})` : "";
|
|
526496
|
+
this.emit({
|
|
526497
|
+
type: "status",
|
|
526498
|
+
content: `REG-61 FIRST-EDIT NUDGE — fired at turn ${turn}${_cyclePart}; reads_in_window=${_readsInWindow}; ${_gapDesc}`,
|
|
526499
|
+
timestamp: (/* @__PURE__ */ new Date()).toISOString()
|
|
526500
|
+
});
|
|
526501
|
+
}
|
|
526386
526502
|
readSessionTodos() {
|
|
526387
526503
|
try {
|
|
526388
526504
|
const sid = process.env["OA_SESSION_ID"] || this._sessionId || "default";
|
|
@@ -528620,70 +528736,7 @@ TASK: ${task}` : task;
|
|
|
528620
528736
|
timestamp: (/* @__PURE__ */ new Date()).toISOString()
|
|
528621
528737
|
});
|
|
528622
528738
|
}
|
|
528623
|
-
|
|
528624
|
-
const REG61_MIN_READS = 3;
|
|
528625
|
-
const REG61_NO_WRITE_GAP = 5;
|
|
528626
|
-
const REG61_COOLDOWN = 5;
|
|
528627
|
-
if (turn >= REG61_TURN_FLOOR && turn > this._reg61CooldownUntilTurn && process.env["OA_DISABLE_REG61"] !== "1") {
|
|
528628
|
-
const _writeGate = this._lastFileWriteTurn < 0 || turn - this._lastFileWriteTurn >= REG61_NO_WRITE_GAP;
|
|
528629
|
-
if (_writeGate) {
|
|
528630
|
-
const _readClassNames = /* @__PURE__ */ new Set([
|
|
528631
|
-
"file_read",
|
|
528632
|
-
"file_explore",
|
|
528633
|
-
"list_directory",
|
|
528634
|
-
"grep_search",
|
|
528635
|
-
"glob_find",
|
|
528636
|
-
"find_files",
|
|
528637
|
-
"code_neighbors",
|
|
528638
|
-
"repo_map",
|
|
528639
|
-
"codebase_map",
|
|
528640
|
-
"semantic_map",
|
|
528641
|
-
"symbol_search",
|
|
528642
|
-
"todo_read",
|
|
528643
|
-
"memory_read",
|
|
528644
|
-
"memory_search"
|
|
528645
|
-
]);
|
|
528646
|
-
let _readsInWindow = 0;
|
|
528647
|
-
const _editClassNames = /* @__PURE__ */ new Set([
|
|
528648
|
-
"file_write",
|
|
528649
|
-
"file_edit",
|
|
528650
|
-
"batch_edit",
|
|
528651
|
-
"file_patch"
|
|
528652
|
-
]);
|
|
528653
|
-
for (let k = toolCallLog.length - 1; k >= 0; k--) {
|
|
528654
|
-
const _name = toolCallLog[k].name;
|
|
528655
|
-
if (_editClassNames.has(_name))
|
|
528656
|
-
break;
|
|
528657
|
-
if (_readClassNames.has(_name))
|
|
528658
|
-
_readsInWindow++;
|
|
528659
|
-
}
|
|
528660
|
-
if (_readsInWindow >= REG61_MIN_READS) {
|
|
528661
|
-
this._reg61CooldownUntilTurn = turn + REG61_COOLDOWN;
|
|
528662
|
-
const _gapDesc = this._lastFileWriteTurn < 0 ? `no creative edits yet this run` : `${turn - this._lastFileWriteTurn} turns since last creative edit (turn ${this._lastFileWriteTurn})`;
|
|
528663
|
-
const reg61Msg = `[FIRST-EDIT NUDGE — REG-61]
|
|
528664
|
-
You have made ${_readsInWindow} read/exploration calls in the trailing window — ${_gapDesc}. Reading is preparation; writing is progress. Runs that stay in pure-read mode produce zero deliverables.
|
|
528665
|
-
|
|
528666
|
-
Your NEXT tool call MUST be EXACTLY ONE of:
|
|
528667
|
-
• file_write — create a new file
|
|
528668
|
-
• file_edit — modify an existing file (find/replace)
|
|
528669
|
-
• batch_edit — multiple find/replace edits in one call
|
|
528670
|
-
• file_patch — apply a unified diff
|
|
528671
|
-
|
|
528672
|
-
These are the ONLY four tools that count as creative edits. The following do NOT count and will NOT satisfy this directive:
|
|
528673
|
-
• todo_write (only updates the todo list, not the filesystem)
|
|
528674
|
-
• memory_write (writes to memory store, not the project)
|
|
528675
|
-
• list_directory / file_read / file_explore / grep_search / shell
|
|
528676
|
-
|
|
528677
|
-
Pick the SMALLEST concrete deliverable from the spec — typically the project entry point or the file most other modules depend on. Write a stub or skeleton if the full implementation is too large; you can iterate later. Do NOT issue another read or todo update before producing the next file_write/file_edit/batch_edit/file_patch.`;
|
|
528678
|
-
messages2.push({ role: "system", content: reg61Msg });
|
|
528679
|
-
this.emit({
|
|
528680
|
-
type: "status",
|
|
528681
|
-
content: `REG-61 FIRST-EDIT NUDGE — fired at turn ${turn}; reads_in_window=${_readsInWindow}; ${_gapDesc}`,
|
|
528682
|
-
timestamp: (/* @__PURE__ */ new Date()).toISOString()
|
|
528683
|
-
});
|
|
528684
|
-
}
|
|
528685
|
-
}
|
|
528686
|
-
}
|
|
528739
|
+
this._runReg61Check(turn, toolCallLog, messages2);
|
|
528687
528740
|
const REG58_NO_WRITE_BUDGET = 30;
|
|
528688
528741
|
if (turn > stagnationCooldownUntilTurn && this._lastFileWriteTurn >= 0 && turn - this._lastFileWriteTurn >= REG58_NO_WRITE_BUDGET && process.env["OA_DISABLE_REG58"] !== "1") {
|
|
528689
528742
|
const gap = turn - this._lastFileWriteTurn;
|
|
@@ -532118,6 +532171,7 @@ Your most recent tool calls SUCCEEDED. If the task is complete, call task_comple
|
|
|
532118
532171
|
content: `Re-engaging — cycle ${bruteForceCycle} (${totalTurns} turns, ${toolCallCount} tool calls so far)`,
|
|
532119
532172
|
timestamp: (/* @__PURE__ */ new Date()).toISOString()
|
|
532120
532173
|
});
|
|
532174
|
+
this._reg61CooldownUntilTurn = -1;
|
|
532121
532175
|
messages2.push({
|
|
532122
532176
|
role: "user",
|
|
532123
532177
|
content: `[CONTINUATION — Cycle ${bruteForceCycle}]
|
|
@@ -532151,6 +532205,7 @@ You have ${this.options.maxTurns} more turns. Continue making progress. Call tas
|
|
|
532151
532205
|
this.emit({ type: "error", content: "Task aborted by user", timestamp: (/* @__PURE__ */ new Date()).toISOString() });
|
|
532152
532206
|
break;
|
|
532153
532207
|
}
|
|
532208
|
+
this._runReg61Check(turn, toolCallLog, messages2, `bf-cycle ${bruteForceCycle}`);
|
|
532154
532209
|
const bfNow = Date.now();
|
|
532155
532210
|
if (bfNow > nextSelfEval) {
|
|
532156
532211
|
selfEvalCount++;
|
package/npm-shrinkwrap.json
CHANGED
|
@@ -1,12 +1,12 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "open-agents-ai",
|
|
3
|
-
"version": "0.187.
|
|
3
|
+
"version": "0.187.524",
|
|
4
4
|
"lockfileVersion": 3,
|
|
5
5
|
"requires": true,
|
|
6
6
|
"packages": {
|
|
7
7
|
"": {
|
|
8
8
|
"name": "open-agents-ai",
|
|
9
|
-
"version": "0.187.
|
|
9
|
+
"version": "0.187.524",
|
|
10
10
|
"hasInstallScript": true,
|
|
11
11
|
"license": "CC-BY-NC-4.0",
|
|
12
12
|
"dependencies": {
|
package/package.json
CHANGED