Self-hosted, Self-Improving, Federated Everything App — dev management, agent orchestration, persistent mind, digital twin, second brain, and creative pipeline from a single dashboard over Tailscale to port the user into the OS
JavaScript ai-agentsdashboarddeveloper-toolsdigital-twinquantified-selfsecond-brainself-hostedtailscale
import { join } from 'path';
import { z } from 'zod';
import { ServerError } from '../lib/errorHandler.js';
import { createFileWriteQueue } from '../lib/fileWriteQueue.js';
import { atomicWrite, ensureDir, PATHS, readJSONFileStrict } from '../lib/fileUtils.js';
const RULES_FILE = join(PATHS.messages, 'triage-rules.json');
const MISSING_STORE = Symbol('missing triage rules');
const rulesStoreSchema = z.object({
rules: z.array(z.object({
senderPattern: z.string(),
correctedAction: z.string(),
count: z.number().int().positive().optional()
}).passthrough())
}).passthrough();
// One tail for the shared rules file: each load→mutate→save cycle must see the
// previous cycle's result or overlapping corrections/deletions drop writes.
const queueWrite = createFileWriteQueue();
async function loadRules() {
await ensureDir(PATHS.messages);
const { ok, value } = await readJSONFileStrict(RULES_FILE, MISSING_STORE, { logError: false });
if (ok && value === MISSING_STORE) return { rules: [] };
if (!ok || !rulesStoreSchema.safeParse(value).success) {
throw new ServerError('Message triage rules storage is unavailable or invalid; original data preserved', {
status: 503, code: 'MESSAGE_TRIAGE_RULES_UNAVAILABLE'
});
}
return value;
}
async function saveRules(data) {
await ensureDir(PATHS.messages);
await atomicWrite(RULES_FILE, data);
}
/**
* Get all triage rules for injection into the LLM prompt.
*/
export async function getTriageRules() {
const { rules } = await loadRules();
return rules;
}
/**
* Record a user correction: when they take a different action than the AI recommended.
* Deduplicates by pattern — if the same sender/pattern already has a rule, update it.
*/
export function recordCorrection({ from, subject, triaged, corrected }) {
return queueWrite(() => applyCorrection({ from, subject, triaged, corrected }));
}
async function applyCorrection({ from, subject, triaged, corrected }) {
const data = await loadRules();
// Build a pattern from the sender — strip email-specific parts for generalization
const senderPattern = from || 'Unknown';
// Check if we already have a rule for this sender+action combo
const existing = data.rules.find(r =>
r.senderPattern === senderPattern && r.correctedAction === corrected
);
if (existing) {
existing.count = (existing.count || 1) + 1;
existing.lastSeen = new Date().toISOString();
existing.exampleSubject = subject;
} else {
data.rules.push({
senderPattern,
exampleSubject: subject || '',
originalAction: triaged,
correctedAction: corrected,
count: 1,
createdAt: new Date().toISOString(),
lastSeen: new Date().toISOString()
});
}
await saveRules(data);
console.log(`📧 Triage rule recorded: "${senderPattern}" ${triaged} -> ${corrected}`);
}
/**
* Delete a specific rule by index.
*/
export function deleteRule(index) {
return queueWrite(async () => {
const data = await loadRules();
if (index < 0 || index >= data.rules.length) return false;
data.rules.splice(index, 1);
await saveRules(data);
return true;
});
}
/**
* Get all rules (for UI display).
*/
export async function listRules() {
const { rules } = await loadRules();
return rules;
}