codekin 0.4.0 → 0.5.0

package/server/dist/orchestrator-learning.d.ts

@@ -0,0 +1,134 @@
+/**
+ * Orchestrator self-improving memory — extraction, deduplication, aging,
+ * pattern learning, and user skill modeling.
+ *
+ * This module adds the intelligence layer on top of OrchestratorMemory:
+ * - Extracts memory candidates from session interactions
+ * - Deduplicates against existing memories using FTS similarity
+ * - Ages and decays stale items on a schedule
+ * - Tracks finding outcomes to improve future triage
+ * - Maintains a user skill model that adapts guidance
+ * - Records decisions with outcome tracking
+ */
+import { OrchestratorMemory, type MemoryType, type MemoryItem } from './orchestrator-memory.js';
+/** A candidate extracted from a session interaction. */
+export interface MemoryCandidate {
+    memoryType: MemoryType;
+    title: string;
+    content: string;
+    scope: string | null;
+    tags: string[];
+    confidence: number;
+}
+/** Finding outcome — tracks what happened when a finding was acted on or skipped. */
+export interface FindingOutcome {
+    findingId: string;
+    repo: string;
+    category: string;
+    severity: string;
+    action: 'implemented' | 'skipped' | 'deferred';
+    reason: string;
+    sessionId: string | null;
+    outcome: 'success' | 'failure' | 'pending' | null;
+    timestamp: string;
+}
+/** User skill assessment per domain. */
+export interface SkillLevel {
+    domain: string;
+    level: 'beginner' | 'intermediate' | 'advanced' | 'expert';
+    confidence: number;
+    signals: string[];
+    lastUpdated: string;
+}
+/** A decision record with outcome tracking. */
+export interface DecisionRecord {
+    id: string;
+    decision: string;
+    rationale: string;
+    repo: string | null;
+    relatedFinding: string | null;
+    expectedOutcome: string;
+    actualOutcome: string | null;
+    outcomeAssessedAt: string | null;
+    timestamp: string;
+}
+/**
+ * Extract memory candidates from an interaction transcript.
+ *
+ * This is a rule-based extractor that looks for patterns indicating
+ * durable information worth remembering. In future, this could be
+ * enhanced with an LLM call for richer extraction.
+ */
+export declare function extractMemoryCandidates(userMessage: string, assistantResponse: string, currentRepo: string | null): MemoryCandidate[];
+/**
+ * Check if a candidate is a duplicate of an existing memory item.
+ * Uses FTS search to find similar items, then compares content overlap.
+ */
+export declare function findDuplicate(memory: OrchestratorMemory, candidate: MemoryCandidate, threshold?: number): MemoryItem | null;
+/**
+ * Smart upsert: insert candidate if no duplicate exists, update if similar.
+ * Returns the memory item ID and whether it was new or updated.
+ */
+export declare function smartUpsert(memory: OrchestratorMemory, candidate: MemoryCandidate, sourceRef?: string | null): {
+    id: string;
+    action: 'inserted' | 'updated' | 'skipped';
+};
+/**
+ * Run the aging/decay cycle:
+ * 1. Expire items past their TTL
+ * 2. Compact old journal entries into monthly summaries
+ * 3. Decay confidence of items that haven't been accessed recently
+ */
+export declare function runAgingCycle(memory: OrchestratorMemory): {
+    expired: number;
+    compacted: number;
+    decayed: number;
+};
+/**
+ * Record the outcome of a finding triage decision.
+ */
+export declare function recordFindingOutcome(memory: OrchestratorMemory, outcome: FindingOutcome): string;
+/**
+ * Analyze past finding outcomes to compute a triage recommendation.
+ * Returns the likely action and confidence based on historical patterns.
+ */
+export declare function getTriageRecommendation(memory: OrchestratorMemory, category: string, severity: string, repo: string | null): {
+    action: 'implement' | 'skip' | 'defer' | 'unknown';
+    confidence: number;
+    reason: string;
+};
+/** Load the user's skill profile from disk. */
+export declare function loadSkillProfile(): SkillLevel[];
+/** Save the skill profile to disk. */
+export declare function saveSkillProfile(profile: SkillLevel[]): void;
+/**
+ * Update the user's skill level for a domain based on observed signals.
+ * Signals are things like: "used advanced git rebase", "asked basic TypeScript question",
+ * "configured CI pipeline without help".
+ */
+export declare function updateSkillLevel(domain: string, signal: string, indicatedLevel: 'beginner' | 'intermediate' | 'advanced' | 'expert'): SkillLevel;
+/**
+ * Get the user's assessed skill level for a domain.
+ * Returns null if we have no data for this domain.
+ */
+export declare function getSkillLevel(domain: string): SkillLevel | null;
+/**
+ * Get a guidance style recommendation based on the user's overall skill profile.
+ */
+export declare function getGuidanceStyle(): {
+    tone: 'tutorial' | 'collaborative' | 'concise';
+    explainLevel: 'detailed' | 'moderate' | 'minimal';
+    domains: Record<string, SkillLevel['level']>;
+};
+/**
+ * Record a decision with expected outcome.
+ */
+export declare function recordDecision(memory: OrchestratorMemory, decision: Omit<DecisionRecord, 'id' | 'timestamp' | 'actualOutcome' | 'outcomeAssessedAt'>): string;
+/**
+ * Update a decision's actual outcome.
+ */
+export declare function assessDecisionOutcome(memory: OrchestratorMemory, decisionId: string, actualOutcome: string): boolean;
+/**
+ * Get decisions that need outcome assessment (older than 7 days, no outcome yet).
+ */
+export declare function getPendingOutcomeAssessments(memory: OrchestratorMemory): DecisionRecord[];

package/server/dist/orchestrator-learning.js

@@ -0,0 +1,567 @@
+/**
+ * Orchestrator self-improving memory — extraction, deduplication, aging,
+ * pattern learning, and user skill modeling.
+ *
+ * This module adds the intelligence layer on top of OrchestratorMemory:
+ * - Extracts memory candidates from session interactions
+ * - Deduplicates against existing memories using FTS similarity
+ * - Ages and decays stale items on a schedule
+ * - Tracks finding outcomes to improve future triage
+ * - Maintains a user skill model that adapts guidance
+ * - Records decisions with outcome tracking
+ */
+import { readFileSync, writeFileSync, existsSync, readdirSync } from 'fs';
+import { join } from 'path';
+import { ORCHESTRATOR_DIR } from './orchestrator-manager.js';
+// ---------------------------------------------------------------------------
+// Default TTLs (in days)
+// ---------------------------------------------------------------------------
+const TTL_DAYS = {
+    user_preference: null, // permanent (pinned)
+    repo_context: null, // permanent until repo removed
+    decision: 90,
+    finding_outcome: 180,
+    session_summary: 60,
+    journal: 30,
+};
+// ---------------------------------------------------------------------------
+// Memory Extraction
+// ---------------------------------------------------------------------------
+/**
+ * Extract memory candidates from an interaction transcript.
+ *
+ * This is a rule-based extractor that looks for patterns indicating
+ * durable information worth remembering. In future, this could be
+ * enhanced with an LLM call for richer extraction.
+ */
+export function extractMemoryCandidates(userMessage, assistantResponse, currentRepo) {
+    const candidates = [];
+    // Pattern 1: User states a preference ("I prefer...", "always use...", "don't...")
+    const prefPatterns = [
+        /\bi\s+prefer\b/i,
+        /\balways\s+(use|do|run|make)\b/i,
+        /\bdon'?t\s+(ever|always)\b/i,
+        /\bnever\s+(do|use|run|make)\b/i,
+        /\bplease\s+always\b/i,
+        /\bmy\s+preference\s+is\b/i,
+    ];
+    for (const pat of prefPatterns) {
+        if (pat.test(userMessage)) {
+            candidates.push({
+                memoryType: 'user_preference',
+                title: 'User preference',
+                content: userMessage.slice(0, 500),
+                scope: null,
+                tags: ['preference', 'user-stated'],
+                confidence: 0.9,
+            });
+            break;
+        }
+    }
+    // Pattern 2: User describes their role or expertise
+    const rolePatterns = [
+        /\bi(?:'m| am)\s+a\s+(senior|junior|lead|staff|principal|intern)?\s*\w*/i,
+        /\bmy\s+role\s+is\b/i,
+        /\bi\s+work\s+(?:as|on|in|with)\b/i,
+        /\bi(?:'ve| have)\s+been\s+(?:doing|working|coding|programming)\b/i,
+        /\bfirst\s+time\s+(?:using|working|touching|doing)\b/i,
+        /\bnew\s+to\s+(?:this|react|go|python|typescript|rust|devops)\b/i,
+    ];
+    for (const pat of rolePatterns) {
+        if (pat.test(userMessage)) {
+            candidates.push({
+                memoryType: 'user_preference',
+                title: 'User background',
+                content: userMessage.slice(0, 500),
+                scope: null,
+                tags: ['background', 'skill-signal'],
+                confidence: 0.7,
+            });
+            break;
+        }
+    }
+    // Pattern 3: Explicit "remember" requests
+    if (/\bremember\s+(?:that|this|to)\b/i.test(userMessage)) {
+        candidates.push({
+            memoryType: 'user_preference',
+            title: 'User asked to remember',
+            content: userMessage.slice(0, 500),
+            scope: currentRepo,
+            tags: ['explicit-remember'],
+            confidence: 1.0,
+        });
+    }
+    // Pattern 4: Decision made ("let's go with...", "decided to...")
+    const decisionPatterns = [
+        /\blet'?s?\s+(?:go\s+with|use|do|pick|choose)\b/i,
+        /\bdecided?\s+(?:to|on|that)\b/i,
+        /\bwe(?:'ll| will)\s+(?:go|use|do)\b/i,
+        /\bsounds?\s+good\b/i,
+        /\byes,?\s+(?:go\s+ahead|do\s+it|proceed|let's)\b/i,
+    ];
+    for (const pat of decisionPatterns) {
+        if (pat.test(userMessage)) {
+            candidates.push({
+                memoryType: 'decision',
+                title: 'Decision',
+                content: `User: ${userMessage.slice(0, 250)}\nContext: ${assistantResponse.slice(0, 250)}`,
+                scope: currentRepo,
+                tags: ['decision'],
+                confidence: 0.7,
+            });
+            break;
+        }
+    }
+    // Pattern 5: Repo-specific context from assistant responses
+    if (currentRepo && assistantResponse.length > 200) {
+        const repoSignals = [
+            /\bthis\s+(?:repo|repository|project|codebase)\s+(?:is|uses|has|was)\b/i,
+            /\bthe\s+(?:main|primary)\s+(?:stack|framework|language)\b/i,
+            /\barchitecture\s+(?:is|uses|follows)\b/i,
+        ];
+        for (const pat of repoSignals) {
+            if (pat.test(assistantResponse)) {
+                candidates.push({
+                    memoryType: 'repo_context',
+                    title: `Repo context: ${currentRepo.split('/').pop()}`,
+                    content: assistantResponse.slice(0, 500),
+                    scope: currentRepo,
+                    tags: ['repo-context', 'auto-extracted'],
+                    confidence: 0.6,
+                });
+                break;
+            }
+        }
+    }
+    return candidates;
+}
+// ---------------------------------------------------------------------------
+// Deduplication
+// ---------------------------------------------------------------------------
+/**
+ * Check if a candidate is a duplicate of an existing memory item.
+ * Uses FTS search to find similar items, then compares content overlap.
+ */
+export function findDuplicate(memory, candidate, threshold = 0.85) {
+    // Search for similar items by title/content keywords
+    const searchTerms = (candidate.title + ' ' + candidate.content)
+        .replace(/[^\w\s]/g, ' ')
+        .split(/\s+/)
+        .filter(w => w.length > 3)
+        .slice(0, 5)
+        .join(' ');
+    if (!searchTerms)
+        return null;
+    try {
+        const similar = memory.search(searchTerms, 5);
+        for (const item of similar) {
+            // Same type and scope
+            if (item.memoryType !== candidate.memoryType)
+                continue;
+            if (item.scope !== candidate.scope)
+                continue;
+            // Simple content overlap check
+            const similarity = computeOverlap(item.content, candidate.content);
+            if (similarity >= threshold)
+                return item;
+        }
+    }
+    catch {
+        // FTS query may fail on unusual input — not a problem
+    }
+    return null;
+}
+/** Compute word-level Jaccard similarity between two texts. */
+function computeOverlap(a, b) {
+    const wordsA = new Set(a.toLowerCase().split(/\s+/).filter(w => w.length > 2));
+    const wordsB = new Set(b.toLowerCase().split(/\s+/).filter(w => w.length > 2));
+    if (wordsA.size === 0 || wordsB.size === 0)
+        return 0;
+    let intersection = 0;
+    for (const w of wordsA) {
+        if (wordsB.has(w))
+            intersection++;
+    }
+    return intersection / (wordsA.size + wordsB.size - intersection);
+}
+/**
+ * Smart upsert: insert candidate if no duplicate exists, update if similar.
+ * Returns the memory item ID and whether it was new or updated.
+ */
+export function smartUpsert(memory, candidate, sourceRef = null) {
+    const duplicate = findDuplicate(memory, candidate);
+    if (duplicate) {
+        // If the new candidate has higher confidence, update
+        if (candidate.confidence > duplicate.confidence) {
+            const id = memory.upsert({
+                id: duplicate.id,
+                ...candidate,
+                sourceRef,
+                isPinned: duplicate.isPinned,
+                expiresAt: duplicate.expiresAt,
+            });
+            return { id, action: 'updated' };
+        }
+        return { id: duplicate.id, action: 'skipped' };
+    }
+    // Compute expiry based on type
+    const ttlDays = TTL_DAYS[candidate.memoryType];
+    const expiresAt = ttlDays
+        ? new Date(Date.now() + ttlDays * 24 * 60 * 60 * 1000).toISOString()
+        : null;
+    const id = memory.upsert({
+        ...candidate,
+        sourceRef,
+        isPinned: candidate.memoryType === 'user_preference',
+        expiresAt,
+    });
+    return { id, action: 'inserted' };
+}
+// ---------------------------------------------------------------------------
+// Aging & Decay
+// ---------------------------------------------------------------------------
+/**
+ * Run the aging/decay cycle:
+ * 1. Expire items past their TTL
+ * 2. Compact old journal entries into monthly summaries
+ * 3. Decay confidence of items that haven't been accessed recently
+ */
+export function runAgingCycle(memory) {
+    // 1. Expire stale items
+    const expired = memory.expireStale();
+    // 2. Compact old journal entries (older than 30 days) into monthly summaries
+    const compacted = compactOldJournals();
+    // 3. Decay confidence of old, non-pinned items that haven't been updated
+    const thirtyDaysAgo = new Date(Date.now() - 30 * 24 * 60 * 60 * 1000).toISOString();
+    const oldItems = memory.list({ limit: 100 })
+        .filter(item => !item.isPinned &&
+        item.updatedAt < thirtyDaysAgo &&
+        item.confidence > 0.3);
+    let decayed = 0;
+    for (const item of oldItems) {
+        const newConfidence = Math.max(0.3, item.confidence * 0.95); // 5% decay
+        if (newConfidence < item.confidence) {
+            memory.upsert({
+                id: item.id,
+                memoryType: item.memoryType,
+                scope: item.scope,
+                title: item.title,
+                content: item.content,
+                sourceRef: item.sourceRef,
+                confidence: newConfidence,
+                expiresAt: item.expiresAt,
+                isPinned: item.isPinned,
+                tags: item.tags,
+            });
+            decayed++;
+        }
+    }
+    return { expired, compacted, decayed };
+}
+/** Compact journal entries older than 30 days into monthly summary files. */
+function compactOldJournals() {
+    const journalDir = join(ORCHESTRATOR_DIR, 'journal');
+    if (!existsSync(journalDir))
+        return 0;
+    const thirtyDaysAgo = new Date(Date.now() - 30 * 24 * 60 * 60 * 1000);
+    const cutoffDate = thirtyDaysAgo.toISOString().slice(0, 10);
+    let files;
+    try {
+        files = readdirSync(journalDir).filter(f => f.endsWith('.md'));
+    }
+    catch {
+        return 0;
+    }
+    // Group old files by month
+    const monthGroups = new Map();
+    for (const file of files) {
+        const dateMatch = file.match(/^(\d{4}-\d{2})-\d{2}\.md$/);
+        if (!dateMatch)
+            continue;
+        const fileDate = file.replace('.md', '');
+        if (fileDate >= cutoffDate)
+            continue; // not old enough
+        const month = dateMatch[1];
+        const group = monthGroups.get(month) ?? [];
+        group.push(file);
+        monthGroups.set(month, group);
+    }
+    // Compact each month into a summary file
+    let compacted = 0;
+    for (const [month, monthFiles] of monthGroups) {
+        const summaryFile = join(journalDir, `${month}-summary.md`);
+        if (existsSync(summaryFile))
+            continue; // already compacted
+        const contents = [`# Journal Summary: ${month}\n`];
+        for (const file of monthFiles.sort()) {
+            const filePath = join(journalDir, file);
+            try {
+                const content = readFileSync(filePath, 'utf-8');
+                contents.push(`## ${file.replace('.md', '')}\n${content}\n`);
+            }
+            catch {
+                continue;
+            }
+        }
+        writeFileSync(summaryFile, contents.join('\n'), 'utf-8');
+        compacted += monthFiles.length;
+        // Note: we don't delete originals — the orchestrator can do that if it wants
+    }
+    return compacted;
+}
+// ---------------------------------------------------------------------------
+// Pattern Learning — Finding Outcomes
+// ---------------------------------------------------------------------------
+/**
+ * Record the outcome of a finding triage decision.
+ */
+export function recordFindingOutcome(memory, outcome) {
+    return memory.upsert({
+        memoryType: 'finding_outcome',
+        title: `${outcome.action}: ${outcome.category} finding in ${outcome.repo.split('/').pop()}`,
+        content: JSON.stringify(outcome),
+        scope: outcome.repo,
+        sourceRef: outcome.sessionId,
+        confidence: 0.9,
+        isPinned: false,
+        expiresAt: new Date(Date.now() + 180 * 24 * 60 * 60 * 1000).toISOString(),
+        tags: [outcome.category, outcome.severity, outcome.action],
+    });
+}
+/**
+ * Analyze past finding outcomes to compute a triage recommendation.
+ * Returns the likely action and confidence based on historical patterns.
+ */
+export function getTriageRecommendation(memory, category, severity, repo) {
+    // Get all finding outcomes for this category, prefer repo-specific matches
+    const allOutcomes = memory.list({ memoryType: 'finding_outcome', limit: 50 })
+        .map(item => {
+        try {
+            return JSON.parse(item.content);
+        }
+        catch {
+            return null;
+        }
+    })
+        .filter((o) => o !== null)
+        .filter(o => o.category === category);
+    // Prefer repo-specific outcomes if available
+    const repoSpecific = repo ? allOutcomes.filter(o => o.repo === repo) : [];
+    const outcomes = repoSpecific.length >= 2 ? repoSpecific : allOutcomes;
+    if (outcomes.length < 3) {
+        return { action: 'unknown', confidence: 0, reason: 'Not enough historical data' };
+    }
+    // Filter to matching severity
+    const bySeverity = outcomes.filter(o => o.severity === severity);
+    const pool = bySeverity.length >= 2 ? bySeverity : outcomes;
+    // Count actions
+    const counts = { implemented: 0, skipped: 0, deferred: 0 };
+    for (const o of pool) {
+        counts[o.action]++;
+    }
+    const total = pool.length;
+    const implementRate = counts.implemented / total;
+    const skipRate = counts.skipped / total;
+    // Also factor in success rate of implementations
+    const implementations = pool.filter(o => o.action === 'implemented');
+    const successRate = implementations.length > 0
+        ? implementations.filter(o => o.outcome === 'success').length / implementations.length
+        : 0;
+    if (implementRate > 0.6 && successRate > 0.5) {
+        return {
+            action: 'implement',
+            confidence: Math.min(0.9, implementRate * successRate),
+            reason: `${Math.round(implementRate * 100)}% of similar findings were implemented (${Math.round(successRate * 100)}% success rate)`,
+        };
+    }
+    if (skipRate > 0.6) {
+        return {
+            action: 'skip',
+            confidence: Math.min(0.85, skipRate),
+            reason: `${Math.round(skipRate * 100)}% of similar findings were skipped`,
+        };
+    }
+    return {
+        action: 'unknown',
+        confidence: 0.3,
+        reason: `Mixed history: ${counts.implemented} implemented, ${counts.skipped} skipped, ${counts.deferred} deferred`,
+    };
+}
+// ---------------------------------------------------------------------------
+// User Skill Model
+// ---------------------------------------------------------------------------
+const SKILL_PROFILE_FILE = join(ORCHESTRATOR_DIR, 'skill-profile.json');
+/** Load the user's skill profile from disk. */
+export function loadSkillProfile() {
+    if (!existsSync(SKILL_PROFILE_FILE))
+        return [];
+    try {
+        return JSON.parse(readFileSync(SKILL_PROFILE_FILE, 'utf-8'));
+    }
+    catch {
+        return [];
+    }
+}
+/** Save the skill profile to disk. */
+export function saveSkillProfile(profile) {
+    writeFileSync(SKILL_PROFILE_FILE, JSON.stringify(profile, null, 2), 'utf-8');
+}
+/**
+ * Update the user's skill level for a domain based on observed signals.
+ * Signals are things like: "used advanced git rebase", "asked basic TypeScript question",
+ * "configured CI pipeline without help".
+ */
+export function updateSkillLevel(domain, signal, indicatedLevel) {
+    const profile = loadSkillProfile();
+    const existing = profile.find(s => s.domain === domain);
+    const now = new Date().toISOString();
+    if (existing) {
+        // Add signal
+        existing.signals.push(signal);
+        if (existing.signals.length > 20)
+            existing.signals = existing.signals.slice(-20);
+        existing.lastUpdated = now;
+        // Adjust level — only move up, or down if strong evidence
+        const levels = ['beginner', 'intermediate', 'advanced', 'expert'];
+        const currentIdx = levels.indexOf(existing.level);
+        const newIdx = levels.indexOf(indicatedLevel);
+        if (newIdx > currentIdx) {
+            // Level up — require confidence
+            existing.confidence = Math.min(1, existing.confidence + 0.15);
+            if (existing.confidence >= 0.7) {
+                existing.level = indicatedLevel;
+                existing.confidence = 0.6; // reset after level change
+            }
+        }
+        else if (newIdx < currentIdx) {
+            // Level down — be cautious
+            existing.confidence = Math.max(0, existing.confidence - 0.1);
+            if (existing.confidence < 0.3) {
+                existing.level = indicatedLevel;
+                existing.confidence = 0.5;
+            }
+        }
+        else {
+            // Same level — increase confidence
+            existing.confidence = Math.min(1, existing.confidence + 0.1);
+        }
+        saveSkillProfile(profile);
+        return existing;
+    }
+    // New domain
+    const newSkill = {
+        domain,
+        level: indicatedLevel,
+        confidence: 0.5,
+        signals: [signal],
+        lastUpdated: now,
+    };
+    profile.push(newSkill);
+    saveSkillProfile(profile);
+    return newSkill;
+}
+/**
+ * Get the user's assessed skill level for a domain.
+ * Returns null if we have no data for this domain.
+ */
+export function getSkillLevel(domain) {
+    const profile = loadSkillProfile();
+    return profile.find(s => s.domain === domain) ?? null;
+}
+/**
+ * Get a guidance style recommendation based on the user's overall skill profile.
+ */
+export function getGuidanceStyle() {
+    const profile = loadSkillProfile();
+    if (profile.length === 0) {
+        return { tone: 'collaborative', explainLevel: 'moderate', domains: {} };
+    }
+    // Compute average skill level
+    const levels = { beginner: 0, intermediate: 1, advanced: 2, expert: 3 };
+    const avg = profile.reduce((sum, s) => sum + levels[s.level], 0) / profile.length;
+    const domains = {};
+    for (const s of profile)
+        domains[s.domain] = s.level;
+    if (avg < 0.8) {
+        return { tone: 'tutorial', explainLevel: 'detailed', domains };
+    }
+    if (avg < 2.0) {
+        return { tone: 'collaborative', explainLevel: 'moderate', domains };
+    }
+    return { tone: 'concise', explainLevel: 'minimal', domains };
+}
+// ---------------------------------------------------------------------------
+// Decision History
+// ---------------------------------------------------------------------------
+/**
+ * Record a decision with expected outcome.
+ */
+export function recordDecision(memory, decision) {
+    const record = {
+        ...decision,
+        id: `decision-${Date.now()}`,
+        timestamp: new Date().toISOString(),
+        actualOutcome: null,
+        outcomeAssessedAt: null,
+    };
+    return memory.upsert({
+        memoryType: 'decision',
+        title: `Decision: ${decision.decision.slice(0, 100)}`,
+        content: JSON.stringify(record),
+        scope: decision.repo,
+        sourceRef: decision.relatedFinding,
+        confidence: 0.8,
+        isPinned: false,
+        expiresAt: new Date(Date.now() + 90 * 24 * 60 * 60 * 1000).toISOString(),
+        tags: ['decision'],
+    });
+}
+/**
+ * Update a decision's actual outcome.
+ */
+export function assessDecisionOutcome(memory, decisionId, actualOutcome) {
+    const item = memory.get(decisionId);
+    if (!item || item.memoryType !== 'decision')
+        return false;
+    try {
+        const record = JSON.parse(item.content);
+        record.actualOutcome = actualOutcome;
+        record.outcomeAssessedAt = new Date().toISOString();
+        memory.upsert({
+            id: item.id,
+            memoryType: 'decision',
+            title: item.title,
+            content: JSON.stringify(record),
+            scope: item.scope,
+            sourceRef: item.sourceRef,
+            confidence: item.confidence,
+            isPinned: item.isPinned,
+            expiresAt: item.expiresAt,
+            tags: item.tags,
+        });
+        return true;
+    }
+    catch {
+        return false;
+    }
+}
+/**
+ * Get decisions that need outcome assessment (older than 7 days, no outcome yet).
+ */
+export function getPendingOutcomeAssessments(memory) {
+    const sevenDaysAgo = new Date(Date.now() - 7 * 24 * 60 * 60 * 1000).toISOString();
+    const decisions = memory.list({ memoryType: 'decision', limit: 50 });
+    return decisions
+        .map(item => {
+        try {
+            return JSON.parse(item.content);
+        }
+        catch {
+            return null;
+        }
+    })
+        .filter((d) => d !== null &&
+        d.actualOutcome === null &&
+        d.timestamp < sevenDaysAgo);
+}
+//# sourceMappingURL=orchestrator-learning.js.map