principles-disciple 1.26.0 → 1.27.0
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/openclaw.plugin.json
CHANGED
|
@@ -2,7 +2,7 @@
|
|
|
2
2
|
"id": "principles-disciple",
|
|
3
3
|
"name": "Principles Disciple",
|
|
4
4
|
"description": "Evolutionary programming agent framework with strategic guardrails and reflection loops.",
|
|
5
|
-
"version": "1.
|
|
5
|
+
"version": "1.27.0",
|
|
6
6
|
"skills": [
|
|
7
7
|
"./skills"
|
|
8
8
|
],
|
|
@@ -76,8 +76,8 @@
|
|
|
76
76
|
}
|
|
77
77
|
},
|
|
78
78
|
"buildFingerprint": {
|
|
79
|
-
"gitSha": "
|
|
80
|
-
"bundleMd5": "
|
|
81
|
-
"builtAt": "2026-04-
|
|
79
|
+
"gitSha": "9037c29faf9a",
|
|
80
|
+
"bundleMd5": "e7e9ebf7f67083f72f8f4328314d5670",
|
|
81
|
+
"builtAt": "2026-04-13T06:10:29.622Z"
|
|
82
82
|
}
|
|
83
83
|
}
|
package/package.json
CHANGED
package/src/core/trajectory.ts
CHANGED
|
@@ -961,6 +961,17 @@ export class TrajectoryDatabase {
|
|
|
961
961
|
throw new SampleNotFoundError(`${sampleId} (after update)`);
|
|
962
962
|
}
|
|
963
963
|
|
|
964
|
+
// #Phase2b: Emit pain event for rejected corrections
|
|
965
|
+
if (status === 'rejected') {
|
|
966
|
+
this.recordCorrectionRejectedPain({
|
|
967
|
+
session_id: record.session_id,
|
|
968
|
+
quality_score: record.quality_score,
|
|
969
|
+
diff_excerpt: record.diff_excerpt,
|
|
970
|
+
principle_ids_json: record.principle_ids_json,
|
|
971
|
+
created_at: record.created_at,
|
|
972
|
+
});
|
|
973
|
+
}
|
|
974
|
+
|
|
964
975
|
return {
|
|
965
976
|
sampleId: String(record.sample_id),
|
|
966
977
|
sessionId: String(record.session_id),
|
|
@@ -977,6 +988,43 @@ export class TrajectoryDatabase {
|
|
|
977
988
|
};
|
|
978
989
|
}
|
|
979
990
|
|
|
991
|
+
/**
|
|
992
|
+
* When a correction sample is rejected, emit a pain event to the trajectory.
|
|
993
|
+
* This feeds rejected corrections into the nocturnal pipeline as a high-fidelity
|
|
994
|
+
* violation signal (human-verified, unlike heuristic pain detection).
|
|
995
|
+
*/
|
|
996
|
+
private recordCorrectionRejectedPain(record: {
|
|
997
|
+
session_id: unknown;
|
|
998
|
+
quality_score: unknown;
|
|
999
|
+
diff_excerpt: unknown;
|
|
1000
|
+
principle_ids_json: unknown;
|
|
1001
|
+
created_at: unknown;
|
|
1002
|
+
}): void {
|
|
1003
|
+
const sessionId = String(record.session_id);
|
|
1004
|
+
const qualityScore = Number(record.quality_score);
|
|
1005
|
+
const diffExcerpt = String(record.diff_excerpt ?? '');
|
|
1006
|
+
const principleIds = String(record.principle_ids_json ?? '[]');
|
|
1007
|
+
// quality_score (0-100) from correction sample → pain score (0-100), clamped
|
|
1008
|
+
const painScore = Math.max(0, Math.min(100, Math.round(Number(qualityScore) || 0)));
|
|
1009
|
+
const reason = `Correction rejected (quality ${qualityScore.toFixed(2)}). Principles: ${principleIds}${diffExcerpt ? ` — ${diffExcerpt.slice(0, 120)}` : ''}`;
|
|
1010
|
+
|
|
1011
|
+
try {
|
|
1012
|
+
this.recordPainEvent({
|
|
1013
|
+
sessionId,
|
|
1014
|
+
source: 'correction_rejected',
|
|
1015
|
+
score: painScore,
|
|
1016
|
+
reason,
|
|
1017
|
+
severity: painScore >= 70 ? 'severe' : painScore >= 40 ? 'moderate' : 'mild',
|
|
1018
|
+
origin: 'system_infer',
|
|
1019
|
+
text: diffExcerpt || undefined,
|
|
1020
|
+
createdAt: String(record.created_at),
|
|
1021
|
+
});
|
|
1022
|
+
} catch (err) {
|
|
1023
|
+
// Non-fatal: pain event recording should not break the review flow
|
|
1024
|
+
console.warn(`[Trajectory] Failed to record correction_rejected pain event: ${String(err)}`);
|
|
1025
|
+
}
|
|
1026
|
+
}
|
|
1027
|
+
|
|
980
1028
|
/**
|
|
981
1029
|
* Export correction samples to JSONL file.
|
|
982
1030
|
*
|
|
@@ -1496,6 +1544,11 @@ export class TrajectoryDatabase {
|
|
|
1496
1544
|
`).get(sessionId) as Record<string, unknown> | undefined;
|
|
1497
1545
|
if (!correctionTurn || !correctionTurn.references_assistant_turn_id) return;
|
|
1498
1546
|
|
|
1547
|
+
// #Phase2b-fix: Tool failure is NOT required for correction samples.
|
|
1548
|
+
// User corrections are the highest-fidelity signal — they indicate the agent
|
|
1549
|
+
// said or did something wrong, regardless of whether tool calls succeeded.
|
|
1550
|
+
// Requiring tool failure excluded the most valuable cases: "agent did something
|
|
1551
|
+
// that technically worked but violated a principle or was logically wrong."
|
|
1499
1552
|
const failedCall = this.db.prepare(`
|
|
1500
1553
|
SELECT id, tool_name, error_type, error_message
|
|
1501
1554
|
FROM tool_calls
|
|
@@ -1503,26 +1556,37 @@ export class TrajectoryDatabase {
|
|
|
1503
1556
|
ORDER BY id DESC
|
|
1504
1557
|
LIMIT 1
|
|
1505
1558
|
`).get(sessionId) as Record<string, unknown> | undefined;
|
|
1506
|
-
if (!failedCall) return;
|
|
1507
1559
|
|
|
1508
|
-
const
|
|
1509
|
-
SELECT id, tool_name
|
|
1560
|
+
const recentCalls = this.db.prepare(`
|
|
1561
|
+
SELECT id, tool_name, outcome
|
|
1510
1562
|
FROM tool_calls
|
|
1511
|
-
WHERE session_id = ?
|
|
1563
|
+
WHERE session_id = ?
|
|
1512
1564
|
ORDER BY id DESC
|
|
1513
|
-
LIMIT
|
|
1565
|
+
LIMIT 5
|
|
1514
1566
|
`).all(sessionId) as Record<string, unknown>[];
|
|
1515
|
-
if (successfulCalls.length === 0) return;
|
|
1516
1567
|
|
|
1517
|
-
const
|
|
1568
|
+
const successfulCalls = recentCalls.filter(c => c.outcome === 'success');
|
|
1569
|
+
|
|
1570
|
+
// Generate sample ID from correction turn + first recent call (or correction id if no calls)
|
|
1571
|
+
const refForHash = successfulCalls[0]?.id ?? correctionTurn.id;
|
|
1572
|
+
const sampleId = `sample_${crypto.createHash('md5').update(`${sessionId}:${correctionTurn.id}:${refForHash}`).digest('hex').slice(0, 12)}`;
|
|
1518
1573
|
const userRawText = this.restoreRawText(correctionTurn.raw_text as string | null, correctionTurn.blob_ref as string | null);
|
|
1574
|
+
|
|
1575
|
+
// Quality scoring: correction cue is always valuable
|
|
1576
|
+
// Tool failure adds context (20pts), successful calls add context (up to 15pts)
|
|
1577
|
+
// Pure conversation corrections still score 55-75 (high enough to review)
|
|
1519
1578
|
const qualityScore = [
|
|
1520
1579
|
correctionTurn.references_assistant_turn_id ? 35 : 0,
|
|
1521
1580
|
correctionTurn.correction_cue ? 20 : 0,
|
|
1522
1581
|
failedCall ? 20 : 0,
|
|
1523
|
-
successfulCalls.length
|
|
1582
|
+
Math.min(successfulCalls.length, 3) * 5,
|
|
1524
1583
|
].reduce((sum, value) => sum + value, 0);
|
|
1525
1584
|
|
|
1585
|
+
// Diff excerpt: prefer user correction text, fallback to error info, fallback to cue
|
|
1586
|
+
const diffText = userRawText
|
|
1587
|
+
|| (failedCall ? String(failedCall.error_message ?? failedCall.error_type ?? failedCall.tool_name) : '')
|
|
1588
|
+
|| String(correctionTurn.correction_cue ?? 'user correction');
|
|
1589
|
+
|
|
1526
1590
|
this.withWrite(() => {
|
|
1527
1591
|
this.db.prepare(`
|
|
1528
1592
|
INSERT OR IGNORE INTO correction_samples (
|
|
@@ -1535,8 +1599,8 @@ export class TrajectoryDatabase {
|
|
|
1535
1599
|
sessionId,
|
|
1536
1600
|
Number(correctionTurn.references_assistant_turn_id),
|
|
1537
1601
|
Number(correctionTurn.id),
|
|
1538
|
-
safeJson(successfulCalls.map((call) => ({ id: call.id, toolName: call.tool_name }))),
|
|
1539
|
-
summarizeForDiff(
|
|
1602
|
+
safeJson(successfulCalls.map((call) => ({ id: call.id, toolName: call.tool_name, outcome: call.outcome }))),
|
|
1603
|
+
summarizeForDiff(diffText),
|
|
1540
1604
|
'[]',
|
|
1541
1605
|
qualityScore,
|
|
1542
1606
|
nowIso(),
|
|
@@ -0,0 +1,180 @@
|
|
|
1
|
+
import { describe, it, expect, beforeEach, afterEach } from 'vitest';
|
|
2
|
+
import * as fs from 'fs';
|
|
3
|
+
import * as os from 'os';
|
|
4
|
+
import * as path from 'path';
|
|
5
|
+
import { TrajectoryDatabase } from '../../src/core/trajectory.js';
|
|
6
|
+
|
|
7
|
+
function safeRmDir(dir: string): void {
|
|
8
|
+
try { fs.rmSync(dir, { recursive: true, force: true }); } catch { /* ignore */ }
|
|
9
|
+
}
|
|
10
|
+
|
|
11
|
+
describe('Trajectory — correction_rejected pain event (Phase 2b)', () => {
|
|
12
|
+
let workspaceDir: string;
|
|
13
|
+
let trajectory: TrajectoryDatabase;
|
|
14
|
+
|
|
15
|
+
beforeEach(() => {
|
|
16
|
+
workspaceDir = fs.mkdtempSync(path.join(os.tmpdir(), 'pd-correction-pain-'));
|
|
17
|
+
trajectory = new TrajectoryDatabase({ workspaceDir });
|
|
18
|
+
});
|
|
19
|
+
|
|
20
|
+
afterEach(() => {
|
|
21
|
+
trajectory?.dispose();
|
|
22
|
+
safeRmDir(workspaceDir);
|
|
23
|
+
});
|
|
24
|
+
|
|
25
|
+
it('emits a pain event when a correction sample is rejected', async () => {
|
|
26
|
+
// Step 1: Create a session
|
|
27
|
+
trajectory.recordSession({
|
|
28
|
+
sessionId: 'test-session-001',
|
|
29
|
+
startedAt: new Date().toISOString(),
|
|
30
|
+
});
|
|
31
|
+
|
|
32
|
+
// Step 2: Create an assistant turn (to be referenced)
|
|
33
|
+
const assistantTurnId = trajectory.recordAssistantTurn({
|
|
34
|
+
sessionId: 'test-session-001',
|
|
35
|
+
turnIndex: 0,
|
|
36
|
+
rawText: 'Here is some code I wrote',
|
|
37
|
+
createdAt: new Date().toISOString(),
|
|
38
|
+
});
|
|
39
|
+
|
|
40
|
+
// Step 3: Create a user turn with correction_cue (triggers auto-creation)
|
|
41
|
+
trajectory.recordUserTurn({
|
|
42
|
+
sessionId: 'test-session-001',
|
|
43
|
+
turnIndex: 1,
|
|
44
|
+
rawText: 'This is wrong. Fix it properly.',
|
|
45
|
+
correctionDetected: true,
|
|
46
|
+
correctionCue: 'This is wrong. Fix it properly.',
|
|
47
|
+
referencesAssistantTurnId: assistantTurnId,
|
|
48
|
+
createdAt: new Date().toISOString(),
|
|
49
|
+
});
|
|
50
|
+
|
|
51
|
+
// Wait for async sample creation
|
|
52
|
+
await new Promise(resolve => setTimeout(resolve, 50));
|
|
53
|
+
|
|
54
|
+
// Verify sample was created as pending
|
|
55
|
+
const pendingSamples = trajectory.listCorrectionSamples('pending');
|
|
56
|
+
expect(pendingSamples.length).toBe(1);
|
|
57
|
+
const sampleId = pendingSamples[0].sampleId;
|
|
58
|
+
|
|
59
|
+
// Verify no pain events yet
|
|
60
|
+
const painEventsBefore = trajectory.listPainEventsForSession('test-session-001');
|
|
61
|
+
expect(painEventsBefore.length).toBe(0);
|
|
62
|
+
|
|
63
|
+
// Step 4: Review as rejected
|
|
64
|
+
trajectory.reviewCorrectionSample(sampleId, 'rejected', 'Does not match requirements');
|
|
65
|
+
|
|
66
|
+
// Step 5: Verify pain event was created
|
|
67
|
+
const painEventsAfter = trajectory.listPainEventsForSession('test-session-001');
|
|
68
|
+
expect(painEventsAfter.length).toBe(1);
|
|
69
|
+
|
|
70
|
+
const painEvent = painEventsAfter[0];
|
|
71
|
+
expect(painEvent.source).toBe('correction_rejected');
|
|
72
|
+
expect(painEvent.reason).toContain('Correction rejected');
|
|
73
|
+
expect(painEvent.origin).toBe('system_infer');
|
|
74
|
+
});
|
|
75
|
+
|
|
76
|
+
it('does NOT emit a pain event when a correction sample is approved', async () => {
|
|
77
|
+
// Setup: session + assistant turn + user correction turn
|
|
78
|
+
trajectory.recordSession({
|
|
79
|
+
sessionId: 'test-session-002',
|
|
80
|
+
startedAt: new Date().toISOString(),
|
|
81
|
+
});
|
|
82
|
+
const assistantTurnId = trajectory.recordAssistantTurn({
|
|
83
|
+
sessionId: 'test-session-002',
|
|
84
|
+
turnIndex: 0,
|
|
85
|
+
rawText: 'Here is some code',
|
|
86
|
+
createdAt: new Date().toISOString(),
|
|
87
|
+
});
|
|
88
|
+
trajectory.recordUserTurn({
|
|
89
|
+
sessionId: 'test-session-002',
|
|
90
|
+
turnIndex: 1,
|
|
91
|
+
rawText: 'This needs work',
|
|
92
|
+
correctionDetected: true,
|
|
93
|
+
correctionCue: 'This needs work',
|
|
94
|
+
referencesAssistantTurnId: assistantTurnId,
|
|
95
|
+
createdAt: new Date().toISOString(),
|
|
96
|
+
});
|
|
97
|
+
|
|
98
|
+
// Wait for async sample creation
|
|
99
|
+
await new Promise(resolve => setTimeout(resolve, 50));
|
|
100
|
+
|
|
101
|
+
// Get pending sample
|
|
102
|
+
const pendingSamples = trajectory.listCorrectionSamples('pending');
|
|
103
|
+
expect(pendingSamples.length).toBe(1);
|
|
104
|
+
const sampleId = pendingSamples[0].sampleId;
|
|
105
|
+
|
|
106
|
+
// Review as approved - should NOT trigger pain event
|
|
107
|
+
trajectory.reviewCorrectionSample(sampleId, 'approved', 'Looks good');
|
|
108
|
+
|
|
109
|
+
// Verify NO pain event was created (approved != rejected)
|
|
110
|
+
const painEvents = trajectory.listPainEventsForSession('test-session-002');
|
|
111
|
+
expect(painEvents.length).toBe(0);
|
|
112
|
+
});
|
|
113
|
+
|
|
114
|
+
it('maps quality_score to pain_score correctly (0-100 range)', async () => {
|
|
115
|
+
// Setup: with quality score components
|
|
116
|
+
trajectory.recordSession({
|
|
117
|
+
sessionId: 'test-session-003',
|
|
118
|
+
startedAt: new Date().toISOString(),
|
|
119
|
+
});
|
|
120
|
+
const assistantTurnId = trajectory.recordAssistantTurn({
|
|
121
|
+
sessionId: 'test-session-003',
|
|
122
|
+
turnIndex: 0,
|
|
123
|
+
rawText: 'Code here',
|
|
124
|
+
createdAt: new Date().toISOString(),
|
|
125
|
+
});
|
|
126
|
+
|
|
127
|
+
// Create user turn with correction_cue (adds 20 points)
|
|
128
|
+
trajectory.recordUserTurn({
|
|
129
|
+
sessionId: 'test-session-003',
|
|
130
|
+
turnIndex: 1,
|
|
131
|
+
rawText: 'Wrong approach. Try a different algorithm.',
|
|
132
|
+
correctionDetected: true,
|
|
133
|
+
correctionCue: 'Wrong approach. Try a different algorithm.',
|
|
134
|
+
referencesAssistantTurnId: assistantTurnId,
|
|
135
|
+
createdAt: new Date().toISOString(),
|
|
136
|
+
});
|
|
137
|
+
|
|
138
|
+
// Add a failed tool call (adds 20 points)
|
|
139
|
+
trajectory.recordToolCall({
|
|
140
|
+
sessionId: 'test-session-003',
|
|
141
|
+
turnIndex: 2,
|
|
142
|
+
toolName: 'write',
|
|
143
|
+
toolCallIndex: 0,
|
|
144
|
+
paramsJson: { path: '/tmp/test.txt', content: 'test' },
|
|
145
|
+
outcome: 'failure',
|
|
146
|
+
errorMessage: 'Permission denied',
|
|
147
|
+
errorType: 'PermissionError',
|
|
148
|
+
createdAt: new Date().toISOString(),
|
|
149
|
+
});
|
|
150
|
+
|
|
151
|
+
// Add successful calls (adds 25 points)
|
|
152
|
+
trajectory.recordToolCall({
|
|
153
|
+
sessionId: 'test-session-003',
|
|
154
|
+
turnIndex: 3,
|
|
155
|
+
toolName: 'read',
|
|
156
|
+
toolCallIndex: 1,
|
|
157
|
+
paramsJson: { path: '/tmp/test.txt' },
|
|
158
|
+
outcome: 'success',
|
|
159
|
+
resultJson: { content: 'file content' },
|
|
160
|
+
createdAt: new Date().toISOString(),
|
|
161
|
+
});
|
|
162
|
+
|
|
163
|
+
// Wait for async sample creation
|
|
164
|
+
await new Promise(resolve => setTimeout(resolve, 50));
|
|
165
|
+
|
|
166
|
+
// Get pending sample (quality_score ~65: 20 + 20 + 25)
|
|
167
|
+
const pendingSamples = trajectory.listCorrectionSamples('pending');
|
|
168
|
+
expect(pendingSamples.length).toBe(1);
|
|
169
|
+
const sampleId = pendingSamples[0].sampleId;
|
|
170
|
+
|
|
171
|
+
// Review as rejected
|
|
172
|
+
trajectory.reviewCorrectionSample(sampleId, 'rejected', 'Test rejection');
|
|
173
|
+
|
|
174
|
+
// Verify pain score is clamped to 0-100
|
|
175
|
+
const painEvents = trajectory.listPainEventsForSession('test-session-003');
|
|
176
|
+
expect(painEvents.length).toBe(1);
|
|
177
|
+
expect(painEvents[0].score).toBeGreaterThanOrEqual(0);
|
|
178
|
+
expect(painEvents[0].score).toBeLessThanOrEqual(100);
|
|
179
|
+
});
|
|
180
|
+
});
|