@intutic/sync-daemon 2.0.0 → 2.2.0

package/dist/syncLoop.js

@@ -1,878 +0,0 @@
-/**
- * syncLoop.ts — Core bidirectional sync loop.
- *
- * The sync loop is the heartbeat of the daemon. Each iteration:
- * 1. Fetches config from the control plane (POST /api/v1/sync/config)
- * 2. If configVersion > local → writes SOP content to harness files
- * 3. Computes SHA-256 hashes of local config files
- * 4. Reports hashes + status back to the control plane
- * 5. Calls the onSync callback with the result
- * 6. Sleeps for pollIntervalMs (AbortSignal-aware)
- * 7. Repeats until signal.aborted
- *
- * Design: never crash on a single iteration failure. Errors are
- * caught per-iteration, logged, and the loop continues.
- *
- * HLD §3.14 — Real-Time State Mirroring
- * LLD #8 — Sync Daemon / CLI
- *
- * @module
- */
-import * as crypto from 'node:crypto';
-import * as path from 'node:path';
-import * as fs from 'node:fs';
-import * as os from 'node:os';
-import { newIso } from '@intutic/id';
-import { deriveEnforcementInputs } from '@intutic/shared-types';
-import { writeConfigFiles, HARNESS_FILES, applyConfigEdits, loadLocalSopEntries } from './configWriter.js';
-import { injectMcpServer } from './harness/mcpAutoWrite.js';
-import { discoverWorktrees } from './lib/gitWorktrees.js';
-import { computeFileHashes } from './hashReporter.js';
-import { loadIntegrity, saveIntegrity } from './integrityStore.js';
-import { drainHookEvents, drainReviewRequests, REVIEW_REQUESTS_LOG, parseSopConstraints, } from './harness/claudeCodeHooks.js';
-import { writeRuntimeEnv } from './lib/runtimeEnv.js';
-import { refreshPolicySnapshot } from './lib/policySnapshot.js';
-import { refreshApprovedBypasses } from './lib/approvedBypasses.js';
-import { refreshEgressPolicy } from './lib/egressPolicy.js';
-import { refreshDecisionsDigest } from './lib/decisionsDigest.js';
-import { runComplianceProbes } from './lib/complianceProbes.js';
-import { startWatcher } from './watcher/driftWatcher.js';
-import { shouldCaptureThisIteration, captureAndUpload } from './configReader.js';
-import { watch } from 'chokidar';
-// Named import, not default: this package is ESM ("type": "module") and under
-// Node16 resolution the default import of ioredis's CJS typings resolves to the
-// module namespace, which is neither constructable nor usable as a type. The
-// named export is the class itself. (The old `new (Redis as any)(...)` cast
-// silenced exactly this, at the cost of also erasing the instance type.)
-import { Redis } from 'ioredis';
-import { TrajectoryMonitor } from './trajectoryMonitor.js';
-import { collectAgentReport, reportAgent, fetchLocalProxyInstanceId } from './agentReporter.js';
-import { startHarnessSession, endAllOpenSessions } from './sessionReporter.js';
-/** Narrow an unknown thrown value to a printable message. */
-function errorMessage(err) {
-    return err instanceof Error ? err.message : String(err);
-}
-/**
- * Structural guard for trace events arriving over the `trace:live:*` Valkey
- * channel. The payload is JSON off the wire, so it is genuinely unknown here;
- * `handleTraceEvent` indexes `sessionId`/`workspaceId` immediately and a
- * malformed publish would otherwise fault inside the buffer bookkeeping.
- */
-function isTraceEvent(value) {
-    if (typeof value !== 'object' || value === null)
-        return false;
-    const e = value;
-    return typeof e['sessionId'] === 'string' && typeof e['workspaceId'] === 'string';
-}
-/**
- * Turns a non-clean `scanSkillContent` result — already computed as part of
- * `collectAgentReport`'s `facets.skills` — into a `skill_flagged` hook event
- * (TD-358).
- *
- * Deliberately reuses the report the daemon already built for this cycle
- * rather than re-scanning: `agentReporter.ts`'s `collectSkills` runs
- * `scanSkillContent` on every `.agents/skills/**\/SKILL.md` as part of
- * building `facets.skills` already, so this only turns a result the daemon
- * already has into a visible signal — it does not add a second scan pass.
- *
- * Appends directly to `.intutic/events/hook-events.jsonl`, the SAME file
- * every emitted PreToolUse gate's `log_event`/`logEvent` writes to and the
- * SAME mechanism `onDriftDetected` below already uses for `config_tamper` —
- * not a new event pipeline. `drainHookEvents` (`claudeCodeHooks.ts`) picks
- * this line up on its next pass exactly like a gate-written one, because it
- * only ever reads the file; it does not care who appended a given line.
- *
- * Report-only, per `scanSkillContent`'s own doc comment on the unmeasured
- * false-positive rate against real, benign skill markdown: this NEVER
- * blocks, mutates, or removes a skill. `hookEvents.ts`'s
- * `HookEventSchema.event` enum (control plane) counts `skill_flagged`
- * exactly like `tool_flagged` — advisory telemetry, not an incident.
- *
- * A failure to append is caught and warned, never thrown: a governance
- * telemetry write must not be able to abort the sync cycle it rides in on.
- */
-export function emitSkillFlaggedEvents(opts) {
-    const eventsLog = path.join(opts.workspaceRoot, '.intutic', 'events', 'hook-events.jsonl');
-    for (const skill of opts.skills) {
-        if (!skill.scanned || skill.clean)
-            continue;
-        const key = `${skill.source}:${skill.name}`;
-        if (opts.alreadyEmitted.has(key))
-            continue;
-        opts.alreadyEmitted.add(key);
-        try {
-            const ts = newIso();
-            const entry = JSON.stringify({
-                event: 'skill_flagged',
-                toolName: `skill:${skill.name}`,
-                reason: `scanSkillContent found ${skill.findingsCount} finding(s) in ` +
-                    `${skill.source}/${skill.name}/SKILL.md — advisory only, see skillScan.ts`,
-                workspaceId: opts.workspaceId,
-                harnessType: opts.harnessType,
-                timestamp: ts,
-                incidentId: crypto
-                    .createHash('sha1')
-                    .update(ts + skill.name + opts.workspaceId)
-                    .digest('hex')
-                    .slice(0, 16),
-                filePath: `${skill.source}/${skill.name}/SKILL.md`,
-            }) + '\n';
-            fs.appendFileSync(eventsLog, entry, { flag: 'a' });
-        }
-        catch (err) {
-            console.warn('[sync-daemon] failed to write skill_flagged event (non-fatal):', err);
-        }
-    }
-}
-// ─── Default constants ───────────────────────────────────────────────
-const DEFAULT_POLL_INTERVAL_MS = 30_000;
-/**
- * Best-effort reachability probe for the control plane, modelled on
- * `agentReporter.fetchEgressStatus()`'s bounded-fetch-and-swallow shape.
- *
- * The trajectory monitor used to start on `VALKEY_URL || apiKey` alone —
- * never checking `controlPlaneUrl` at all, so an unset URL produced
- * `undefined/api/v1/...` on every submit attempt. This is the missing check:
- * a monitor that will publish its summaries to a URL that does not resolve
- * (or does not answer) should not start believing it has somewhere to send
- * them.
- */
-async function controlPlaneIsReachable(controlPlaneUrl) {
-    if (!controlPlaneUrl)
-        return false;
-    try {
-        const res = await fetch(`${controlPlaneUrl}/healthz`, { signal: AbortSignal.timeout(2000) });
-        return res.ok;
-    }
-    catch {
-        return false;
-    }
-}
-/**
- * The `review_before:` tokens the snapshot compiles into hold rules: from the
- * synced SOPs' front matter and JSON blocks, the workspace settings, and the
- * local `.intutic/sops`. The same parse the Claude Code writer runs for its
- * pattern blacklist, so the two cannot disagree about what is held.
- */
-export async function localHoldTokensFor(workspaceRoot, sops, settings, harnesses = []) {
-    const local = await loadLocalSopEntries(workspaceRoot, harnesses);
-    return parseSopConstraints([...sops, ...local], settings).reviewBefore;
-}
-/**
- * Refresh the three gate-side caches every hook reads: the policy snapshot
- * (`~/.intutic/hooks/policy-snapshot.rules`), the approved review-hold
- * bypasses and the central egress policy. Called on EVERY sync cycle by both
- * loops — this one and `intutic connect`'s — never only when the config
- * version moved: policy changes without the version changing, and before
- * this helper existed `connect` refreshed once at startup and never again, so
- * a guardrail promoted mid-session reached a connected machine only on
- * restart (TD-488). None of the three throw; the return says which of them
- * landed.
- */
-export async function refreshGateCaches(opts) {
-    const snapshot = (await refreshPolicySnapshot(opts)) !== null;
-    const bypasses = (await refreshApprovedBypasses(opts)) !== null;
-    const egress = (await refreshEgressPolicy(opts)) !== null;
-    return { snapshot, bypasses, egress };
-}
-export async function startSyncLoop(options) {
-    const { controlPlaneUrl, apiKey, workspaceId, workspaceRoot, pollIntervalMs = DEFAULT_POLL_INTERVAL_MS, onSync, signal, } = options;
-    // Step -1: Start the Trajectory Monitor & Valkey Subscriber
-    let trajectoryMonitor = null;
-    let trajectorySubscriber = null;
-    const valkeyUrl = process.env.VALKEY_URL ?? 'redis://127.0.0.1:6379';
-    if ((process.env.VALKEY_URL || apiKey) && (await controlPlaneIsReachable(controlPlaneUrl))) {
-        trajectoryMonitor = new TrajectoryMonitor({
-            valkeyUrl,
-            controlPlaneUrl,
-            apiKey,
-            windowMs: 300_000,
-            submitIntervalMs: 60_000,
-        });
-        try {
-            await trajectoryMonitor.start();
-            trajectorySubscriber = new Redis(valkeyUrl);
-            await trajectorySubscriber.psubscribe('trace:live:*');
-            trajectorySubscriber.on('pmessage', (_pattern, _channel, message) => {
-                try {
-                    const event = JSON.parse(message);
-                    if (!isTraceEvent(event)) {
-                        console.warn('[sync-daemon] Discarding malformed trajectory trace event');
-                        return;
-                    }
-                    trajectoryMonitor?.handleTraceEvent(event);
-                }
-                catch (err) {
-                    // Non-blocking warning
-                    console.warn('[sync-daemon] Failed to parse trajectory trace event:', err);
-                }
-            });
-            console.log('[sync-daemon] Trajectory monitor & subscriber started successfully');
-        }
-        catch (err) {
-            console.warn('[sync-daemon] Could not start trajectory monitor:', err);
-        }
-    }
-    // Tracking latest settings for use in drift handler
-    let latestSettings;
-    let latestSops = [];
-    let latestProxyUrl = '';
-    let latestHarnesses = [];
-    let latestGovernanceInputs = {};
-    // Step 0: Write runtime env file (hook scripts source this for credentials)
-    // Runs once at startup and then again on every iteration.
-    try {
-        // The Valkey URL rides along only when this loop is itself using one
-        // (the trajectory monitor connected): the MCP proxies share their session
-        // window through it (Wave 5.3, TD-437).
-        await writeRuntimeEnv({ controlPlaneUrl, apiKey, workspaceId, valkeyUrl: trajectoryMonitor ? valkeyUrl : undefined });
-    }
-    catch (err) {
-        console.warn('[sync-daemon] Could not write runtime env file (non-fatal):', err);
-    }
-    // Step 0b–0d: Refresh the three gate-side caches every hook reads —
-    // policy snapshot, approved review_before bypasses, central egress policy.
-    //
-    // Deliberately here beside writeRuntimeEnv, and deliberately NOT inside
-    // writeConfigFiles: that path is gated on the config version, so a snapshot
-    // written there would go stale by construction on every cycle where the
-    // version did not move. Policy changes without the config version changing,
-    // and a bypass an operator just approved has to reach the gate before the
-    // developer's very next retry. One helper, shared with `intutic connect`'s
-    // own loop, so the two cannot drift apart again (TD-488).
-    await refreshGateCaches({
-        controlPlaneUrl, apiKey, workspaceId,
-        localHoldTokens: await localHoldTokensFor(workspaceRoot, [], undefined),
-    });
-    // Try to sync offline traces back to PostgreSQL
-    try {
-        await syncOfflineTraces(controlPlaneUrl, apiKey);
-    }
-    catch (err) {
-        console.warn('[sync-daemon] Could not sync offline traces (non-fatal):', err);
-    }
-    // Set up FSEvents-driven hook event drain (replaces 30s fixed poll for drain step).
-    // Drains immediately when the hook-events.jsonl file changes (written by hook scripts).
-    // Falls back to a 60s safety-net interval for NFS/Docker mounts where inotify may not fire.
-    const hookEventsLog = `${workspaceRoot}/.intutic/events/hook-events.jsonl`;
-    // The review-hold log, watched alongside it. Two named files rather than the
-    // directory: a directory watch would also fire on `.rejected.jsonl`, which the
-    // drain itself writes, and a drain that retriggers itself is a loop.
-    const reviewHoldsLog = `${workspaceRoot}/${REVIEW_REQUESTS_LOG}`;
-    let fsWatcher = null;
-    let drainSafetyTimer = null;
-    const runDrain = async () => {
-        try {
-            const drained = await drainHookEvents(workspaceRoot, controlPlaneUrl, apiKey);
-            if (drained > 0) {
-                console.log(`[sync-daemon] Drained ${drained} hook governance events to control plane`);
-            }
-        }
-        catch (err) {
-            console.warn('[sync-daemon] Hook event drain error (non-fatal):', err);
-        }
-        // Separate try: a failure draining events must not skip the holds. They go
-        // to different endpoints and one being down says nothing about the other.
-        try {
-            const held = await drainReviewRequests(workspaceRoot, controlPlaneUrl, apiKey);
-            if (held > 0) {
-                console.log(`[sync-daemon] Drained ${held} review hold(s) to control plane`);
-            }
-        }
-        catch (err) {
-            console.warn('[sync-daemon] Review hold drain error (non-fatal):', err);
-        }
-    };
-    try {
-        fsWatcher = watch([hookEventsLog, reviewHoldsLog], { ignoreInitial: true, persistent: false });
-        fsWatcher.on('change', runDrain);
-        fsWatcher.on('add', runDrain);
-    }
-    catch {
-        // chokidar unavailable — rely on the 60s safety-net only
-    }
-    // 60-second safety-net drain poll (catches events missed by FSEvents on NFS/Docker)
-    drainSafetyTimer = setInterval(runDrain, 60_000);
-    signal?.addEventListener('abort', () => {
-        fsWatcher?.close();
-        if (drainSafetyTimer)
-            clearInterval(drainSafetyTimer);
-        trajectoryMonitor?.stop();
-        trajectorySubscriber?.disconnect();
-    }, { once: true });
-    // WS-5 — Drift watcher with onDriftDetected callback
-    // Wired once after first successful sync (when we have harness list).
-    // The callback fires a config_tamper governance event and immediately rewrites config.
-    let driftWatcher = null;
-    const onDriftDetected = async (changedPath) => {
-        // 1. Immediately rewrite config (if settings allow)
-        if (latestSettings?.bypassEnforcementTier !== 'alert-only') {
-            try {
-                await writeConfigFiles(workspaceRoot, latestSops, latestProxyUrl, latestHarnesses, workspaceId, latestSettings?.bypassEnforcementTier);
-            }
-            catch (err) {
-                console.warn('[sync-daemon] onDriftDetected: writeConfigFiles failed (non-fatal):', err);
-            }
-        }
-        // 2. Append config_tamper event to hook-events JSONL
-        try {
-            const tamperEntry = JSON.stringify({
-                event: 'config_tamper',
-                toolName: 'config_file',
-                reason: 'Harness config file modified outside sync-daemon',
-                workspaceId,
-                filePath: changedPath,
-                timestamp: new Date().toISOString(),
-                incidentId: crypto.createHash('sha1').update(changedPath + Date.now()).digest('hex').slice(0, 16),
-            }) + '\n';
-            const hookEventsJsonl = path.join(os.homedir(), '.intutic', 'events', 'hook-events.jsonl');
-            fs.appendFileSync(hookEventsJsonl, tamperEntry, { flag: 'a' });
-        }
-        catch (err) {
-            console.warn('[sync-daemon] onDriftDetected: failed to write tamper event (non-fatal):', err);
-        }
-    };
-    let iterationCount = 0;
-    while (!signal?.aborted) {
-        try {
-            // Refresh runtime env on every iteration (key rotation)
-            const result = await runSyncIteration({
-                controlPlaneUrl,
-                apiKey,
-                workspaceId,
-                workspaceRoot,
-                onSync,
-            });
-            // Update cached state for drift handler
-            if (result) {
-                latestSettings = result.settings;
-                latestSops = result.sops ?? [];
-                latestProxyUrl = result.proxyUrl ?? '';
-                latestHarnesses = result.harnesses ?? [];
-                latestGovernanceInputs = result.harnessGovernanceInputs ?? {};
-                await writeRuntimeEnv({
-                    controlPlaneUrl,
-                    apiKey,
-                    workspaceId,
-                    mcpProxyFailBehavior: result.settings?.mcpProxyFailBehavior,
-                    mcpProxyMode: result.settings?.mcpProxyMode,
-                    bypassEnforcementTier: result.settings?.bypassEnforcementTier,
-                    valkeyUrl: trajectoryMonitor ? valkeyUrl : undefined,
-                });
-                // Same reasoning as Step 0b–0d: policy rides the sync cycle, not the
-                // config version.
-                await refreshGateCaches({
-                    controlPlaneUrl, apiKey, workspaceId,
-                    localHoldTokens: await localHoldTokensFor(workspaceRoot, latestSops, latestSettings, latestHarnesses),
-                });
-                // Governed decisions log — opt-in only (WorkspaceSettings.decisionsLogEnabled,
-                // default off: a growing context file is token spend the product must not
-                // silently impose on every workspace). No Step-0-equivalent unconditional
-                // call before the loop, unlike egress/policy/bypasses above: this gate can
-                // only be evaluated once `result.settings` is known, which is not yet true
-                // before the first iteration completes. When disabled, this is simply not
-                // called — existing files (if any, from before the workspace disabled the
-                // feature) are left as last written rather than force-deleted or overwritten
-                // with a stale/wrong "disabled" digest.
-                if (result.settings?.decisionsLogEnabled) {
-                    await refreshDecisionsDigest({
-                        controlPlaneUrl,
-                        apiKey,
-                        workspaceId,
-                        workspaceRoot,
-                        harnesses: latestHarnesses,
-                    });
-                }
-            }
-            // Start the drift watcher on first successful sync (once harnesses are known)
-            if (!driftWatcher && latestHarnesses.length > 0) {
-                try {
-                    driftWatcher = startWatcher(workspaceRoot, latestHarnesses, onDriftDetected);
-                    signal?.addEventListener('abort', () => driftWatcher?.stop(), { once: true });
-                }
-                catch (err) {
-                    console.warn('[sync-daemon] Could not start drift watcher (non-fatal):', err);
-                }
-            }
-            // Run network compliance probes to check for proxy bypass
-            try {
-                const probeResults = await runComplianceProbes(workspaceId);
-                for (const res of probeResults) {
-                    if (!res.contained && res.incident) {
-                        const entry = JSON.stringify(res.incident) + '\n';
-                        fs.appendFileSync(hookEventsLog, entry, { flag: 'a' });
-                    }
-                }
-            }
-            catch (err) {
-                console.warn('[sync-daemon] Compliance probes failed (non-fatal):', err);
-            }
-            // 4b. Capture harness configs back to control plane (every Nth iteration)
-            if (shouldCaptureThisIteration(iterationCount)) {
-                try {
-                    await captureAndUpload(controlPlaneUrl, apiKey, workspaceId, workspaceRoot, latestHarnesses, latestGovernanceInputs);
-                }
-                catch (err) {
-                    console.warn('[sync-daemon] Config capture failed (non-fatal):', err);
-                }
-            }
-            iterationCount++;
-        }
-        catch (err) {
-            console.error('[sync-daemon] iteration error:', err);
-        }
-        await sleep(pollIntervalMs, signal);
-    }
-    // Cleanup on exit
-    fsWatcher?.close();
-    driftWatcher?.stop();
-    if (drainSafetyTimer)
-        clearInterval(drainSafetyTimer);
-    // Mark this run's harness sessions ended so the dashboard's live list
-    // reflects reality.
-    await endAllOpenSessions(controlPlaneUrl, apiKey);
-}
-/**
- * Execute a single sync iteration:
- * 1. Fetch config from control plane
- * 2. Write config files if version is newer
- * 3. Compute file hashes
- * 4. Report status back to control plane
- * 5. Update local integrity store
- * 6. Call onSync callback
- *
- * Exported (in addition to being used internally by `startSyncLoop`'s
- * polling loop) as a test seam — it is the unit that exercises exactly one
- * cycle's worth of work, including the injectMcpServer call (step 3c),
- * without the surrounding while-loop's extra one-time/Nth-iteration network
- * calls (policy snapshot, approved bypasses, egress policy, compliance
- * probes, drift watcher, decisions digest, config capture).
- */
-export async function runSyncIteration(ctx) {
-    const { controlPlaneUrl, apiKey, workspaceId, workspaceRoot, onSync } = ctx;
-    // 1. Fetch config from control plane
-    const config = await fetchConfig(controlPlaneUrl, apiKey, workspaceId);
-    // 2. Load local integrity store
-    const integrity = await loadIntegrity(workspaceRoot);
-    const localVersion = integrity?.configVersion ?? -1;
-    let sopsWritten = 0;
-    // 3. If remote version is newer → write config files
-    if (config.configVersion > localVersion) {
-        const harnesses = extractHarnesses(config);
-        const writeResult = await writeConfigFiles(workspaceRoot, config.sops, config.proxyUrl, harnesses);
-        sopsWritten = writeResult.filesWritten.length;
-    }
-    // 3b. Apply custom config edits from SkillOpt suggestions
-    if (config.appliedEdits && config.appliedEdits.length > 0) {
-        const forceApply = sopsWritten > 0;
-        const appliedSuggestionsPath = path.join(workspaceRoot, '.intutic', 'applied-suggestions.json');
-        let localAppliedIds = [];
-        try {
-            if (fs.existsSync(appliedSuggestionsPath)) {
-                localAppliedIds = JSON.parse(fs.readFileSync(appliedSuggestionsPath, 'utf-8'));
-            }
-        }
-        catch {
-            // ignore
-        }
-        const toApply = forceApply
-            ? config.appliedEdits
-            : config.appliedEdits.filter(edit => !localAppliedIds.includes(edit.suggestionId));
-        if (toApply.length > 0) {
-            try {
-                const results = await applyConfigEdits(workspaceRoot, toApply, config.settings.bypassEnforcementTier);
-                // Ack every result back to control-plane (TD-349). Best-effort per
-                // suggestion — one failed ack must not block the others or abort
-                // the sync cycle; see reportApplyResult's own non-fatal handling.
-                for (const result of results) {
-                    await reportApplyResult(controlPlaneUrl, apiKey, result.suggestionId, result);
-                }
-                // Only remember ids that fully landed (every operation for that
-                // suggestion applied). Recording every ATTEMPTED id regardless of
-                // outcome (the old behavior) meant a fuzzy-match miss was silently
-                // never retried — the id was in applied-suggestions.json forever,
-                // and forceApply is the only other thing that re-attempts an id
-                // already in this file. A failed edit must remain retryable.
-                const landedIds = results.filter((r) => r.ok).map((r) => r.suggestionId);
-                const newlyAppliedIds = Array.from(new Set([...localAppliedIds, ...landedIds]));
-                try {
-                    fs.mkdirSync(path.dirname(appliedSuggestionsPath), { recursive: true });
-                    fs.writeFileSync(appliedSuggestionsPath, JSON.stringify(newlyAppliedIds, null, 2), 'utf-8');
-                }
-                catch {
-                    // ignore
-                }
-            }
-            catch (err) {
-                console.warn('[sync-daemon] Failed to apply custom SkillOpt config edits:', err);
-            }
-        }
-    }
-    // 3c. Proxy-wrap MCP servers across all supported harness configs. This is
-    // a continuous invariant, not a one-shot done only at `connect` time: a
-    // server a user adds to a harness config after their first `connect` would
-    // otherwise never get wrapped and stay invisible to governance. Safe to run
-    // every cycle because mcpAutoWrite's writeJsonFile is write-if-changed — an
-    // already-wrapped, unchanged config writes zero bytes. Non-fatal, matching
-    // step 3b above: a failure here must not abort the sync cycle.
-    try {
-        await injectMcpServer(workspaceRoot, workspaceId);
-    }
-    catch (err) {
-        console.warn('[sync-daemon] injectMcpServer failed (non-fatal):', err);
-    }
-    // 3d. Worktree propagation (O1). `workspaceRoot` above only ever covers the
-    // ONE checkout this daemon was pointed at — a `git worktree add`-created
-    // sibling checkout has its own independent working tree, and every
-    // project-tier governance file this product writes is untracked, so none
-    // of it exists in a worktree by default (see gitWorktrees.ts's module doc
-    // for the full mechanics and why Xirp is what surfaced this). Discovering
-    // worktrees fresh every cycle — never caching the list across
-    // iterations — is what makes worktree REMOVAL fall out for free: a
-    // worktree `git worktree remove`d since the last cycle simply stops being
-    // returned here, with no stale-entry cleanup logic required.
-    //
-    // Config-content writes reuse the SAME `config.configVersion > localVersion`
-    // gate step 3 already used — every worktree of one repo shares the one
-    // control-plane config, so re-deriving per-worktree versioning would just
-    // reproduce a decision already made above. `injectMcpServer` runs
-    // unconditionally per worktree, same as it does for the main root: it is
-    // write-if-changed internally (mcpAutoWrite's writeJsonFile), so an
-    // unchanged worktree costs a few file reads and no writes.
-    try {
-        const worktrees = await discoverWorktrees(workspaceRoot);
-        for (const worktreeRoot of worktrees) {
-            if (worktreeRoot === workspaceRoot)
-                continue; // already covered above
-            try {
-                if (config.configVersion > localVersion) {
-                    await writeConfigFiles(worktreeRoot, config.sops, config.proxyUrl, extractHarnesses(config));
-                }
-                await injectMcpServer(worktreeRoot, workspaceId);
-            }
-            catch (err) {
-                console.warn(`[sync-daemon] worktree propagation failed for ${worktreeRoot} (non-fatal):`, err);
-            }
-        }
-    }
-    catch (err) {
-        console.warn('[sync-daemon] discoverWorktrees failed (non-fatal):', err);
-    }
-    // 4. Compute file hashes for drift detection
-    const harnesses = extractHarnesses(config);
-    const canonicalHashes = buildCanonicalHashMap(config);
-    const fileHashes = await computeFileHashes(workspaceRoot, harnesses, canonicalHashes);
-    const driftCount = fileHashes.filter((f) => f.drifted).length;
-    // 5. Report status back to control plane per active harness
-    for (const harness of harnesses) {
-        const filename = HARNESS_FILES[harness];
-        if (!filename)
-            continue;
-        const harnessHashes = fileHashes.filter((f) => f.filePath === filename);
-        if (harnessHashes.length > 0) {
-            await reportStatus(controlPlaneUrl, apiKey, workspaceId, harness, harnessHashes);
-        }
-    }
-    // 5b. Register each harness as a durable agent with its facets, so the
-    // dashboard agent graph + posture ring stay live. Best-effort: a failed
-    // report never blocks the sync loop.
-    //
-    // Skills scanned this cycle whose finding has already been emitted, keyed
-    // `source:name`. `collectAgentReport` recomputes the same workspace-wide
-    // skill scan once per harness in this loop — a skill is not harness-
-    // specific — so without this a workspace with two active harnesses would
-    // double-emit every flagged skill every cycle.
-    const skillFlaggedThisCycle = new Set();
-    // Per-harness governance-coverage enforcement signals for this cycle,
-    // derived from the same facets `collectAgentReport` just computed — see
-    // `SyncResult.harnessGovernanceInputs`'s doc comment. Uses the shared
-    // `deriveEnforcementInputs` (`@intutic/shared-types`, TD-443) so this
-    // mapping cannot drift from `harnessGradeSweep.ts`'s control-plane-side
-    // sweep the way the two hand-kept copies previously did — safe to depend
-    // on since `@intutic/shared-types` is a leaf package, not
-    // `services/control-plane` itself, so this publicly-mirrored module still
-    // never imports from the enterprise-only service.
-    const harnessGovernanceInputs = {};
-    // The local proxy's instance id, read once per iteration (not per harness):
-    // step 5c registers each harness's git/task context onto that process's own
-    // session row (TD-231, Wave 5.6). `null` — no local proxy, or a shared
-    // gateway — keeps the `ses_` fallback. Skipped when there is no harness to
-    // report, so a workspace with no SOP targets never probes the proxy.
-    const proxyInstanceId = harnesses.length > 0 ? await fetchLocalProxyInstanceId() : null;
-    for (const harness of harnesses) {
-        const filename = HARNESS_FILES[harness];
-        const configSynced = filename
-            ? !fileHashes.some((f) => f.filePath === filename && f.drifted)
-            : false;
-        const report = await collectAgentReport({
-            workspaceRoot,
-            harnessType: harness,
-            configSynced,
-            dlpEnabled: true, // the daemon-managed proxy ships DLP on by default
-            policyEnforced: Boolean(controlPlaneUrl),
-            allowLocalVaults: config.settings?.allowLocalMemoryVaults,
-        });
-        await reportAgent(controlPlaneUrl, apiKey, workspaceId, report);
-        harnessGovernanceInputs[harness] = deriveEnforcementInputs(report.facets);
-        // 5b-i. Skill-content scan findings, surfaced as `skill_flagged` events
-        // (TD-358). See `emitSkillFlaggedEvents`'s own doc comment for why this
-        // rides the existing hook-events drain rather than a new pipeline.
-        emitSkillFlaggedEvents({
-            workspaceRoot,
-            workspaceId,
-            harnessType: harness,
-            skills: report.facets.skills,
-            alreadyEmitted: skillFlaggedThisCycle,
-        });
-        // 5c. Report the harness session (once per harness per proxy process —
-        // the reporter dedupes). This is what switches on branch/commit capture
-        // and the task-context cascade. With a local proxy the context lands on
-        // the proxy's own row, the one its traces are filed under; without one a
-        // `ses_` row is opened (see sessionReporter.ts).
-        await startHarnessSession({
-            controlPlaneUrl,
-            apiKey,
-            workspaceId,
-            harnessType: harness,
-            workspaceRoot,
-            ...(proxyInstanceId ? { proxyInstanceId } : {}),
-        });
-    }
-    // 6. Update local integrity store
-    const newFiles = {};
-    for (const fh of fileHashes) {
-        if (fh.localHash) {
-            newFiles[fh.filePath] = fh.localHash;
-        }
-    }
-    await saveIntegrity(workspaceRoot, {
-        lastSyncAt: newIso(),
-        configVersion: config.configVersion,
-        files: newFiles,
-    });
-    // 6b. ContextGraph indexing — REMOVED.
-    //
-    // This block scanned the whole workspace every sync cycle and POSTed the
-    // delta to /api/v1/context/sync, an ingest stripped from the control plane
-    // with the Context Graph feature (commit 9cfc0200). Worse than dead: the
-    // integrity file is only written on a 2xx, so every cycle re-scanned and
-    // re-posted the same delta forever on every developer machine.
-    // 6c. Hook event drain is now FSEvents-driven (see startSyncLoop).
-    //     No per-iteration drain needed here; the chokidar watcher handles it.
-    // 7. Invoke onSync callback
-    const result = {
-        configVersion: config.configVersion,
-        sopsWritten,
-        driftCount,
-        timestamp: newIso(),
-        settings: config.settings,
-        sops: config.sops,
-        proxyUrl: config.proxyUrl,
-        harnesses,
-        harnessGovernanceInputs,
-    };
-    onSync?.(result);
-    return result;
-}
-// ─── HTTP helpers ────────────────────────────────────────────────────
-/**
- * Fetch the sync config from the control plane.
- *
- * POST /api/v1/sync/config
- */
-async function fetchConfig(controlPlaneUrl, apiKey, workspaceId) {
-    const url = `${controlPlaneUrl}/api/v1/sync/config`;
-    const res = await fetch(url, {
-        method: 'POST',
-        headers: {
-            'Content-Type': 'application/json',
-            Authorization: `Bearer ${apiKey}`,
-        },
-        body: JSON.stringify({ workspaceId }),
-        signal: AbortSignal.timeout(15_000), // 15s request timeout
-    });
-    if (!res.ok) {
-        throw new Error(`[sync-daemon] fetchConfig failed: ${res.status} ${res.statusText}`);
-    }
-    return (await res.json());
-}
-/**
- * Report file hashes and status back to the control plane.
- *
- * POST /api/v1/sync/sop-hash
- */
-async function reportStatus(controlPlaneUrl, apiKey, workspaceId, harnessType, files) {
-    const url = `${controlPlaneUrl}/api/v1/sync/sop-hash`;
-    const res = await fetch(url, {
-        method: 'POST',
-        headers: {
-            'Content-Type': 'application/json',
-            Authorization: `Bearer ${apiKey}`,
-        },
-        body: JSON.stringify({
-            workspaceId,
-            harnessType,
-            files,
-            reportedAt: newIso(),
-        }),
-        signal: AbortSignal.timeout(15_000),
-    });
-    if (!res.ok) {
-        // Non-fatal: log but don't throw (control plane might be temporarily down)
-        console.warn(`[sync-daemon] reportStatus failed for ${harnessType}: ${res.status} ${res.statusText}`);
-    }
-}
-/**
- * Ack a SkillOpt config-edit apply outcome back to the control plane
- * (TD-349) — POST /api/v1/skillopt/:suggestionId/apply-result.
- *
- * Same HTTP call shape as `reportGovernanceCoverageSnapshot`
- * (configReader.ts): `fetch` + `AbortSignal.timeout(10_000)`, fully
- * non-fatal. This ack closing the loop is important, but it must never be
- * allowed to abort a sync cycle that otherwise succeeded — a dropped ack
- * just means the suggestion re-syncs and gets acked again next cycle
- * (getSyncConfig re-pushes 'applied' and 'apply_unconfirmed' rows alike).
- */
-async function reportApplyResult(controlPlaneUrl, apiKey, suggestionId, result) {
-    try {
-        const res = await fetch(`${controlPlaneUrl}/api/v1/skillopt/${suggestionId}/apply-result`, {
-            method: 'POST',
-            headers: {
-                'Content-Type': 'application/json',
-                Authorization: `Bearer ${apiKey}`,
-            },
-            body: JSON.stringify({
-                ok: result.ok,
-                perOperation: result.perOperation,
-            }),
-            signal: AbortSignal.timeout(10_000),
-        });
-        if (!res.ok) {
-            const body = await res.text().catch(() => '');
-            console.warn(`[sync-daemon] reportApplyResult failed for ${suggestionId} (${res.status}): ${body}`);
-        }
-    }
-    catch (err) {
-        console.warn(`[sync-daemon] reportApplyResult request failed for ${suggestionId}:`, err);
-    }
-}
-// ─── Helpers ─────────────────────────────────────────────────────────
-/**
- * Extract unique harness types from a sync config payload.
- */
-function extractHarnesses(config) {
-    const set = new Set();
-    for (const sop of config.sops) {
-        for (const h of sop.harnessTargets) {
-            set.add(h);
-        }
-    }
-    return [...set];
-}
-/**
- * Build a filePath → canonical hash map from the config's SOPs.
- * Uses the first canonical hash found for each file path.
- */
-function buildCanonicalHashMap(config) {
-    const map = {};
-    for (const sop of config.sops) {
-        // Each SOP has a single contentHash; in a multi-harness scenario
-        // the same hash applies to the content source (not the formatted output).
-        // For per-file canonical hashes, the control plane would need to
-        // provide them per-harness. For now we use sop.contentHash as a
-        // proxy keyed by sopId.
-        map[sop.sopId] = sop.contentHash;
-    }
-    return map;
-}
-/**
- * Sleep for the given duration with AbortSignal support.
- *
- * Resolves immediately if the signal is already aborted.
- * Otherwise, resolves after `ms` milliseconds or when the
- * signal fires, whichever comes first.
- */
-function sleep(ms, signal) {
-    if (signal?.aborted)
-        return Promise.resolve();
-    return new Promise((resolve) => {
-        const timer = setTimeout(() => {
-            signal?.removeEventListener('abort', onAbort);
-            resolve();
-        }, ms);
-        function onAbort() {
-            clearTimeout(timer);
-            resolve();
-        }
-        signal?.addEventListener('abort', onAbort, { once: true });
-    });
-}
-export async function syncOfflineTraces(controlPlaneUrl, apiKey) {
-    const logsDir = path.join(os.homedir(), '.intutic', 'logs');
-    if (!fs.existsSync(logsDir))
-        return;
-    try {
-        const files = fs.readdirSync(logsDir);
-        const traceFiles = files.filter(f => f.startsWith('traces-') && f.endsWith('.jsonl'));
-        if (traceFiles.length === 0)
-            return;
-        for (const file of traceFiles) {
-            const originalPath = path.join(logsDir, file);
-            const syncingPath = originalPath + '.syncing';
-            // Rename to avoid write race conditions with the Rust proxy
-            try {
-                fs.renameSync(originalPath, syncingPath);
-            }
-            catch (renameErr) {
-                console.error(`[sync-daemon] Failed to lock/rename file ${file}:`, errorMessage(renameErr));
-                continue;
-            }
-            try {
-                const raw = fs.readFileSync(syncingPath, 'utf-8');
-                const lines = raw.split('\n').map(l => l.trim()).filter(Boolean);
-                if (lines.length === 0) {
-                    // Empty file, just clean it up
-                    fs.unlinkSync(syncingPath);
-                    continue;
-                }
-                const traces = lines.map(line => JSON.parse(line));
-                console.log(`[sync-daemon] Found ${traces.length} offline traces to sync back in ${file}.`);
-                // Batch in groups of 100
-                const batchSize = 100;
-                for (let i = 0; i < traces.length; i += batchSize) {
-                    const batch = traces.slice(i, i + batchSize);
-                    const res = await fetch(`${controlPlaneUrl}/api/v1/traces/sync-back`, {
-                        method: 'POST',
-                        headers: {
-                            'Content-Type': 'application/json',
-                            Authorization: `Bearer ${apiKey}`,
-                        },
-                        body: JSON.stringify({ traces: batch }),
-                    });
-                    if (!res.ok) {
-                        throw new Error(`Sync-back API returned status ${res.status}`);
-                    }
-                }
-                // Successfully synced this sharded file, delete it
-                fs.unlinkSync(syncingPath);
-                console.log(`[sync-daemon] Successfully synced ${traces.length} offline traces back from ${file}.`);
-            }
-            catch (err) {
-                console.error(`[sync-daemon] Failed to sync offline traces back from ${file}:`, errorMessage(err));
-                // Revert rename on failure to allow retry on next cycle
-                try {
-                    if (fs.existsSync(syncingPath)) {
-                        fs.renameSync(syncingPath, originalPath);
-                    }
-                }
-                catch (revertErr) {
-                    console.error(`[sync-daemon] Failed to revert rename for ${file}:`, errorMessage(revertErr));
-                }
-            }
-        }
-    }
-    catch (err) {
-        console.error(`[sync-daemon] Failed to scan offline traces directory:`, errorMessage(err));
-    }
-}
-//# sourceMappingURL=syncLoop.js.map