principles-disciple 1.71.0 → 1.73.0

package/src/tools/write-pain-flag.ts

@@ -1,215 +0,0 @@
-import type { OpenClawPluginApi } from '../openclaw-sdk.js';
-import { Type } from '@sinclair/typebox';
-import { buildPainFlag } from '../core/pain.js';
-import { resolveWorkspaceDirFromApi } from '../core/path-resolver.js';
-import { TrajectoryRegistry } from '../core/trajectory.js';
-import * as fs from 'fs';
-import * as path from 'path';
-import { atomicWriteFileSync } from '../utils/io.js';
-
-// Pain flag contract required fields
-const PAIN_FLAG_REQUIRED_FIELDS = ['source', 'score', 'time', 'reason'] as const;
-
-/**
- * Atomic file write: write to temp file then rename.
- * Prevents corruption if process crashes mid-write.
- */
-function writePainFlagAtomic(filePath: string, content: string): void {
-    const dir = path.dirname(filePath);
-    if (!fs.existsSync(dir)) {
-        fs.mkdirSync(dir, { recursive: true });
-    }
-    atomicWriteFileSync(filePath, content);
-}
-
-/**
- * Creates the `write_pain_flag` tool.
- *
- * This tool allows the agent to record a pain signal when it recognizes
- * that it made a mistake, violated a principle, or needs to flag an issue
- * for later reflection.
- *
- * The tool wraps `buildPainFlag` + atomic `writePainFlag` to ensure:
- * - Correct KV format serialization (never [object Object] corruption)
- * - Atomic writes (temp file + rename, crash-safe)
- * - Full contract compliance (source, score, time, reason)
- *
- * The agent should NEVER write to .pain_flag directly.
- */
-export function createWritePainFlagTool(api: OpenClawPluginApi) {
-    return {
-        name: 'write_pain_flag',
-        description:
-            'Record a pain signal to flag mistakes, principle violations, or issues for later reflection. ' +
-            'Use this tool INSTEAD of writing .pain_flag directly. ' +
-            'Pain signals are processed by the evolution system on the next heartbeat cycle.',
-        parameters: Type.Object({
-            reason: Type.String({
-                description:
-                    'Describe specifically what went wrong. ' +
-                    'Include the error, the violated principle, or the issue. ' +
-                    'Be concrete: "I edited config.ts without reading it first, breaking the export" ' +
-                    'is better than "I made a mistake".',
-            }),
-            score: Type.Optional(Type.Number({
-                description:
-                    'Pain severity score (0-100). Default: 80. ' +
-                    'Guidelines: 30-50 (minor issue), 50-70 (moderate error), ' +
-                    '70-100 (severe principle violation or data loss risk).',
-                minimum: 0,
-                maximum: 100,
-            })),
-            source: Type.Optional(Type.String({
-                description:
-                    'Source of the pain signal. ' +
-                    'Values: manual (user flagged), tool_failure (tool error), ' +
-                    'user_empathy (user frustration), principle_violation (principle broken), ' +
-                    'human_intervention (user manually intervened). ' +
-                    'Default: manual.',
-            })),
-            session_id: Type.Optional(Type.String({
-                description:
-                    'Session ID where the pain occurred. ' +
-                    'If not provided, the system will use the current session.',
-            })),
-            is_risky: Type.Optional(Type.Boolean({
-                description:
-                    'Whether this involves a high-risk operation (e.g., writing to sensitive files). ' +
-                    'Default: false.',
-            })),
-        }),
-
-        async execute(
-            _toolCallId: string,
-            rawParams: Record<string, unknown>
-        ): Promise<{ content: { type: string; text: string }[] }> {
-            const reason = typeof rawParams.reason === 'string' ? rawParams.reason.trim() : '';
-            const score = typeof rawParams.score === 'number' ? Math.max(0, Math.min(100, Math.round(rawParams.score))) : 80;
-            const source = typeof rawParams.source === 'string' && rawParams.source.trim() ? rawParams.source.trim() : 'manual';
-            const sessionId = typeof rawParams.session_id === 'string' ? rawParams.session_id.trim() : '';
-            const isRisky = rawParams.is_risky === true;
-
-            // ── Validate required fields ──
-            if (!reason) {
-                api.logger?.warn?.('[PD:write_pain_flag] Missing required field: reason');
-                return {
-                    content: [{
-                        type: 'text',
-                        text: '❌ Error: The `reason` parameter is required.\n' +
-                            'Describe specifically what went wrong. Example:\n' +
-                            '"I edited config.ts without reading it first, breaking the export"',
-                    }],
-                };
-            }
-
-            // ── Resolve workspace ──
-            const workspaceDir = resolveWorkspaceDirFromApi(api);
-            if (!workspaceDir) {
-                api.logger?.error?.('[PD:write_pain_flag] Cannot resolve workspace directory');
-                return {
-                    content: [{
-                        type: 'text',
-                        text: '❌ Error: Cannot determine the workspace directory. ' +
-                            'Please ensure you are in an active workspace.',
-                    }],
-                };
-            }
-
-            try {
-                // ── Record pain event to trajectory DB first (before flag file) ──
-                // This ensures we have a real AUTOINCREMENT ID that flows through
-                // the entire pain→principle→compile chain (painEventId propagation).
-                let painEventId: number | undefined;
-                try {
-                    const trajectory = TrajectoryRegistry.get(workspaceDir);
-                    painEventId = trajectory.recordPainEvent({
-                        sessionId: sessionId || 'unknown',
-                        source,
-                        score,
-                        reason,
-                        severity: null,
-                        origin: 'manual',
-                        confidence: null,
-                        text: undefined,
-                    });
-                } catch (trajErr) {
-                    // If trajectory write fails, log but continue — the pain flag
-                    // itself is still valid and should be processed. The pain event
-                    // ID simply won't be available for the compiler's exact matching.
-                    api.logger?.warn?.(`[PD:write_pain_flag] Failed to record pain event to trajectory: ${String(trajErr)}`);
-                }
-
-                // ── Build pain flag data (KV format) ──
-                const painData = buildPainFlag({
-                    source,
-                    score: String(score),
-                    reason,
-                    session_id: sessionId,
-                    is_risky: isRisky,
-                    pain_event_id: painEventId !== undefined ? String(painEventId) : undefined,
-                });
-
-                // ── Validate contract compliance ──
-                const missingFields: string[] = [];
-                for (const field of PAIN_FLAG_REQUIRED_FIELDS) {
-                    if (!painData[field] || painData[field].trim() === '') {
-                        missingFields.push(field);
-                    }
-                }
-                if (missingFields.length > 0) {
-                    api.logger?.error?.(`[PD:write_pain_flag] Pain flag missing required fields: ${missingFields.join(', ')}`);
-                    return {
-                        content: [{
-                            type: 'text',
-                            text: `❌ Error: Pain flag is missing required fields: ${missingFields.join(', ')}. ` +
-                                'This is an internal error — please report it.',
-                        }],
-                    };
-                }
-
-                // ── Atomic write (temp file + rename) ──
-                const painFlagPath = path.join(workspaceDir, '.state', '.pain_flag');
-                const { serializeKvLines } = await import('../utils/io.js');
-                const content = serializeKvLines(painData);
-                writePainFlagAtomic(painFlagPath, content);
-
-                // ── Log success ──
-                api.logger?.info?.(
-                    `[PD:write_pain_flag] Pain signal recorded: source=${source}, score=${score}, ` +
-                    `reason="${reason.slice(0, 80)}"${reason.length > 80 ? '...' : ''}"`
-                );
-
-                // ── Agent feedback ──
-                return {
-                    content: [{
-                        type: 'text',
-                        text: `✅ Pain signal recorded successfully.\n\n` +
-                            `- **Reason**: ${reason}\n` +
-                            `- **Score**: ${score}/100\n` +
-                            `- **Source**: ${source}\n` +
-                            `- **Risk**: ${isRisky ? 'Yes' : 'No'}\n` +
-                            `- **Session**: ${sessionId || '(current)'}\n\n` +
-                            `The evolution system will process this signal on the next heartbeat cycle ` +
-                            `(typically within 60 seconds).`,
-                    }],
-                };
-            } catch (err) {
-                // ── Log failure with stack trace ──
-                const errorMsg = err instanceof Error ? err.message : String(err);
-                const stack = err instanceof Error ? err.stack?.split('\n').slice(0, 3).join(' → ') : '';
-                api.logger?.error?.(
-                    `[PD:write_pain_flag] Failed to write pain flag: ${errorMsg}` +
-                    (stack ? `\n  Stack: ${stack}` : '')
-                );
-
-                return {
-                    content: [{
-                        type: 'text',
-                        text: `❌ Failed to record pain signal: ${errorMsg}\n\n` +
-                            'The error has been logged. Please try again or report this issue.',
-                    }],
-                };
-            }
-        },
-    };
-}

package/tests/commands/nocturnal-review.test.ts

@@ -1,448 +0,0 @@
-import { describe, it, expect, beforeEach, afterEach } from 'vitest';
-import * as fs from 'fs';
-import * as path from 'path';
-import * as os from 'os';
-import { handleNocturnalReviewCommand } from '../../src/commands/nocturnal-review.js';
-import type { PluginCommandContext } from '../../src/openclaw-sdk.js';
-import {
-  registerSample,
-  updateReviewStatus,
-  getDatasetRecord,
-} from '../../src/core/nocturnal-dataset.js';
-import type { NocturnalArtifact } from '../../src/core/nocturnal-arbiter.js';
-
-// ---------------------------------------------------------------------------
-// Fixtures
-// ---------------------------------------------------------------------------
-
-function makeArtifact(overrides: Partial<NocturnalArtifact> = {}): NocturnalArtifact {
-  return {
-    artifactId: 'a1b2c3d4-e5f6-7890-abcd-ef1234567890',
-    sessionId: 'session-abc123',
-    principleId: 'T-08',
-    sourceSnapshotRef: 'snapshot-2026-03-27-001',
-    badDecision: 'After bash command failed, immediately retried without diagnosing',
-    betterDecision: 'Check the error message before retrying',
-    rationale: 'Diagnosing failures prevents repeated failures and respects cost of each attempt',
-    createdAt: '2026-03-27T12:00:00.000Z',
-    ...overrides,
-  };
-}
-
-function makeTmpDir(): string {
-  return fs.mkdtempSync(path.join(os.tmpdir(), 'pd-nocturnal-review-test-'));
-}
-
-function rmdir(dir: string): void {
-  try {
-    if (fs.existsSync(dir)) {
-      fs.rmSync(dir, { recursive: true, force: true });
-    }
-  } catch {
-    // Ignore
-  }
-}
-
-function makeCtx(workspaceDir: string, args = ''): PluginCommandContext {
-  return {
-    config: { workspaceDir, language: 'en' },
-    args,
-  } as unknown as PluginCommandContext;
-}
-
-function setupSample(
-  workspaceDir: string,
-  artifactId: string,
-  family: string | null = 'gpt-4'
-): { fingerprint: string; artifactPath: string } {
-  const artifact = makeArtifact({ artifactId });
-  const artifactPath = path.join(
-    workspaceDir,
-    '.state',
-    'nocturnal',
-    'samples',
-    `${artifactId}.json`
-  );
-  fs.mkdirSync(path.dirname(artifactPath), { recursive: true });
-  fs.writeFileSync(artifactPath, JSON.stringify({ ...artifact, status: 'approved' }), 'utf-8');
-  const result = registerSample(workspaceDir, artifact, artifactPath, family);
-  return { fingerprint: result.record.sampleFingerprint, artifactPath };
-}
-
-// ---------------------------------------------------------------------------
-// Tests: list
-// ---------------------------------------------------------------------------
-
-describe('NocturnalReviewCommand list', () => {
-  let tmpDir: string;
-
-  beforeEach(() => {
-    tmpDir = makeTmpDir();
-  });
-
-  afterEach(() => {
-    rmdir(tmpDir);
-  });
-
-  it('shows pending samples', () => {
-    const { fingerprint } = setupSample(tmpDir, 'art-pending-list');
-
-    const result = handleNocturnalReviewCommand(makeCtx(tmpDir, 'list'));
-
-    expect(result.text).toContain('pending_review');
-    expect(result.text).toContain(fingerprint.substring(0, 16));
-  });
-
-  it('shows empty when no pending samples', () => {
-    const { fingerprint } = setupSample(tmpDir, 'art-approved-list');
-    updateReviewStatus(tmpDir, fingerprint, 'approved_for_training', 'Approved');
-
-    const result = handleNocturnalReviewCommand(makeCtx(tmpDir, 'list'));
-
-    expect(result.text).toContain('No pending');
-  });
-});
-
-// ---------------------------------------------------------------------------
-// Tests: approve
-// ---------------------------------------------------------------------------
-
-describe('NocturnalReviewCommand approve', () => {
-  let tmpDir: string;
-
-  beforeEach(() => {
-    tmpDir = makeTmpDir();
-  });
-
-  afterEach(() => {
-    rmdir(tmpDir);
-  });
-
-  it('approves a pending sample', () => {
-    const { fingerprint } = setupSample(tmpDir, 'art-approve-test');
-
-    const result = handleNocturnalReviewCommand(
-      makeCtx(tmpDir, `approve ${fingerprint} Looks good for training`)
-    );
-
-    expect(result.text).toContain('approved for training');
-    const record = getDatasetRecord(tmpDir, fingerprint);
-    expect(record?.reviewStatus).toBe('approved_for_training');
-    expect(record?.reviewReason).toBe('Looks good for training');
-  });
-
-  it('approves with default reason if not provided', () => {
-    const { fingerprint } = setupSample(tmpDir, 'art-approve-default');
-
-    const result = handleNocturnalReviewCommand(
-      makeCtx(tmpDir, `approve ${fingerprint}`)
-    );
-
-    expect(result.text).toContain('approved for training');
-    const record = getDatasetRecord(tmpDir, fingerprint);
-    expect(record?.reviewStatus).toBe('approved_for_training');
-    expect(record?.reviewReason).toBeTruthy();
-  });
-
-  it('rejects already approved sample', () => {
-    const { fingerprint } = setupSample(tmpDir, 'art-already-approved');
-    updateReviewStatus(tmpDir, fingerprint, 'approved_for_training', 'Already approved');
-
-    const result = handleNocturnalReviewCommand(
-      makeCtx(tmpDir, `approve ${fingerprint}`)
-    );
-
-    expect(result.text).toContain('already approved');
-  });
-
-  it('rejects rejected sample without reset', () => {
-    const { fingerprint } = setupSample(tmpDir, 'art-already-rejected');
-    updateReviewStatus(tmpDir, fingerprint, 'rejected', 'Rejected');
-
-    const result = handleNocturnalReviewCommand(
-      makeCtx(tmpDir, `approve ${fingerprint}`)
-    );
-
-    expect(result.text).toContain('rejected');
-  });
-
-  it('rejects superseded sample', () => {
-    const { fingerprint } = setupSample(tmpDir, 'art-superseded');
-    updateReviewStatus(tmpDir, fingerprint, 'approved_for_training', 'Approved');
-    updateReviewStatus(tmpDir, fingerprint, 'superseded', 'Superseded by better');
-
-    const result = handleNocturnalReviewCommand(
-      makeCtx(tmpDir, `approve ${fingerprint}`)
-    );
-
-    expect(result.text).toContain('superseded');
-  });
-
-  it('returns error for unknown fingerprint', () => {
-    const result = handleNocturnalReviewCommand(
-      makeCtx(tmpDir, 'approve unknown-fingerprint')
-    );
-
-    expect(result.text).toContain('not found');
-  });
-
-  it('returns error when missing fingerprint', () => {
-    const result = handleNocturnalReviewCommand(makeCtx(tmpDir, 'approve'));
-
-    expect(result.text).toContain('Usage');
-  });
-});
-
-// ---------------------------------------------------------------------------
-// Tests: reject
-// ---------------------------------------------------------------------------
-
-describe('NocturnalReviewCommand reject', () => {
-  let tmpDir: string;
-
-  beforeEach(() => {
-    tmpDir = makeTmpDir();
-  });
-
-  afterEach(() => {
-    rmdir(tmpDir);
-  });
-
-  it('rejects a pending sample', () => {
-    const { fingerprint } = setupSample(tmpDir, 'art-reject-test');
-
-    const result = handleNocturnalReviewCommand(
-      makeCtx(tmpDir, `reject ${fingerprint} Not suitable for training`)
-    );
-
-    expect(result.text).toContain('rejected');
-    const record = getDatasetRecord(tmpDir, fingerprint);
-    expect(record?.reviewStatus).toBe('rejected');
-    expect(record?.reviewReason).toBe('Not suitable for training');
-  });
-
-  it('rejects with default reason if not provided', () => {
-    const { fingerprint } = setupSample(tmpDir, 'art-reject-default');
-
-    const result = handleNocturnalReviewCommand(
-      makeCtx(tmpDir, `reject ${fingerprint}`)
-    );
-
-    expect(result.text).toContain('rejected');
-    const record = getDatasetRecord(tmpDir, fingerprint);
-    expect(record?.reviewStatus).toBe('rejected');
-  });
-
-  it('rejects already rejected sample', () => {
-    const { fingerprint } = setupSample(tmpDir, 'art-already-rejected');
-    updateReviewStatus(tmpDir, fingerprint, 'rejected', 'Already rejected');
-
-    const result = handleNocturnalReviewCommand(
-      makeCtx(tmpDir, `reject ${fingerprint}`)
-    );
-
-    expect(result.text).toContain('already rejected');
-  });
-
-  it('rejects superseded sample', () => {
-    const { fingerprint } = setupSample(tmpDir, 'art-superseded-reject');
-    updateReviewStatus(tmpDir, fingerprint, 'approved_for_training', 'Approved');
-    updateReviewStatus(tmpDir, fingerprint, 'superseded', 'Superseded');
-
-    const result = handleNocturnalReviewCommand(
-      makeCtx(tmpDir, `reject ${fingerprint}`)
-    );
-
-    expect(result.text).toContain('superseded');
-  });
-
-  it('returns error for unknown fingerprint', () => {
-    const result = handleNocturnalReviewCommand(
-      makeCtx(tmpDir, 'reject unknown-fingerprint')
-    );
-
-    expect(result.text).toContain('not found');
-  });
-});
-
-// ---------------------------------------------------------------------------
-// Tests: show
-// ---------------------------------------------------------------------------
-
-describe('NocturnalReviewCommand show', () => {
-  let tmpDir: string;
-
-  beforeEach(() => {
-    tmpDir = makeTmpDir();
-  });
-
-  afterEach(() => {
-    rmdir(tmpDir);
-  });
-
-  it('shows sample details', () => {
-    const { fingerprint } = setupSample(tmpDir, 'art-show-test');
-
-    const result = handleNocturnalReviewCommand(
-      makeCtx(tmpDir, `show ${fingerprint}`)
-    );
-
-    expect(result.text).toContain(fingerprint.substring(0, 16));
-    expect(result.text).toContain('T-08');
-    expect(result.text).toContain('session-abc123');
-  });
-
-  it('returns error for unknown fingerprint', () => {
-    const result = handleNocturnalReviewCommand(
-      makeCtx(tmpDir, 'show unknown-fingerprint')
-    );
-
-    expect(result.text).toContain('not found');
-  });
-
-  it('returns error when missing fingerprint', () => {
-    const result = handleNocturnalReviewCommand(makeCtx(tmpDir, 'show'));
-
-    expect(result.text).toContain('Usage');
-  });
-});
-
-// ---------------------------------------------------------------------------
-// Tests: set-family
-// ---------------------------------------------------------------------------
-
-describe('NocturnalReviewCommand set-family', () => {
-  let tmpDir: string;
-
-  beforeEach(() => {
-    tmpDir = makeTmpDir();
-  });
-
-  afterEach(() => {
-    rmdir(tmpDir);
-  });
-
-  it('sets target model family', () => {
-    const { fingerprint } = setupSample(tmpDir, 'art-family-test', null);
-
-    const result = handleNocturnalReviewCommand(
-      makeCtx(tmpDir, `set-family ${fingerprint} claude-3`)
-    );
-
-    expect(result.text).toContain('claude-3');
-    const record = getDatasetRecord(tmpDir, fingerprint);
-    expect(record?.targetModelFamily).toBe('claude-3');
-  });
-
-  it('updates existing family', () => {
-    const { fingerprint } = setupSample(tmpDir, 'art-family-update-test', 'gpt-4');
-
-    const result = handleNocturnalReviewCommand(
-      makeCtx(tmpDir, `set-family ${fingerprint} gpt-4o`)
-    );
-
-    expect(result.text).toContain('gpt-4o');
-    const record = getDatasetRecord(tmpDir, fingerprint);
-    expect(record?.targetModelFamily).toBe('gpt-4o');
-  });
-
-  it('returns error for unknown fingerprint', () => {
-    const result = handleNocturnalReviewCommand(
-      makeCtx(tmpDir, 'set-family unknown-fingerprint gpt-4')
-    );
-
-    expect(result.text).toContain('not found');
-  });
-
-  it('returns error when missing args', () => {
-    const { fingerprint } = setupSample(tmpDir, 'art-family-missing-test');
-
-    const result = handleNocturnalReviewCommand(
-      makeCtx(tmpDir, `set-family ${fingerprint}`)
-    );
-
-    expect(result.text).toContain('Usage');
-  });
-});
-
-// ---------------------------------------------------------------------------
-// Tests: stats
-// ---------------------------------------------------------------------------
-
-describe('NocturnalReviewCommand stats', () => {
-  let tmpDir: string;
-
-  beforeEach(() => {
-    tmpDir = makeTmpDir();
-  });
-
-  afterEach(() => {
-    rmdir(tmpDir);
-  });
-
-  it('shows dataset statistics', () => {
-    setupSample(tmpDir, 'art-stats-1', 'gpt-4');
-    const { fingerprint: fp2 } = setupSample(tmpDir, 'art-stats-2', 'gpt-4');
-    updateReviewStatus(tmpDir, fp2, 'approved_for_training', 'Approved');
-
-    const result = handleNocturnalReviewCommand(makeCtx(tmpDir, 'stats'));
-
-    expect(result.text).toContain('Total');
-    expect(result.text).toContain('2');
-  });
-
-  it('shows export-ready counts by family', () => {
-    const { fingerprint: fp1 } = setupSample(tmpDir, 'art-stats-family-1', 'gpt-4');
-    updateReviewStatus(tmpDir, fp1, 'approved_for_training', 'Approved');
-
-    const result = handleNocturnalReviewCommand(makeCtx(tmpDir, 'stats'));
-
-    expect(result.text).toContain('gpt-4');
-    expect(result.text).toContain('1');
-  });
-});
-
-// ---------------------------------------------------------------------------
-// Integration: full review flow
-// ---------------------------------------------------------------------------
-
-describe('NocturnalReviewCommand full review flow', () => {
-  let tmpDir: string;
-
-  beforeEach(() => {
-    tmpDir = makeTmpDir();
-  });
-
-  afterEach(() => {
-    rmdir(tmpDir);
-  });
-
-  it('full flow: list → set-family → approve → export-ready', () => {
-    // 1. Register sample without family
-    const { fingerprint } = setupSample(tmpDir, 'art-full-flow', null);
-
-    // 2. Verify it's pending
-    let result = handleNocturnalReviewCommand(makeCtx(tmpDir, 'list'));
-    expect(result.text).toContain(fingerprint.substring(0, 16));
-
-    // 3. Set family
-    result = handleNocturnalReviewCommand(
-      makeCtx(tmpDir, `set-family ${fingerprint} gpt-4`)
-    );
-    expect(result.text).toContain('gpt-4');
-
-    // 4. Approve
-    result = handleNocturnalReviewCommand(
-      makeCtx(tmpDir, `approve ${fingerprint} Good sample`)
-    );
-    expect(result.text).toContain('approved for training');
-
-    // 5. Verify stats show approved
-    result = handleNocturnalReviewCommand(makeCtx(tmpDir, 'stats'));
-    expect(result.text).toContain('Approved for training');
-
-    // 6. Verify not in pending list anymore
-    result = handleNocturnalReviewCommand(makeCtx(tmpDir, 'list'));
-    expect(result.text).not.toContain(fingerprint.substring(0, 16));
-  });
-});