atris 2.6.3 → 3.0.0

package/commands/publish.js

@@ -0,0 +1,129 @@
+const fs = require('fs');
+const path = require('path');
+const os = require('os');
+const { loadCredentials } = require('../utils/auth');
+const { apiRequestJson } = require('../utils/api');
+
+async function publishAtris() {
+  if (process.argv[3] === '--help') {
+    console.log('Usage: atris publish [--name <name>] [--description <desc>]');
+    console.log('');
+    console.log('  atris publish                         Publish current workspace as a template');
+    console.log('  atris publish --name "CRM Starter"    Publish with a specific name');
+    console.log('  atris publish --name crm --description "Sales CRM template"');
+    process.exit(0);
+  }
+
+  const creds = loadCredentials();
+  if (!creds || !creds.token) {
+    console.error('Not logged in. Run: atris login');
+    process.exit(1);
+  }
+
+  // Parse --name
+  let name = null;
+  const nameEq = process.argv.find(a => a.startsWith('--name='));
+  if (nameEq) { name = nameEq.slice(7); }
+  else {
+    const ni = process.argv.indexOf('--name');
+    if (ni !== -1 && process.argv[ni + 1] && !process.argv[ni + 1].startsWith('-')) name = process.argv[ni + 1];
+  }
+
+  // Parse --description
+  let description = null;
+  const descEq = process.argv.find(a => a.startsWith('--description='));
+  if (descEq) { description = descEq.slice(14); }
+  else {
+    const di = process.argv.indexOf('--description');
+    if (di !== -1 && process.argv[di + 1] && !process.argv[di + 1].startsWith('-')) description = process.argv[di + 1];
+  }
+
+  // Resolve name from .atris/business.json or directory name
+  if (!name) {
+    const bizFile = path.join(process.cwd(), '.atris', 'business.json');
+    if (fs.existsSync(bizFile)) {
+      try {
+        const biz = JSON.parse(fs.readFileSync(bizFile, 'utf8'));
+        name = biz.name || biz.slug;
+      } catch {}
+    }
+    if (!name) name = path.basename(process.cwd());
+  }
+
+  // Collect files from atris/ directory
+  const atrisDir = path.join(process.cwd(), 'atris');
+  if (!fs.existsSync(atrisDir)) {
+    console.error('No atris/ directory found. Run from a workspace root.');
+    process.exit(1);
+  }
+
+  const EXCLUDE = ['logs', '.env', 'secrets'];
+  const files = [];
+
+  function collectFiles(dir, prefix) {
+    const entries = fs.readdirSync(dir);
+    for (const entry of entries) {
+      if (EXCLUDE.includes(entry)) continue;
+      if (entry.startsWith('.env') || entry === 'secrets') continue;
+      const full = path.join(dir, entry);
+      const rel = prefix ? `${prefix}/${entry}` : entry;
+      const stat = fs.statSync(full);
+      if (stat.isDirectory()) {
+        collectFiles(full, rel);
+      } else if (stat.isFile() && stat.size < 512 * 1024) {
+        try {
+          const content = fs.readFileSync(full, 'utf8');
+          files.push({ path: `/${rel}`, content });
+        } catch {}
+      }
+    }
+  }
+
+  collectFiles(atrisDir, '');
+
+  if (files.length === 0) {
+    console.error('No publishable files found in atris/.');
+    process.exit(1);
+  }
+
+  console.log('');
+  console.log(`Publishing "${name}" (${files.length} files)...`);
+
+  // POST to API
+  const body = { name, description: description || '', files };
+  const result = await apiRequestJson('/workspace/templates', {
+    method: 'POST',
+    token: creds.token,
+    body,
+  });
+
+  if (!result.ok) {
+    console.error(`\n  Publish failed: ${result.errorMessage || result.error || result.status}`);
+    process.exit(1);
+  }
+
+  // Save local copy
+  const localDir = path.join(os.homedir(), '.atris', 'templates', name.toLowerCase().replace(/[^a-z0-9-]/g, '-'));
+  fs.mkdirSync(localDir, { recursive: true });
+
+  for (const f of files) {
+    const filePath = path.join(localDir, f.path.replace(/^\//, ''));
+    fs.mkdirSync(path.dirname(filePath), { recursive: true });
+    fs.writeFileSync(filePath, f.content);
+  }
+
+  // Save metadata
+  fs.writeFileSync(path.join(localDir, 'template.json'), JSON.stringify({
+    name,
+    description: description || '',
+    file_count: files.length,
+    published_at: new Date().toISOString(),
+  }, null, 2));
+
+  console.log('');
+  console.log(`  Published as '${name}'. Others can fork with: atris fork ${name}`);
+  console.log(`  Local copy saved to ${localDir}`);
+  console.log('');
+}
+
+module.exports = { publishAtris };

package/commands/pull.js

@@ -8,10 +8,39 @@ const { getLogPath } = require('../lib/file-ops');
 const { parseJournalSections, mergeSections, reconstructJournal } = require('../lib/journal');
 const { loadBusinesses } = require('./business');
 const { loadManifest, saveManifest, computeFileHash, buildManifest, computeLocalHashes, threeWayCompare } = require('../lib/manifest');
+const { normalizeWikiOnlyPrefix } = require('../lib/wiki');
+
+function pruneEmptyParentDirs(filePath, stopDir) {
+  let current = path.dirname(filePath);
+  const boundary = path.resolve(stopDir);
+  while (current.startsWith(boundary) && current !== boundary) {
+    try {
+      if (fs.readdirSync(current).length > 0) break;
+      fs.rmdirSync(current);
+      current = path.dirname(current);
+    } catch {
+      break;
+    }
+  }
+}
 
 async function pullAtris() {
   let arg = process.argv[3];
 
+  if (arg === '--help') {
+    console.log('Usage: atris pull [business] [--into <path>] [--only <prefix>] [--keep-local] [--timeout <seconds>]');
+    console.log('');
+    console.log('  Pull is force-overwrite by default. Cloud is the source of truth.');
+    console.log('  Local files that conflict with cloud are replaced by the cloud version.');
+    console.log('');
+    console.log('  atris pull                   Pull into current business workspace');
+    console.log('  atris pull doordash          Pull a business into ./doordash or --into <path>');
+    console.log('  atris pull doordash --into /tmp/doordash');
+    console.log('  atris pull doordash --only atris/wiki/');
+    console.log('  atris pull --keep-local      Preserve conflicting local edits as .remote files (legacy)');
+    return;
+  }
+
   // Auto-detect business from .atris/business.json in current dir
   if (!arg || arg.startsWith('--')) {
     const bizFile = path.join(process.cwd(), '.atris', 'business.json');
@@ -91,7 +120,10 @@ async function pullBusiness(slug) {
     process.exit(1);
   }
 
-  const force = process.argv.includes('--force');
+  // Pull is force-overwrite by default (cloud = source of truth).
+  // --keep-local opts back into the legacy three-way merge with .remote conflict files.
+  // --force is still accepted as an alias for the default for muscle-memory.
+  const force = !process.argv.includes('--keep-local');
 
   // Parse --only flag: comma-separated directory prefixes to filter
   // Supports both --only=team/,context/ and --only team/,context/
@@ -108,6 +140,8 @@ async function pullBusiness(slug) {
   const onlyPrefixes = onlyRaw
     ? onlyRaw.split(',').map(p => {
         let norm = p.replace(/^\//, '');
+        const wikiPrefix = normalizeWikiOnlyPrefix(norm);
+        if (wikiPrefix) return wikiPrefix;
         if (norm && !norm.endsWith('/') && !norm.includes('.')) norm += '/';
         return norm;
       }).filter(Boolean)
@@ -132,20 +166,22 @@ async function pullBusiness(slug) {
   let outputDir;
   if (intoIdx !== -1 && process.argv[intoIdx + 1]) {
     outputDir = path.resolve(process.argv[intoIdx + 1]);
+  } else if (fs.existsSync(path.join(process.cwd(), '.atris', 'business.json'))) {
+    // Inside a pulled workspace — pull into current dir (no nesting)
+    outputDir = process.cwd();
+  } else if (fs.existsSync(path.join(process.cwd(), 'atris')) && fs.statSync(path.join(process.cwd(), 'atris')).isDirectory()) {
+    // Inside an atris init'd workspace — merge business into current dir
+    outputDir = process.cwd();
   } else {
-    const atrisDir = path.join(process.cwd(), 'atris');
-    if (fs.existsSync(atrisDir)) {
-      outputDir = path.join(atrisDir, slug);
-    } else {
-      outputDir = path.join(process.cwd(), slug);
-    }
+    // Default: ./{slug}/ in current directory
+    outputDir = path.join(process.cwd(), slug);
   }
 
   // Resolve business ID — always refresh from API to avoid stale workspace_id
   let businessId, workspaceId, businessName, resolvedSlug;
   const businesses = loadBusinesses();
 
-  const listResult = await apiRequestJson('/businesses/', { method: 'GET', token: creds.token });
+  const listResult = await apiRequestJson('/business/', { method: 'GET', token: creds.token });
   if (!listResult.ok) {
     // Fall back to local cache if API fails
     if (businesses[slug]) {
@@ -187,6 +223,29 @@ async function pullBusiness(slug) {
     process.exit(1);
   }
 
+  // Auto-wake the EC2 computer if --auto-wake is set.
+  // Without this, pull silently serves stale data from agent_files cache when
+  // the computer is asleep — the bug that confused us all night.
+  const autoWake = process.argv.includes('--auto-wake');
+  if (autoWake) {
+    const statusResult = await apiRequestJson(`/business/${businessId}/ai-computer/status`, { method: 'GET', token: creds.token });
+    const computerStatus = statusResult.ok && statusResult.data ? statusResult.data.status : 'unknown';
+    if (computerStatus !== 'running' || !(statusResult.data && statusResult.data.endpoint)) {
+      process.stdout.write('  Waking EC2 computer... ');
+      await apiRequestJson(`/business/${businessId}/ai-computer/wake`, { method: 'POST', token: creds.token });
+      const wakeStart = Date.now();
+      while (Date.now() - wakeStart < 90000) {
+        await new Promise((r) => setTimeout(r, 3000));
+        const s = await apiRequestJson(`/business/${businessId}/ai-computer/status`, { method: 'GET', token: creds.token });
+        if (s.ok && s.data && s.data.status === 'running' && s.data.endpoint) {
+          const elapsed = Math.floor((Date.now() - wakeStart) / 1000);
+          console.log(`awake (${elapsed}s)`);
+          break;
+        }
+      }
+    }
+  }
+
   // Load manifest (last sync state)
   const manifest = loadManifest(resolvedSlug || slug);
   const timeSince = manifest ? _timeSince(manifest.last_sync) : null;
@@ -203,21 +262,104 @@ async function pullBusiness(slug) {
     process.stdout.write(`\r  Fetching workspace... ${spinner[spinIdx++ % 4]} ${elapsed}s`);
   }, 250);
 
-  // Get remote snapshot — pass --only prefixes to server for faster response
-  let snapshotUrl = `/businesses/${businessId}/workspaces/${workspaceId}/snapshot?include_content=true`;
-  if (onlyPrefixes) {
-    snapshotUrl += `&paths=${encodeURIComponent(onlyPrefixes.map(p => p.replace(/\/$/, '')).join(','))}`;
-  }
-  const result = await apiRequestJson(snapshotUrl, { method: 'GET', token: creds.token, timeoutMs });
+  // Smart pull: if we have a manifest (not first sync), fetch hashes first, then only changed content
+  const hasManifest = manifest && manifest.files && Object.keys(manifest.files).length > 0 && !force;
+  let result;
+
+  const pathsParam = onlyPrefixes ? `&paths=${encodeURIComponent(onlyPrefixes.map(p => p.replace(/\/$/, '')).join(','))}` : '';
 
-  clearInterval(loading);
-  const totalSec = Math.floor((Date.now() - startTime) / 1000);
-  process.stdout.write(`\r  Fetched in ${totalSec}s.${' '.repeat(20)}\n`);
+  if (hasManifest) {
+    // Phase 1: fetch hashes only (fast — no file content transferred)
+    const hashUrl = `/business/${businessId}/workspaces/${workspaceId}/snapshot?include_content=false${pathsParam}`;
+    const hashResult = await apiRequestJson(hashUrl, { method: 'GET', token: creds.token, timeoutMs });
+
+    if (hashResult.ok && hashResult.data && hashResult.data.files) {
+      // Diff against manifest to find changed files
+      const remoteHashes = {};
+      for (const f of hashResult.data.files) {
+        if (f.path && f.hash) remoteHashes[f.path] = f.hash;
+      }
+      const changedPaths = [];
+      const manifestFiles = manifest.files || {};
+      for (const [p, hash] of Object.entries(remoteHashes)) {
+        const prev = manifestFiles[p];
+        if (!prev || prev.hash !== hash) changedPaths.push(p);
+      }
+
+      if (changedPaths.length === 0) {
+        clearInterval(loading);
+        process.stdout.write(`\r  Checked ${Object.keys(remoteHashes).length} files in ${Math.floor((Date.now() - startTime) / 1000)}s.${' '.repeat(10)}\n`);
+        // Still need full result for diff logic below — build it from hash-only data
+        result = { ok: true, data: { files: hashResult.data.files } };
+      } else {
+        // Phase 2: fetch ONLY changed files via batch endpoint (not full snapshot)
+        clearInterval(loading);
+        const checkSec = Math.floor((Date.now() - startTime) / 1000);
+        console.log(`\r  Checked in ${checkSec}s — ${changedPaths.length} changed, ${Object.keys(remoteHashes).length - changedPaths.length} unchanged.${' '.repeat(10)}`);
+
+        const startPhase2 = Date.now();
+        const loading2 = setInterval(() => {
+          const elapsed = Math.floor((Date.now() - startPhase2) / 1000);
+          process.stdout.write(`\r  Fetching ${changedPaths.length} changed files... ${spinner[spinIdx++ % 4]} ${elapsed}s`);
+        }, 250);
+
+        // Try batch file read first (fast — only changed files)
+        const batchUrl = `/business/${businessId}/workspaces/${workspaceId}/files/batch`;
+        const batchResult = await apiRequestJson(batchUrl, {
+          method: 'POST',
+          token: creds.token,
+          body: { paths: changedPaths },
+          timeoutMs,
+        });
+
+        clearInterval(loading2);
+        const phase2Sec = Math.floor((Date.now() - startPhase2) / 1000);
+
+        if (batchResult.ok && batchResult.data && batchResult.data.files) {
+          process.stdout.write(`\r  Fetched ${batchResult.data.files.length} files in ${phase2Sec}s.${' '.repeat(10)}\n`);
+          // Merge: hash-only results + content for changed files
+          const contentMap = {};
+          for (const f of batchResult.data.files) {
+            if (f.path) contentMap[f.path] = f;
+          }
+          // Build merged file list: all hash-only entries + inject content for changed ones
+          const mergedFiles = hashResult.data.files.map(f => {
+            const withContent = contentMap[f.path];
+            return withContent || f;
+          });
+          result = { ok: true, data: { files: mergedFiles } };
+        } else {
+          // Batch not available — fall back to full snapshot
+          process.stdout.write(`\r  Batch unavailable, fetching full snapshot...${' '.repeat(10)}\n`);
+          const contentUrl = `/business/${businessId}/workspaces/${workspaceId}/snapshot?include_content=true${pathsParam}`;
+          result = await apiRequestJson(contentUrl, { method: 'GET', token: creds.token, timeoutMs });
+          const fullSec = Math.floor((Date.now() - startPhase2) / 1000);
+          process.stdout.write(`\r  Fetched in ${fullSec}s.${' '.repeat(20)}\n`);
+        }
+      }
+    } else {
+      // Hash-only fetch failed — fall back to full snapshot
+      const fullUrl = `/business/${businessId}/workspaces/${workspaceId}/snapshot?include_content=true${pathsParam}`;
+      result = await apiRequestJson(fullUrl, { method: 'GET', token: creds.token, timeoutMs });
+      clearInterval(loading);
+      process.stdout.write(`\r  Fetched in ${Math.floor((Date.now() - startTime) / 1000)}s.${' '.repeat(20)}\n`);
+    }
+  } else {
+    // First sync or --force — full snapshot with content
+    const snapshotUrl = `/business/${businessId}/workspaces/${workspaceId}/snapshot?include_content=true${pathsParam}`;
+    result = await apiRequestJson(snapshotUrl, { method: 'GET', token: creds.token, timeoutMs });
+    clearInterval(loading);
+    const totalSec = Math.floor((Date.now() - startTime) / 1000);
+    process.stdout.write(`\r  Fetched in ${totalSec}s.${' '.repeat(20)}\n`);
+  }
 
   if (!result.ok) {
     const msg = result.errorMessage || result.error || `HTTP ${result.status}`;
     if (result.status === 0 || (typeof msg === 'string' && msg.toLowerCase().includes('timeout'))) {
       console.error(`\n  Workspace timed out (large workspaces can take 60s+). Try: atris pull ${slug} --timeout=600`);
+    } else if (result.status === 502) {
+      console.error(`\n  Computer didn't respond in time. It may be waking up or the workspace is large.`);
+      console.error(`  Try again in 30s, or use: atris pull ${slug} --only=team/,context/`);
     } else if (result.status === 409) {
       console.error(`\n  Computer is sleeping. Wake it first, then pull again.`);
     } else if (result.status === 403) {
@@ -233,11 +375,15 @@ async function pullBusiness(slug) {
   let files = result.data.files || [];
   if (files.length === 0) {
     console.log('  Workspace is empty.');
-    return;
+    // Don't early-return in force mode: we still need to fall through to the
+    // mirror sweep so a genuinely-emptied cloud can clear local files. The
+    // sweep itself has a safety guard that refuses to wipe local content
+    // when remote reports empty (the snapshot-glitch case), so this is safe.
+    if (!force) return;
+  } else {
+    console.log(`  Processing ${files.length} files...`);
   }
 
-  console.log(`  Processing ${files.length} files...`);
-
   // Apply --only filter if specified
   if (onlyPrefixes) {
     files = files.filter(file => {
@@ -247,23 +393,46 @@ async function pullBusiness(slug) {
     });
     if (files.length === 0) {
       console.log(`  No files matched --only filter: ${onlyPrefixes.join(', ')}`);
-      return;
+      // Don't early-return: we still need to update the manifest so paths
+      // that USED to be in the scoped subtree but were deleted on cloud
+      // get evicted from the manifest. Without this, the next push freshness
+      // check would forever flag those paths as drift and demand a pull —
+      // but the pull would early-return again, creating a deadlock.
+    } else {
+      console.log(`  Filtered to ${files.length} files matching: ${onlyPrefixes.join(', ')}`);
     }
-    console.log(`  Filtered to ${files.length} files matching: ${onlyPrefixes.join(', ')}`);
   }
 
-  // Build remote file map {path: {hash, size, content}}
+  // Build remote file map {path: {hash, size}} and content map {path: content}.
+  //
+  // CRITICAL: smart-pull (hash-only fetch) returns files with `path`+`hash`+`size`
+  // but no `content`. Phase-2 batch fetch only adds content for CHANGED files —
+  // unchanged files stay hash-only. We must include hash-only entries in remoteFiles
+  // so threeWayCompare doesn't see them as missing-from-remote (deletedRemote).
+  // The previous version skipped any file without content, which caused every
+  // smart-pull to mark every unchanged file as deleted-on-cloud and rmSync them.
   const remoteFiles = {};
   const remoteContent = {};
+  const crypto = require('crypto');
   for (const file of files) {
-    if (!file.path || file.binary || file.content === null || file.content === undefined) continue;
-    // Skip empty files (deleted files that were blanked out)
-    if (file.content === '') continue;
-    // Compute hash from content bytes (matches computeLocalHashes raw byte hashing)
-    const crypto = require('crypto');
-    const rawBytes = Buffer.from(file.content, 'utf-8');
-    remoteFiles[file.path] = { hash: crypto.createHash('sha256').update(rawBytes).digest('hex'), size: rawBytes.length };
-    remoteContent[file.path] = file.content;
+    if (!file.path || file.binary) continue;
+    // An empty string IS valid content (a real, zero-byte file). The earlier
+    // version excluded `content === ''` from the hasContent path, which made
+    // empty files masquerade as hash-only entries; they'd then be recorded in
+    // the manifest (with the empty-string hash) but never written to disk.
+    // A subsequent push would compare local (file missing) to manifest (file
+    // present) and try to delete the file from cloud — silently undoing the
+    // very thing the user just pulled.
+    const hasContent = file.content !== null && file.content !== undefined && typeof file.content === 'string';
+    if (hasContent) {
+      // Full content available — hash from raw bytes (matches computeLocalHashes)
+      const rawBytes = Buffer.from(file.content, 'utf-8');
+      remoteFiles[file.path] = { hash: crypto.createHash('sha256').update(rawBytes).digest('hex'), size: rawBytes.length };
+      remoteContent[file.path] = file.content;
+    } else if (file.hash) {
+      // Hash-only entry from smart pull — trust the cloud-reported hash
+      remoteFiles[file.path] = { hash: file.hash, size: file.size || 0 };
+    }
   }
 
   // Compute local file hashes
@@ -277,6 +446,7 @@ async function pullBusiness(slug) {
 
   // Apply changes
   let pulled = 0;
+  let deleted = 0;
   let conflictCount = 0;
   let unchangedCount = diff.unchanged.length;
 
@@ -320,13 +490,84 @@ async function pullBusiness(slug) {
     }
   }
 
-  // Warn about remote deletions
+  // Apply remote deletions
   for (const p of diff.deletedRemote) {
-    console.log(`  - ${p.replace(/^\//, '')}  deleted on computer`);
+    const baseHash = effectiveManifest && effectiveManifest.files && effectiveManifest.files[p]
+      ? effectiveManifest.files[p].hash
+      : null;
+    const localHash = localFiles[p] ? localFiles[p].hash : null;
+    const localChanged = Boolean(baseHash && localHash && localHash !== baseHash);
+
+    if (force || !localChanged) {
+      const localPath = path.join(outputDir, p.replace(/^\//, ''));
+      fs.rmSync(localPath, { force: true });
+      pruneEmptyParentDirs(localPath, outputDir);
+      console.log(`  - ${p.replace(/^\//, '')}  deleted on computer`);
+      deleted++;
+    } else {
+      console.log(`  \u26A0 ${p.replace(/^\//, '')}  deleted on computer, but you changed it locally`);
+      conflictCount++;
+    }
+  }
+
+  // FORCE MIRROR SWEEP — local must EXACTLY match cloud after a force pull.
+  // The threeWayCompare path with effectiveManifest=null only computes
+  // newLocal/conflicts/newRemote and never marks files as deletedRemote, so
+  // local-only files (created locally, never on cloud) survive a force pull.
+  // That breaks the "cloud is the source of truth" promise. Sweep them now.
+  //
+  // SAFETY GUARDS — without these the sweep can wipe an entire local copy:
+  //   • Scope the sweep: when --only is set, only sweep paths INSIDE the
+  //     prefix(es). Out-of-scope local files must be left alone — the user
+  //     asked for a partial pull, not a workspace-wide reset.
+  //   • Skip when remoteFiles is empty AND local has in-scope content: the
+  //     snapshot endpoint has a known server-side bug where it returns 0
+  //     files for healthy workspaces. If cloud reports empty but local has
+  //     in-scope content we refuse to sweep — the user can re-run with
+  //     --keep-local and investigate, or run `atris align --hard` for an
+  //     explicit nuke.
+  //   • Skip files the server's snapshot filter hides. The warm runner's
+  //     _snapshot_dir (ecs_warm_runner.py) deliberately omits CLAUDE.md and
+  //     other names from snapshots, so they never appear in remoteFiles even
+  //     when they DO exist on cloud. Sweeping them would delete server-managed
+  //     files that aren't actually missing on cloud.
+  const SERVER_HIDDEN_BASENAMES = new Set(['CLAUDE.md']);
+  function basename(p) {
+    const idx = p.lastIndexOf('/');
+    return idx === -1 ? p : p.slice(idx + 1);
+  }
+  function isInScope(p) {
+    if (!onlyPrefixes) return true;
+    const rel = p.replace(/^\//, '');
+    return onlyPrefixes.some((pref) => rel.startsWith(pref));
+  }
+  if (force) {
+    const remotePathSet = new Set(Object.keys(remoteFiles));
+    const inScopeLocal = Object.keys(localFiles).filter(isInScope);
+    if (remotePathSet.size === 0 && inScopeLocal.length > 0) {
+      console.log('');
+      console.log('  ⚠ Cloud reported zero files but local has in-scope content. Refusing to sweep.');
+      console.log('    This usually means the snapshot endpoint glitched. Try again,');
+      console.log('    or run `atris align --hard` if you really want to nuke local.');
+    } else {
+      for (const p of inScopeLocal) {
+        if (remotePathSet.has(p)) continue;
+        if (SERVER_HIDDEN_BASENAMES.has(basename(p))) continue;
+        const localPath = path.join(outputDir, p.replace(/^\//, ''));
+        try {
+          fs.rmSync(localPath, { force: true });
+          pruneEmptyParentDirs(localPath, outputDir);
+          console.log(`  - ${p.replace(/^\//, '')}  not on cloud, removed locally`);
+          deleted++;
+        } catch {
+          // ignore — file might already be gone
+        }
+      }
+    }
   }
 
   // Show unchanged
-  if (unchangedCount > 0 && pulled === 0 && conflictCount === 0 && diff.deletedRemote.length === 0) {
+  if (unchangedCount > 0 && pulled === 0 && deleted === 0 && conflictCount === 0) {
     console.log('  Already up to date.');
   }
 
@@ -334,17 +575,17 @@ async function pullBusiness(slug) {
   console.log('');
   const parts = [];
   if (pulled > 0) parts.push(`${pulled} pulled`);
+  if (deleted > 0) parts.push(`${deleted} deleted`);
   if (diff.newRemote.length > 0 && !parts.some(p => p.includes('pulled'))) parts.push(`${diff.newRemote.length} new`);
   if (unchangedCount > 0) parts.push(`${unchangedCount} unchanged`);
   if (conflictCount > 0) parts.push(`${conflictCount} conflict${conflictCount > 1 ? 's' : ''}`);
-  if (diff.deletedRemote.length > 0) parts.push(`${diff.deletedRemote.length} deleted remotely`);
   if (parts.length > 0) console.log(`  ${parts.join(', ')}.`);
 
   // Get current commit hash from remote (for manifest)
   let commitHash = null;
   try {
     const headResult = await apiRequestJson(
-      `/businesses/${businessId}/workspaces/${workspaceId}/git/head`,
+      `/business/${businessId}/workspaces/${workspaceId}/git/head`,
       { method: 'GET', token: creds.token }
     );
     if (headResult.ok && headResult.data && headResult.data.commit) {
@@ -354,10 +595,43 @@ async function pullBusiness(slug) {
     // Git might not be initialized yet — that's fine
   }
 
-  // Save manifest — when using --only, merge into existing manifest to avoid data loss
+  // ANTI-WIPE GUARD: if cloud reported zero in-scope files but local still
+  // has in-scope content (i.e. the sweep refused), don't overwrite the
+  // manifest with empty data for the scoped subtree. The manifest is the
+  // authoritative record of what we last knew was on cloud — wiping it
+  // because of a transient empty snapshot would force every subsequent
+  // push to flag every file as drift. Better to leave the manifest stale
+  // than to record a never-actually-true "cloud is empty" state.
+  //
+  // Applies to both whole-workspace pulls and scoped (--only) pulls.
+  {
+    const inScopeLocalCount = onlyPrefixes
+      ? Object.keys(localFiles).filter((p) => onlyPrefixes.some((pref) => p.replace(/^\//, '').startsWith(pref))).length
+      : Object.keys(localFiles).length;
+    if (Object.keys(remoteFiles).length === 0 && inScopeLocalCount > 0) {
+      return;
+    }
+  }
+
+  // Save manifest — when using --only, merge into existing manifest so paths
+  // OUTSIDE the scoped prefix don't get dropped. Inside the scoped prefix,
+  // however, we must replace (not merge) so that files deleted on cloud
+  // since the last sync get evicted from the manifest. Without this, the
+  // push freshness check would forever flag those paths as "deleted on
+  // cloud" drift, blocking pushes for no reason.
   let manifestFiles = remoteFiles;
   if (onlyPrefixes && manifest && manifest.files) {
-    manifestFiles = { ...manifest.files, ...remoteFiles };
+    const merged = {};
+    // 1. Keep paths from old manifest that are OUTSIDE the scoped prefix.
+    for (const [p, info] of Object.entries(manifest.files)) {
+      const inScope = onlyPrefixes.some((pref) => p.replace(/^\//, '').startsWith(pref));
+      if (!inScope) merged[p] = info;
+    }
+    // 2. Overwrite the in-scope subtree with what we just pulled (cloud truth).
+    for (const [p, info] of Object.entries(remoteFiles)) {
+      merged[p] = info;
+    }
+    manifestFiles = merged;
   }
   const newManifest = buildManifest(manifestFiles, commitHash);
   saveManifest(resolvedSlug || slug, newManifest);
@@ -371,6 +645,63 @@ async function pullBusiness(slug) {
     workspace_id: workspaceId,
     name: businessName,
   }, null, 2));
+
+  // Wire skills → .claude/skills/ so they work as slash commands
+  const skillsDir = path.join(outputDir, 'skills');
+  const claudeSkillsDir = path.join(outputDir, '.claude', 'skills');
+
+  if (fs.existsSync(skillsDir)) {
+    fs.mkdirSync(claudeSkillsDir, { recursive: true });
+
+    // Recursively find all skill folders (any dir containing SKILL.md, at any depth)
+    const wireSkills = (dir, relPrefix) => {
+      const entries = fs.readdirSync(dir);
+      for (const entry of entries) {
+        const fullPath = path.join(dir, entry);
+        if (!fs.statSync(fullPath).isDirectory()) continue;
+        if (entry === 'README.md' || entry.startsWith('.')) continue;
+
+        const skillFile = path.join(fullPath, 'SKILL.md');
+        if (fs.existsSync(skillFile)) {
+          // This is a leaf skill — wire it
+          const skillName = relPrefix ? `${relPrefix}-${entry}` : entry;
+          const symlinkPath = path.join(claudeSkillsDir, skillName);
+          const relativePath = path.relative(path.dirname(symlinkPath), fullPath);
+
+          // Business skills override init skills (remove existing symlink if present)
+          if (fs.existsSync(symlinkPath)) {
+            try {
+              const stat = fs.lstatSync(symlinkPath);
+              if (stat.isSymbolicLink()) fs.unlinkSync(symlinkPath);
+              else continue; // Don't overwrite real directories
+            } catch { continue; }
+          }
+          try {
+            fs.symlinkSync(relativePath, symlinkPath);
+          } catch (e) {
+            // Fallback: copy
+            fs.mkdirSync(symlinkPath, { recursive: true });
+            fs.copyFileSync(skillFile, path.join(symlinkPath, 'SKILL.md'));
+          }
+        }
+
+        // Recurse into subdirectories (e.g. skills/executive/pipeline-health/)
+        wireSkills(fullPath, relPrefix ? `${relPrefix}-${entry}` : entry);
+      }
+    };
+
+    wireSkills(skillsDir, '');
+
+    // Count wired skills
+    const wiredSkills = fs.readdirSync(claudeSkillsDir).filter(f => {
+      const p = path.join(claudeSkillsDir, f);
+      return fs.statSync(p).isDirectory();
+    });
+    if (wiredSkills.length > 0) {
+      console.log(`  Wired ${wiredSkills.length} skills → .claude/skills/`);
+    }
+  }
+
 }