make strictness of note creation inferred from the data

apps/x/packages/core/src/agents/runtime.ts

@@ -2,6 +2,7 @@ import { jsonSchema, ModelMessage } from "ai";
 import fs from "fs";
 import path from "path";
 import { WorkDir } from "../config/config.js";
+import { getNoteCreationStrictness } from "../config/note_creation_config.js";
 import { Agent, ToolAttachment } from "@x/shared/dist/agent.js";
 import { AssistantContentPart, AssistantMessage, Message, MessageList, ProviderOptions, ToolCallPart, ToolMessage } from "@x/shared/dist/message.js";
 import { LanguageModel, stepCountIs, streamText, tool, Tool, ToolSet } from "ai";
@@ -245,16 +246,24 @@ export async function loadAgent(id: string): Promise<z.infer<typeof Agent>> {
         return CopilotAgent;
     }
 
-    // Special case: load built-in agents from checked-in files
+    // Built-in agents loaded from checked-in files
     const builtinAgents: Record<string, string> = {
-        'note_creation': '../knowledge/note_creation.md',
         'meeting-prep': '../pre_built/meeting-prep.md',
         'email-draft': '../pre_built/email-draft.md',
     };
 
-    if (id in builtinAgents) {
-        const currentDir = path.dirname(new URL(import.meta.url).pathname);
-        const agentFilePath = path.join(currentDir, builtinAgents[id]);
+    // Resolve agent file path (note_creation is dynamic based on strictness config)
+    let agentFilePath: string | null = null;
+    const currentDir = path.dirname(new URL(import.meta.url).pathname);
+
+    if (id === 'note_creation') {
+        const strictness = getNoteCreationStrictness();
+        agentFilePath = path.join(currentDir, `../knowledge/note_creation_${strictness}.md`);
+    } else if (id in builtinAgents) {
+        agentFilePath = path.join(currentDir, builtinAgents[id]);
+    }
+
+    if (agentFilePath) {
         const raw = fs.readFileSync(agentFilePath, "utf8");
 
         let agent: z.infer<typeof Agent> = {

apps/x/packages/core/src/config/config.ts

@@ -13,4 +13,16 @@ function ensureDirs() {
     ensure(path.join(WorkDir, "knowledge"));
 }
 
-ensureDirs();
+function ensureDefaultConfigs() {
+    // Create note_creation.json with default strictness if it doesn't exist
+    const noteCreationConfig = path.join(WorkDir, "config", "note_creation.json");
+    if (!fs.existsSync(noteCreationConfig)) {
+        fs.writeFileSync(noteCreationConfig, JSON.stringify({
+            strictness: "high",
+            configured: false
+        }, null, 2));
+    }
+}
+
+ensureDirs();
+ensureDefaultConfigs();

apps/x/packages/core/src/config/note_creation_config.ts

@@ -0,0 +1,94 @@
+import fs from 'fs';
+import path from 'path';
+import { WorkDir } from './config.js';
+
+export type NoteCreationStrictness = 'low' | 'medium' | 'high';
+
+interface NoteCreationConfig {
+    strictness: NoteCreationStrictness;
+    configured: boolean;
+}
+
+const CONFIG_FILE = path.join(WorkDir, 'config', 'note_creation.json');
+const DEFAULT_STRICTNESS: NoteCreationStrictness = 'high';
+
+/**
+ * Read the full config file.
+ */
+function readConfig(): NoteCreationConfig {
+    try {
+        if (!fs.existsSync(CONFIG_FILE)) {
+            return { strictness: DEFAULT_STRICTNESS, configured: false };
+        }
+        const raw = fs.readFileSync(CONFIG_FILE, 'utf-8');
+        const config = JSON.parse(raw);
+        return {
+            strictness: ['low', 'medium', 'high'].includes(config.strictness)
+                ? config.strictness
+                : DEFAULT_STRICTNESS,
+            configured: config.configured === true,
+        };
+    } catch {
+        return { strictness: DEFAULT_STRICTNESS, configured: false };
+    }
+}
+
+/**
+ * Write the full config file.
+ */
+function writeConfig(config: NoteCreationConfig): void {
+    const configDir = path.dirname(CONFIG_FILE);
+    if (!fs.existsSync(configDir)) {
+        fs.mkdirSync(configDir, { recursive: true });
+    }
+    fs.writeFileSync(CONFIG_FILE, JSON.stringify(config, null, 2));
+}
+
+/**
+ * Get the current note creation strictness setting.
+ * Defaults to 'high' if config doesn't exist.
+ */
+export function getNoteCreationStrictness(): NoteCreationStrictness {
+    return readConfig().strictness;
+}
+
+/**
+ * Set the note creation strictness setting.
+ * Preserves the configured flag.
+ */
+export function setNoteCreationStrictness(strictness: NoteCreationStrictness): void {
+    const config = readConfig();
+    config.strictness = strictness;
+    writeConfig(config);
+}
+
+/**
+ * Check if strictness has been auto-configured based on email analysis.
+ */
+export function isStrictnessConfigured(): boolean {
+    return readConfig().configured;
+}
+
+/**
+ * Mark strictness as configured (after auto-analysis).
+ */
+export function markStrictnessConfigured(): void {
+    const config = readConfig();
+    config.configured = true;
+    writeConfig(config);
+}
+
+/**
+ * Set strictness and mark as configured in one operation.
+ */
+export function setStrictnessAndMarkConfigured(strictness: NoteCreationStrictness): void {
+    writeConfig({ strictness, configured: true });
+}
+
+/**
+ * Get the agent file name suffix based on strictness.
+ */
+export function getNoteCreationAgentSuffix(): string {
+    const strictness = getNoteCreationStrictness();
+    return `note_creation_${strictness}`;
+}

apps/x/packages/core/src/config/strictness_analyzer.ts

@@ -0,0 +1,482 @@
+import fs from 'fs';
+import path from 'path';
+import { WorkDir } from './config.js';
+import {
+    NoteCreationStrictness,
+    setStrictnessAndMarkConfigured,
+    isStrictnessConfigured,
+} from './note_creation_config.js';
+
+const GMAIL_SYNC_DIR = path.join(WorkDir, 'gmail_sync');
+
+interface EmailInfo {
+    threadId: string;
+    subject: string;
+    senders: string[];
+    senderEmails: string[];
+    body: string;
+    date: Date | null;
+}
+
+interface AnalysisResult {
+    totalEmails: number;
+    uniqueSenders: number;
+    newsletterCount: number;
+    automatedCount: number;
+    consumerServiceCount: number;
+    businessCount: number;
+    mediumWouldCreate: number;
+    lowWouldCreate: number;
+    recommendation: NoteCreationStrictness;
+    reason: string;
+}
+
+// Common newsletter/marketing patterns
+const NEWSLETTER_PATTERNS = [
+    /unsubscribe/i,
+    /opt[- ]?out/i,
+    /email preferences/i,
+    /manage.*subscription/i,
+    /via sendgrid/i,
+    /via mailchimp/i,
+    /via hubspot/i,
+    /via constantcontact/i,
+    /list-unsubscribe/i,
+];
+
+const NEWSLETTER_SENDER_PATTERNS = [
+    /^noreply@/i,
+    /^no-reply@/i,
+    /^newsletter@/i,
+    /^marketing@/i,
+    /^hello@/i,
+    /^info@/i,
+    /^team@/i,
+    /^updates@/i,
+    /^news@/i,
+];
+
+// Automated/transactional patterns
+const AUTOMATED_PATTERNS = [
+    /^notifications?@/i,
+    /^alerts?@/i,
+    /^support@/i,
+    /^billing@/i,
+    /^receipts?@/i,
+    /^orders?@/i,
+    /^shipping@/i,
+    /^noreply@/i,
+    /^donotreply@/i,
+    /^mailer-daemon/i,
+    /^postmaster@/i,
+];
+
+const AUTOMATED_SUBJECT_PATTERNS = [
+    /password reset/i,
+    /verify your email/i,
+    /login alert/i,
+    /security alert/i,
+    /your order/i,
+    /order confirmation/i,
+    /shipping confirmation/i,
+    /receipt for/i,
+    /invoice/i,
+    /payment received/i,
+    /\[GitHub\]/i,
+    /\[Jira\]/i,
+    /\[Slack\]/i,
+    /\[Linear\]/i,
+    /\[Notion\]/i,
+];
+
+// Consumer service domains (not business-relevant)
+const CONSUMER_SERVICE_DOMAINS = [
+    'amazon.com', 'amazon.co.uk',
+    'netflix.com',
+    'spotify.com',
+    'uber.com', 'ubereats.com',
+    'doordash.com', 'grubhub.com',
+    'apple.com', 'apple.id',
+    'google.com', 'youtube.com',
+    'facebook.com', 'meta.com', 'instagram.com',
+    'twitter.com', 'x.com',
+    'linkedin.com',
+    'dropbox.com',
+    'paypal.com', 'venmo.com',
+    'chase.com', 'bankofamerica.com', 'wellsfargo.com', 'citi.com',
+    'att.com', 'verizon.com', 't-mobile.com',
+    'comcast.com', 'xfinity.com',
+    'delta.com', 'united.com', 'southwest.com', 'aa.com',
+    'airbnb.com', 'vrbo.com',
+    'walmart.com', 'target.com', 'bestbuy.com',
+    'costco.com',
+];
+
+/**
+ * Parse a synced email markdown file
+ */
+function parseEmailFile(filePath: string): EmailInfo | null {
+    try {
+        const content = fs.readFileSync(filePath, 'utf-8');
+        const lines = content.split('\n');
+
+        // Extract subject from first heading
+        const subjectLine = lines.find(l => l.startsWith('# '));
+        const subject = subjectLine ? subjectLine.slice(2).trim() : '';
+
+        // Extract thread ID
+        const threadIdLine = lines.find(l => l.startsWith('**Thread ID:**'));
+        const threadId = threadIdLine ? threadIdLine.replace('**Thread ID:**', '').trim() : path.basename(filePath, '.md');
+
+        // Extract all senders
+        const senders: string[] = [];
+        const senderEmails: string[] = [];
+        let latestDate: Date | null = null;
+
+        for (const line of lines) {
+            if (line.startsWith('### From:')) {
+                const from = line.replace('### From:', '').trim();
+                senders.push(from);
+
+                // Extract email from "Name <email@domain.com>" format
+                const emailMatch = from.match(/<([^>]+)>/) || from.match(/([^\s<]+@[^\s>]+)/);
+                if (emailMatch) {
+                    senderEmails.push(emailMatch[1].toLowerCase());
+                }
+            }
+            if (line.startsWith('**Date:**')) {
+                const dateStr = line.replace('**Date:**', '').trim();
+                try {
+                    const parsed = new Date(dateStr);
+                    if (!isNaN(parsed.getTime())) {
+                        if (!latestDate || parsed > latestDate) {
+                            latestDate = parsed;
+                        }
+                    }
+                } catch {
+                    // ignore parse errors
+                }
+            }
+        }
+
+        return {
+            threadId,
+            subject,
+            senders,
+            senderEmails,
+            body: content,
+            date: latestDate,
+        };
+    } catch (error) {
+        console.error(`Error parsing email file ${filePath}:`, error);
+        return null;
+    }
+}
+
+/**
+ * Check if email is a newsletter/mass email
+ */
+function isNewsletter(email: EmailInfo): boolean {
+    // Check sender patterns
+    for (const senderEmail of email.senderEmails) {
+        for (const pattern of NEWSLETTER_SENDER_PATTERNS) {
+            if (pattern.test(senderEmail)) {
+                return true;
+            }
+        }
+    }
+
+    // Check body for unsubscribe patterns
+    for (const pattern of NEWSLETTER_PATTERNS) {
+        if (pattern.test(email.body)) {
+            return true;
+        }
+    }
+
+    return false;
+}
+
+/**
+ * Check if email is automated/transactional
+ */
+function isAutomated(email: EmailInfo): boolean {
+    // Check sender patterns
+    for (const senderEmail of email.senderEmails) {
+        for (const pattern of AUTOMATED_PATTERNS) {
+            if (pattern.test(senderEmail)) {
+                return true;
+            }
+        }
+    }
+
+    // Check subject patterns
+    for (const pattern of AUTOMATED_SUBJECT_PATTERNS) {
+        if (pattern.test(email.subject)) {
+            return true;
+        }
+    }
+
+    return false;
+}
+
+/**
+ * Check if email is from a consumer service
+ */
+function isConsumerService(email: EmailInfo): boolean {
+    for (const senderEmail of email.senderEmails) {
+        const domain = senderEmail.split('@')[1];
+        if (domain) {
+            // Check exact match or subdomain match (e.g., mail.amazon.com)
+            for (const consumerDomain of CONSUMER_SERVICE_DOMAINS) {
+                if (domain === consumerDomain || domain.endsWith(`.${consumerDomain}`)) {
+                    return true;
+                }
+            }
+        }
+    }
+    return false;
+}
+
+/**
+ * Categorize an email based on its characteristics.
+ * Returns the category which determines how different strictness levels would handle it.
+ */
+type EmailCategory = 'internal' | 'newsletter' | 'automated' | 'consumer_service' | 'business';
+
+function categorizeEmail(email: EmailInfo, userDomain: string): {
+    category: EmailCategory;
+    externalSenders: string[];
+} {
+    // Filter out user's own domain
+    const externalSenders = email.senderEmails.filter(e => !e.endsWith(`@${userDomain}`));
+    if (externalSenders.length === 0) {
+        return { category: 'internal', externalSenders: [] };
+    }
+
+    if (isNewsletter(email)) {
+        return { category: 'newsletter', externalSenders };
+    }
+
+    if (isAutomated(email)) {
+        return { category: 'automated', externalSenders };
+    }
+
+    if (isConsumerService(email)) {
+        return { category: 'consumer_service', externalSenders };
+    }
+
+    return { category: 'business', externalSenders };
+}
+
+/**
+ * Infer user's domain from email patterns.
+ * Looks for the most common sender domain that appears frequently,
+ * assuming the user's own emails would be the most common sender.
+ */
+function inferUserDomain(emails: EmailInfo[]): string {
+    const domainCounts = new Map<string, number>();
+
+    for (const email of emails) {
+        for (const senderEmail of email.senderEmails) {
+            const domain = senderEmail.split('@')[1];
+            if (domain) {
+                domainCounts.set(domain, (domainCounts.get(domain) || 0) + 1);
+            }
+        }
+    }
+
+    // Find the most frequent domain (likely the user's domain)
+    let maxCount = 0;
+    let userDomain = '';
+
+    for (const [domain, count] of domainCounts) {
+        // Skip known consumer/service domains
+        const isConsumer = CONSUMER_SERVICE_DOMAINS.some(
+            d => domain === d || domain.endsWith(`.${d}`)
+        );
+
+        if (!isConsumer && count > maxCount) {
+            maxCount = count;
+            userDomain = domain;
+        }
+    }
+
+    // Fallback if we couldn't determine
+    return userDomain || 'example.com';
+}
+
+/**
+ * Analyze emails and recommend a strictness level based on email patterns.
+ *
+ * Strictness levels filter emails as follows:
+ * - High: Only creates notes from meetings, emails just update existing notes
+ * - Medium: Creates notes for business emails (filters out consumer services)
+ * - Low: Creates notes for any human sender (only filters newsletters/automated)
+ */
+export function analyzeEmailsAndRecommend(): AnalysisResult {
+    const emails: EmailInfo[] = [];
+
+    // Read all email files from gmail_sync
+    if (fs.existsSync(GMAIL_SYNC_DIR)) {
+        const files = fs.readdirSync(GMAIL_SYNC_DIR).filter(f => f.endsWith('.md'));
+
+        // Filter to last 30 days
+        const thirtyDaysAgo = new Date();
+        thirtyDaysAgo.setDate(thirtyDaysAgo.getDate() - 30);
+
+        for (const file of files) {
+            const filePath = path.join(GMAIL_SYNC_DIR, file);
+            const email = parseEmailFile(filePath);
+            if (email) {
+                // Include if date is within 30 days or if we can't parse the date
+                if (!email.date || email.date >= thirtyDaysAgo) {
+                    emails.push(email);
+                }
+            }
+        }
+    }
+
+    const userDomain = inferUserDomain(emails);
+    console.log(`[StrictnessAnalyzer] Inferred user domain: ${userDomain}`);
+
+    // Track unique senders by category
+    const uniqueSenders = new Set<string>();
+    const newsletterSenders = new Set<string>();
+    const automatedSenders = new Set<string>();
+    const consumerServiceSenders = new Set<string>();
+    const businessSenders = new Set<string>();
+
+    let newsletterCount = 0;
+    let automatedCount = 0;
+    let consumerServiceCount = 0;
+    let businessCount = 0;
+
+    for (const email of emails) {
+        const result = categorizeEmail(email, userDomain);
+
+        for (const sender of result.externalSenders) {
+            uniqueSenders.add(sender);
+        }
+
+        switch (result.category) {
+            case 'newsletter':
+                newsletterCount++;
+                for (const sender of result.externalSenders) newsletterSenders.add(sender);
+                break;
+            case 'automated':
+                automatedCount++;
+                for (const sender of result.externalSenders) automatedSenders.add(sender);
+                break;
+            case 'consumer_service':
+                consumerServiceCount++;
+                for (const sender of result.externalSenders) consumerServiceSenders.add(sender);
+                break;
+            case 'business':
+                businessCount++;
+                for (const sender of result.externalSenders) businessSenders.add(sender);
+                break;
+        }
+    }
+
+    // Calculate what each strictness level would capture:
+    // - Low: business + consumer_service senders (all human, non-automated)
+    // - Medium: business senders only (filters consumer services)
+    // - High: none from emails (only meetings create notes)
+    const lowWouldCreate = businessSenders.size + consumerServiceSenders.size;
+    const mediumWouldCreate = businessSenders.size;
+
+    // Determine recommendation based on email patterns
+    let recommendation: NoteCreationStrictness;
+    let reason: string;
+
+    const totalHumanSenders = lowWouldCreate;
+    const noiseRatio = uniqueSenders.size > 0
+        ? (newsletterSenders.size + automatedSenders.size) / uniqueSenders.size
+        : 0;
+    const consumerRatio = totalHumanSenders > 0
+        ? consumerServiceSenders.size / totalHumanSenders
+        : 0;
+
+    if (totalHumanSenders > 100) {
+        // High volume of contacts - recommend high to avoid noise
+        recommendation = 'high';
+        reason = `High volume of contacts (${totalHumanSenders} potential). High strictness focuses on people you meet, avoiding email overload.`;
+    } else if (totalHumanSenders > 50) {
+        // Moderate volume - recommend medium
+        recommendation = 'medium';
+        reason = `Moderate contact volume (${totalHumanSenders}). Medium strictness captures business contacts (${mediumWouldCreate}) while filtering consumer services.`;
+    } else if (consumerRatio > 0.5) {
+        // Lots of consumer service emails - medium helps filter
+        recommendation = 'medium';
+        reason = `${Math.round(consumerRatio * 100)}% of emails are from consumer services. Medium strictness filters these to focus on business contacts.`;
+    } else if (totalHumanSenders < 30) {
+        // Low volume - comprehensive capture is manageable
+        recommendation = 'low';
+        reason = `Low contact volume (${totalHumanSenders}). Low strictness provides comprehensive capture without overwhelming.`;
+    } else {
+        recommendation = 'medium';
+        reason = `Medium strictness provides a good balance, capturing ${mediumWouldCreate} business contacts.`;
+    }
+
+    return {
+        totalEmails: emails.length,
+        uniqueSenders: uniqueSenders.size,
+        newsletterCount,
+        automatedCount,
+        consumerServiceCount,
+        businessCount,
+        mediumWouldCreate,
+        lowWouldCreate,
+        recommendation,
+        reason,
+    };
+}
+
+/**
+ * Run analysis and auto-configure strictness if not already done.
+ * Returns true if configuration was updated.
+ */
+export function autoConfigureStrictnessIfNeeded(): boolean {
+    if (isStrictnessConfigured()) {
+        return false;
+    }
+
+    // Check if there are any emails to analyze
+    if (!fs.existsSync(GMAIL_SYNC_DIR)) {
+        console.log('[StrictnessAnalyzer] No gmail_sync directory found, skipping auto-configuration');
+        return false;
+    }
+
+    const emailFiles = fs.readdirSync(GMAIL_SYNC_DIR).filter(f => f.endsWith('.md'));
+    if (emailFiles.length === 0) {
+        console.log('[StrictnessAnalyzer] No emails found to analyze, skipping auto-configuration');
+        return false;
+    }
+
+    // Need at least 10 emails for meaningful analysis
+    if (emailFiles.length < 10) {
+        console.log(`[StrictnessAnalyzer] Only ${emailFiles.length} emails found, need at least 10 for meaningful analysis. Using default 'high' strictness.`);
+        setStrictnessAndMarkConfigured('high');
+        return true;
+    }
+
+    console.log('[StrictnessAnalyzer] Running email analysis for auto-configuration...');
+    const result = analyzeEmailsAndRecommend();
+
+    console.log('[StrictnessAnalyzer] Analysis complete:');
+    console.log(`  - Total emails analyzed: ${result.totalEmails}`);
+    console.log(`  - Unique external senders: ${result.uniqueSenders}`);
+    console.log(`  - Newsletters/mass emails: ${result.newsletterCount}`);
+    console.log(`  - Automated/transactional: ${result.automatedCount}`);
+    console.log(`  - Consumer services: ${result.consumerServiceCount}`);
+    console.log(`  - Business emails: ${result.businessCount}`);
+    console.log(`  - Medium strictness would capture: ${result.mediumWouldCreate} contacts`);
+    console.log(`  - Low strictness would capture: ${result.lowWouldCreate} contacts`);
+    console.log(`  - Recommendation: ${result.recommendation.toUpperCase()}`);
+    console.log(`  - Reason: ${result.reason}`);
+
+    setStrictnessAndMarkConfigured(result.recommendation);
+    console.log(`[StrictnessAnalyzer] Auto-configured note creation strictness to: ${result.recommendation}`);
+
+    return true;
+}

apps/x/packages/core/src/knowledge/README.md

@@ -137,7 +137,82 @@ resetGraphState(); // Clears the state file
 
 Or manually delete: `~/.rowboat/knowledge_graph_state.json`
 
-## Configuration
+## Note Creation Strictness
+
+The system supports three strictness levels that control how aggressively notes are created from emails. Meetings always create notes at all levels.
+
+### Configuration
+
+Strictness is configured in `~/.rowboat/config/note_creation.json`:
+
+```json
+{
+  "strictness": "medium",
+  "configured": true
+}
+```
+
+On first run, the system auto-analyzes your emails and recommends a setting based on volume and patterns.
+
+### Strictness Levels
+
+| Level | Philosophy |
+|-------|------------|
+| **High** | "Meetings create notes. Emails enrich them." |
+| **Medium** | "Both create notes, but emails require personalized content." |
+| **Low** | "Capture broadly. Never miss a potentially important contact." |
+
+### What Each Level Filters
+
+| Email Type | High | Medium | Low |
+|------------|------|--------|-----|
+| Mass newsletters | Skip | Skip | Skip |
+| Automated/system emails | Skip | Skip | Skip |
+| Consumer services (Amazon, Netflix, banks) | Skip | Skip | ✅ Create |
+| Generic cold sales | Skip | Skip | ✅ Create |
+| Recruiters | Skip | Skip | ✅ Create |
+| Support reps | Skip | Skip | ✅ Create |
+| Personalized business emails | Skip | ✅ Create | ✅ Create |
+| Warm intros | ✅ Create | ✅ Create | ✅ Create |
+
+### High Strictness
+
+- Emails **never create** new notes (only meetings do)
+- Emails can only **update existing** notes for people you've already met
+- Exception: Warm intros from known contacts can create notes
+- Best for: Users who get lots of emails and want minimal noise
+
+### Medium Strictness
+
+- Emails **can create** notes if personalized and business-relevant
+- Filters out consumer services, mass mail, generic pitches
+- Warm intros from anyone (not just existing contacts) create notes
+- Best for: Balanced capture of relevant business contacts
+
+### Low Strictness
+
+- Creates notes for **any identifiable human sender**
+- Only skips obvious automated emails and newsletters
+- Philosophy: "Better to have a note you don't need than to miss someone important"
+- Best for: Users with low email volume who want comprehensive capture
+
+### Auto-Configuration
+
+On first run, `strictness_analyzer.ts` analyzes your emails and recommends a level:
+
+- **>100 human senders** → Recommends High (avoid overload)
+- **50-100 senders** → Recommends Medium (balanced)
+- **>50% consumer services** → Recommends Medium (filter noise)
+- **<30 senders** → Recommends Low (comprehensive capture is manageable)
+
+### Prompt Files
+
+Each strictness level has its own agent prompt:
+- `note_creation_high.md` - Original strict rules
+- `note_creation_medium.md` - Relaxed for personalized emails
+- `note_creation_low.md` - Minimal filtering
+
+## Other Configuration
 
 ### Batch Size
 Change `BATCH_SIZE` in `build_graph.ts` (currently 25 files per batch)

apps/x/packages/core/src/knowledge/build_graph.ts

@@ -1,6 +1,7 @@
 import fs from 'fs';
 import path from 'path';
 import { WorkDir } from '../config/config.js';
+import { autoConfigureStrictnessIfNeeded } from '../config/strictness_analyzer.js';
 import { createRun, createMessage } from '../runs/runs.js';
 import { bus } from '../runs/bus.js';
 import {
@@ -181,6 +182,9 @@ export async function buildGraph(sourceDir: string): Promise<void> {
 async function processAllSources(): Promise<void> {
     console.log('[GraphBuilder] Checking for new content in all sources...');
 
+    // Auto-configure strictness on first run if not already done
+    autoConfigureStrictnessIfNeeded();
+
     let anyFilesProcessed = false;
 
     for (const folder of SOURCE_FOLDERS) {

apps/x/packages/core/src/knowledge/note_creation_low.md

@@ -0,0 +1,725 @@
+---
+model: gpt-5.2
+tools:
+  workspace-writeFile:
+    type: builtin
+    name: workspace-writeFile
+  workspace-readFile:
+    type: builtin
+    name: workspace-readFile
+  workspace-readdir:
+    type: builtin
+    name: workspace-readdir
+  workspace-mkdir:
+    type: builtin
+    name: workspace-mkdir
+  executeCommand:
+    type: builtin
+    name: executeCommand
+---
+# Task
+
+You are a memory agent. Given a single source file (email or meeting transcript), you will:
+
+1. **Determine source type (meeting or email)**
+2. **Evaluate if the source is worth processing**
+3. **Search for all existing related notes**
+4. **Resolve entities to canonical names**
+5. Identify new entities worth tracking
+6. Extract structured information (decisions, commitments, key facts)
+7. **Detect state changes (status updates, resolved items, role changes)**
+8. Create new notes or update existing notes
+9. **Apply state changes to existing notes**
+
+The core rule: **Capture broadly. Both meetings and emails create notes for most external contacts.**
+
+You have full read access to the existing knowledge directory. Use this extensively to:
+- Find existing notes for people, organizations, projects mentioned
+- Resolve ambiguous names (find existing note for "David")
+- Understand existing relationships before updating
+- Avoid creating duplicate notes
+- Maintain consistency with existing content
+- **Detect when new information changes the state of existing notes**
+
+# Inputs
+
+1. **source_file**: Path to a single file to process (email or meeting transcript)
+2. **knowledge_folder**: Path to Obsidian vault (read/write access)
+3. **user**: Information about the owner of this memory
+   - name: e.g., "Arj"
+   - email: e.g., "arj@rowboat.com"
+   - domain: e.g., "rowboat.com"
+
+# Tools Available
+
+You have access to `executeCommand` to run shell commands:
+```
+executeCommand("ls {path}")                     # List directory contents
+executeCommand("cat {path}")                    # Read file contents
+executeCommand("grep -r '{pattern}' {path}")    # Search across files
+executeCommand("grep -r -l '{pattern}' {path}") # List files containing pattern
+executeCommand("grep -r -i '{pattern}' {path}") # Case-insensitive search
+executeCommand("head -50 {path}")               # Read first 50 lines
+executeCommand("write {path} {content}")        # Create or overwrite file
+```
+
+**Important:** Use shell escaping for paths with spaces:
+```
+executeCommand("cat 'knowledge_folder/People/Sarah Chen.md'")
+executeCommand("grep -r 'David' 'knowledge_folder/People/'")
+```
+
+# Output
+
+Either:
+- **SKIP** with reason, if source should be ignored
+- Updated or new markdown files in notes_folder
+
+---
+
+# The Core Rule: Low Strictness - Capture Broadly
+
+**LOW STRICTNESS MODE**
+
+This mode prioritizes comprehensive capture over selectivity. The goal is to never miss a potentially important contact.
+
+**Meetings create notes for:**
+- All external attendees (anyone not @user.domain)
+
+**Emails create notes for:**
+- Any personalized email from an identifiable sender
+- Anyone who reaches out directly
+- Any external contact who communicates with you
+
+**Only skip:**
+- Obvious automated/system emails (no human sender)
+- Mass newsletters with unsubscribe links
+- Truly anonymous or unidentifiable senders
+
+**Philosophy:** It's better to have a note you don't need than to miss tracking someone important.
+
+---
+
+# Step 0: Determine Source Type
+
+Read the source file and determine if it's a meeting or email.
+```
+executeCommand("cat '{source_file}'")
+```
+
+**Meeting indicators:**
+- Has `Attendees:` field
+- Has `Meeting:` title
+- Transcript format with speaker labels
+- Calendar event metadata
+
+**Email indicators:**
+- Has `From:` and `To:` fields
+- Has `Subject:` field
+- Email signature
+
+**Set processing mode:**
+- `source_type = "meeting"` → Create notes for all external attendees
+- `source_type = "email"` → Create notes for sender if identifiable human
+
+---
+
+# Step 1: Source Filtering (Minimal)
+
+## Skip Only These Sources
+
+### Mass Newsletters
+
+**Indicators (must have MULTIPLE of these):**
+- Unsubscribe link in body or footer
+- From a marketing address (noreply@, newsletter@, marketing@)
+- Sent to multiple recipients or undisclosed-recipients
+- Sent via marketing platforms (via sendgrid, via mailchimp, etc.)
+
+**Action:** SKIP with reason "Mass newsletter"
+
+### Purely Automated (No Human Sender)
+
+**Indicators:**
+- From automated systems with no human behind them (alerts@, notifications@)
+- Password resets, login alerts
+- System notifications (GitHub automated, CI/CD alerts)
+- Receipt confirmations with no human contact info
+
+**Action:** SKIP with reason "Automated system message"
+
+### Truly Low-Signal
+
+**Indicators (must be clearly content-free):**
+- Body is ONLY "Thanks!", "Got it", "OK" with nothing else
+- Auto-replies ("I'm out of office") with no human context
+
+**Action:** SKIP with reason "No substantive content"
+
+## Process Everything Else
+
+**Important:** When in doubt, PROCESS. In low strictness mode, we err on the side of capturing more.
+
+If skipping:
+```
+SKIP
+Reason: {reason}
+```
+
+If processing, continue to Step 2.
+
+---
+
+# Step 2: Read and Parse Source File
+```
+executeCommand("cat '{source_file}'")
+```
+
+Extract metadata:
+
+**For meetings:**
+- **Date:** From header or filename
+- **Title:** Meeting name
+- **Attendees:** List of participants
+- **Duration:** If available
+
+**For emails:**
+- **Date:** From `Date:` header
+- **Subject:** From `Subject:` header
+- **From:** Sender email/name
+- **To/Cc:** Recipients
+
+## 2a: Exclude Self
+
+Never create or update notes for:
+- The user (matches user.name, user.email, or @user.domain)
+- Anyone @{user.domain} (colleagues at user's company)
+
+Filter these out from attendees/participants before proceeding.
+
+## 2b: Extract All Name Variants
+
+From the source, collect every way entities are referenced:
+
+**People variants:**
+- Full names: "Sarah Chen"
+- First names only: "Sarah"
+- Last names only: "Chen"
+- Initials: "S. Chen"
+- Email addresses: "sarah@acme.com"
+- Roles/titles: "their CTO", "the VP of Engineering"
+
+**Organization variants:**
+- Full names: "Acme Corporation"
+- Short names: "Acme"
+- Abbreviations: "AC"
+- Email domains: "@acme.com"
+
+**Project variants:**
+- Explicit names: "Project Atlas"
+- Descriptive references: "the integration", "the pilot", "the deal"
+
+Create a list of all variants found.
+
+---
+
+# Step 3: Search for Existing Notes
+
+For each variant identified, search the notes folder thoroughly.
+
+## 3a: Search by People
+```bash
+executeCommand("grep -r -i -l 'Sarah Chen' '{knowledge_folder}/'")
+executeCommand("grep -r -i -l 'Sarah' '{knowledge_folder}/People/'")
+executeCommand("grep -r -i -l 'sarah@acme.com' '{knowledge_folder}/'")
+executeCommand("grep -r -i -l '@acme.com' '{knowledge_folder}/'")
+executeCommand("grep -r -i 'Aliases.*Sarah' '{knowledge_folder}/People/'")
+```
+
+## 3b: Search by Organizations
+```bash
+executeCommand("ls '{knowledge_folder}/Organizations/'")
+executeCommand("grep -r -i -l 'Acme' '{knowledge_folder}/Organizations/'")
+executeCommand("grep -r -i 'Domain.*acme.com' '{knowledge_folder}/Organizations/'")
+```
+
+## 3c: Search by Projects and Topics
+```bash
+executeCommand("ls '{knowledge_folder}/Projects/'")
+executeCommand("grep -r -i 'Acme' '{knowledge_folder}/Projects/'")
+executeCommand("ls '{knowledge_folder}/Topics/'")
+```
+
+## 3d: Read Candidate Notes
+
+For every note file found in searches, read it to understand context.
+
+## 3e: Matching Criteria
+
+Use standard matching criteria for names, emails, and organizations.
+
+---
+
+# Step 4: Resolve Entities to Canonical Names
+
+Using the search results from Step 3, resolve each variant to a canonical name.
+
+## 4a: Build Resolution Map
+
+Create a mapping from every source reference to its canonical form.
+
+## 4b: Apply Source Type Rules (Low Strictness)
+
+**If source_type == "meeting":**
+- Resolved entities → Update existing notes
+- New entities → Create new notes for ALL external attendees
+
+**If source_type == "email" (LOW STRICTNESS):**
+- Resolved entities → Update existing notes
+- New entities → Create notes for the sender and any mentioned contacts
+
+## 4c: Disambiguation Rules
+
+When multiple candidates match a variant, disambiguate by:
+1. Email match (definitive)
+2. Organization context (strong signal)
+3. Role match
+4. Recency (tiebreaker)
+
+## 4d: Resolution Map Output
+
+Final resolution map before proceeding:
+```
+RESOLVED (use canonical name with absolute path):
+- "Sarah", "Sarah Chen", "sarah@acme.com" → [[People/Sarah Chen]]
+
+NEW ENTITIES (create notes):
+- "Jennifer" (CTO, Acme Corp) → Create [[People/Jennifer]]
+
+AMBIGUOUS (create with disambiguation note):
+- "Mike" (no context) → Create [[People/Mike]] with note about ambiguity
+```
+
+---
+
+# Step 5: Identify New Entities (Low Strictness - Capture Broadly)
+
+For entities not resolved to existing notes, create notes for most of them.
+
+## People
+
+### Who Gets a Note (Low Strictness)
+
+**CREATE a note for:**
+- ALL external meeting attendees (not @user.domain)
+- ALL email senders with identifiable names/emails
+- Anyone CC'd on emails who seems relevant
+- Anyone mentioned by name in conversations
+- Cold outreach senders (even if unsolicited)
+- Sales reps, recruiters, service providers
+- Anyone who might be useful to remember later
+
+**DO NOT create notes for:**
+- Internal colleagues (@user.domain)
+- Truly anonymous/unidentifiable senders
+- System-generated sender names with no human behind them
+
+### The Low Strictness Test
+
+Ask: Could this person ever be useful to remember?
+
+- Sarah Chen, VP Engineering → **Yes, create note**
+- James from HSBC → **Yes, create note** (might need banking help again)
+- Random recruiter → **Yes, create note** (might want to contact later)
+- Cold sales person → **Yes, create note** (might be relevant someday)
+- Support rep → **Yes, create note** (might need them again)
+
+### Role Inference
+
+If role is not explicitly stated, infer from context. Write "Unknown" only if truly impossible to infer anything.
+
+### Relationship Type Guide (Low Strictness)
+
+| Relationship Type | Create People Notes? | Create Org Note? |
+|-------------------|----------------------|------------------|
+| Customer | Yes — all contacts | Yes |
+| Prospect | Yes — all contacts | Yes |
+| Investor | Yes | Yes |
+| Partner | Yes — all contacts | Yes |
+| Vendor | Yes — all contacts | Yes |
+| Bank/Financial | Yes | Yes |
+| Candidate | Yes | No |
+| Recruiter | Yes | Optional |
+| Service provider | Yes | Optional |
+| Cold outreach | Yes | Optional |
+| Support interaction | Yes | Optional |
+
+## Organizations
+
+**CREATE a note if:**
+- Anyone from that org is mentioned or contacted you
+- The org is mentioned in any context
+
+**Only skip:**
+- Organizations you genuinely can't identify
+
+## Projects
+
+**CREATE a note if:**
+- Discussed in meeting or email
+- Any indication of ongoing work or collaboration
+
+## Topics
+
+**CREATE a note if:**
+- Mentioned more than once
+- Seems like a recurring theme
+
+---
+
+# Step 6: Extract Content
+
+For each entity that has or will have a note, extract relevant content.
+
+## Decisions
+
+Extract what was decided, when, by whom, and why.
+
+## Commitments
+
+Extract who committed to what, and any deadlines.
+
+## Key Facts
+
+Key facts should be **substantive information** — not commentary about missing data.
+
+**Extract if:**
+- Specific numbers, dates, or metrics
+- Preferences or working style
+- Background information
+- Authority or decision process
+- Concerns or constraints
+- What they're working on or interested in
+
+**Never include:**
+- Meta-commentary about missing data
+- Obvious facts already in Info section
+- Placeholder text
+
+**If there are no substantive key facts, leave the section empty.**
+
+## Open Items
+
+**Include:**
+- Commitments made
+- Requests received
+- Next steps discussed
+- Follow-ups agreed
+
+**Never include:**
+- Data gaps or research tasks
+- Wishes or hypotheticals
+
+## Summary
+
+The summary should answer: **"Who is this person and why do I know them?"**
+
+Write 2-3 sentences covering their role/function, context of the relationship, and what you're discussing.
+
+## Activity Summary
+
+One line summarizing this source's relevance to the entity:
+```
+**{YYYY-MM-DD}** ({meeting|email}): {Summary with [[links]]}
+```
+
+---
+
+# Step 7: Detect State Changes
+
+Review the extracted content for signals that existing note fields should be updated.
+
+## 7a: Project Status Changes
+
+Look for signals like "approved", "on hold", "cancelled", "completed", etc.
+
+## 7b: Open Item Resolution
+
+Look for signals that tracked items are now complete.
+
+## 7c: Role/Title Changes
+
+Look for new titles in signatures or explicit announcements.
+
+## 7d: Organization/Relationship Changes
+
+Look for company changes, partnership announcements, etc.
+
+## 7e: Build State Change List
+
+Compile all detected state changes before writing.
+
+---
+
+# Step 8: Check for Duplicates and Conflicts
+
+Before writing:
+- Check if already processed this source
+- Skip duplicate key facts
+- Handle conflicting information by noting both versions
+
+---
+
+# Step 9: Write Updates
+
+## 9a: Create and Update Notes
+
+**For new entities:**
+```bash
+executeCommand("write '{knowledge_folder}/People/Jennifer.md' '{content}'")
+```
+
+**For existing entities:**
+- Read current content first
+- Add activity entry at TOP (reverse chronological)
+- Update "Last seen" date
+- Add new key facts (skip duplicates)
+- Add new open items
+
+## 9b: Apply State Changes
+
+Update all fields identified in Step 7.
+
+## 9c: Update Aliases
+
+Add newly discovered name variants to Aliases field.
+
+## 9d: Writing Rules
+
+- **Always use absolute paths** with format `[[Folder/Name]]` for all links
+- Use YYYY-MM-DD format for dates
+- Be concise: one line per activity entry
+- Escape quotes properly in shell commands
+
+---
+
+# Step 10: Ensure Bidirectional Links
+
+After writing, verify links go both ways.
+
+## Absolute Link Format
+
+**IMPORTANT:** Always use absolute links:
+```markdown
+[[People/Sarah Chen]]
+[[Organizations/Acme Corp]]
+[[Projects/Acme Integration]]
+[[Topics/Security Compliance]]
+```
+
+## Bidirectional Link Rules
+
+| If you add... | Then also add... |
+|---------------|------------------|
+| Person → Organization | Organization → Person |
+| Person → Project | Project → Person |
+| Project → Organization | Organization → Project |
+| Project → Topic | Topic → Project |
+| Person → Person | Person → Person (reverse) |
+
+---
+
+# Note Templates
+
+## People
+```markdown
+# {Full Name}
+
+## Info
+**Role:** {role, inferred role, or Unknown}
+**Organization:** [[Organizations/{organization}]] or leave blank
+**Email:** {email or leave blank}
+**Aliases:** {comma-separated: first name, nicknames, email}
+**First met:** {YYYY-MM-DD}
+**Last seen:** {YYYY-MM-DD}
+
+## Summary
+{2-3 sentences: Who they are, why you know them.}
+
+## Connected to
+- [[Organizations/{Organization}]] — works at
+- [[People/{Person}]] — {relationship}
+- [[Projects/{Project}]] — {role}
+
+## Activity
+- **{YYYY-MM-DD}** ({meeting|email}): {Summary with [[Folder/Name]] links}
+
+## Key facts
+{Substantive facts only. Leave empty if none.}
+
+## Open items
+{Commitments and next steps only. Leave empty if none.}
+```
+
+## Organizations
+```markdown
+# {Organization Name}
+
+## Info
+**Type:** {company|team|institution|other}
+**Industry:** {industry or leave blank}
+**Relationship:** {customer|prospect|partner|competitor|vendor|other}
+**Domain:** {primary email domain}
+**Aliases:** {short names, abbreviations}
+**First met:** {YYYY-MM-DD}
+**Last seen:** {YYYY-MM-DD}
+
+## Summary
+{2-3 sentences: What this org is, what your relationship is.}
+
+## People
+- [[People/{Person}]] — {role}
+
+## Contacts
+{For contacts who have their own notes}
+
+## Projects
+- [[Projects/{Project}]] — {relationship}
+
+## Activity
+- **{YYYY-MM-DD}** ({meeting|email}): {Summary}
+
+## Key facts
+{Substantive facts only. Leave empty if none.}
+
+## Open items
+{Commitments and next steps only. Leave empty if none.}
+```
+
+## Projects
+```markdown
+# {Project Name}
+
+## Info
+**Type:** {deal|product|initiative|hiring|other}
+**Status:** {active|planning|on hold|completed|cancelled}
+**Started:** {YYYY-MM-DD or leave blank}
+**Last activity:** {YYYY-MM-DD}
+
+## Summary
+{2-3 sentences: What this project is, goal, current state.}
+
+## People
+- [[People/{Person}]] — {role}
+
+## Organizations
+- [[Organizations/{Org}]] — {relationship}
+
+## Related
+- [[Topics/{Topic}]] — {relationship}
+
+## Timeline
+**{YYYY-MM-DD}** ({meeting|email})
+{What happened.}
+
+## Decisions
+- **{YYYY-MM-DD}**: {Decision}
+
+## Open items
+{Commitments and next steps only.}
+
+## Key facts
+{Substantive facts only.}
+```
+
+## Topics
+```markdown
+# {Topic Name}
+
+## About
+{1-2 sentences: What this topic covers.}
+
+**Keywords:** {comma-separated}
+**Aliases:** {other references}
+**First mentioned:** {YYYY-MM-DD}
+**Last mentioned:** {YYYY-MM-DD}
+
+## Related
+- [[People/{Person}]] — {relationship}
+- [[Organizations/{Org}]] — {relationship}
+- [[Projects/{Project}]] — {relationship}
+
+## Log
+**{YYYY-MM-DD}** ({meeting|email}: {title})
+{Summary}
+
+## Decisions
+- **{YYYY-MM-DD}**: {Decision}
+
+## Open items
+{Commitments and next steps only.}
+
+## Key facts
+{Substantive facts only.}
+```
+
+---
+
+# Summary: Low Strictness Rules
+
+| Source Type | Creates Notes? | Updates Notes? | Detects State Changes? |
+|-------------|---------------|----------------|------------------------|
+| Meeting | Yes — ALL external attendees | Yes | Yes |
+| Email (any human sender) | Yes | Yes | Yes |
+| Email (automated/newsletter) | No (SKIP) | No | No |
+
+**Philosophy:** Capture broadly, filter later if needed.
+
+---
+
+# Error Handling
+
+1. **Missing data:** Leave blank or write "Unknown"
+2. **Ambiguous names:** Create note with disambiguation note
+3. **Conflicting info:** Note both versions
+4. **grep returns nothing:** Create new notes
+5. **State change unclear:** Log in activity but don't change the field
+6. **Note file malformed:** Log warning, attempt partial update
+7. **Shell command fails:** Log error, continue
+
+---
+
+# Quality Checklist
+
+Before completing, verify:
+
+**Source Type:**
+- [ ] Correctly identified as meeting or email
+- [ ] Applied low strictness rules (capture broadly)
+
+**Resolution:**
+- [ ] Extracted all name variants
+- [ ] Searched existing notes
+- [ ] Built resolution map
+- [ ] Used absolute paths `[[Folder/Name]]`
+
+**Filtering:**
+- [ ] Excluded only self and @user.domain
+- [ ] Created notes for all external contacts
+- [ ] Only skipped obvious automated/newsletters
+
+**Content Quality:**
+- [ ] Summaries describe relationship
+- [ ] Roles inferred where possible
+- [ ] Key facts are substantive
+- [ ] Open items are commitments/next steps
+
+**State Changes:**
+- [ ] Detected and applied state changes
+- [ ] Logged changes in activity
+
+**Structure:**
+- [ ] All links use `[[Folder/Name]]` format
+- [ ] Activity entries reverse chronological
+- [ ] Dates are YYYY-MM-DD
+- [ ] Bidirectional links consistent