@roulabs/mx 1.2.0 → 1.2.2

package/bin/mx.js

@@ -167,11 +167,23 @@ function initRuntime(target0, templatesDir2) {
   removeStaleRuntimeReadme(target);
   return { runtime: target, created };
 }
+function ensureWorkScaffolding(root, workName) {
+  const created = [];
+  const sessions = path3.join(workDir(root, workName), "sessions");
+  if (!exists(sessions)) {
+    fs4.mkdirSync(sessions, { recursive: true });
+    created.push(sessions);
+  }
+  return created;
+}
 function updateRuntime(root, templatesDir2) {
-  const dest = stampClaudeMd(root, templatesDir2);
-  const updated = [dest];
+  const updated = [];
+  updated.push(stampClaudeMd(root, templatesDir2));
   const ctxIndex = stampContextIndex(root, templatesDir2);
   if (ctxIndex) updated.push(ctxIndex);
+  for (const workName of listWorkNames(root)) {
+    updated.push(...ensureWorkScaffolding(root, workName));
+  }
   removeStaleRuntimeReadme(root);
   return { runtime: root, updated };
 }
@@ -309,10 +321,10 @@ function workNew(root, name, description = "") {
   const dir = workDir(root, name);
   if (exists(dir)) throw new MxError(`work already exists: ${name}`, "EXISTS");
   fs6.mkdirSync(dir, { recursive: true });
-  fs6.mkdirSync(path4.join(dir, "sessions"), { recursive: true });
   const work = { name, description, worktrees: [] };
   writeWork(root, work);
   writeJson(workspaceFile(root, name), { folders: [], settings: {} });
+  ensureWorkScaffolding(root, name);
   return { ...work, path: dir };
 }
 function listWorksInfo(root, opts = {}) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@roulabs/mx",
-  "version": "1.2.0",
+  "version": "1.2.2",
   "description": "mx — run several features in parallel across shared repos using git worktrees",
   "type": "module",
   "bin": {

package/templates/CLAUDE.md

@@ -115,7 +115,7 @@ Example entry:
 
 Primary path:
 
-1. Read `<runtime>/context/INDEX.json` — every entry's metadata in one Read.
+1. **Read `<runtime>/context/INDEX.json` on every task** — trivial or not, small or large. Skimming a metadata index is cheap; the cost of missing a relevant entry is high. This is a hard rule, not a heuristic.
 2. Open files at `<runtime>/context/<path>.md` for entries whose metadata matches the current task.
 
 When INDEX descriptions don't surface what you need — and often they won't — fall back to anything that works:
@@ -123,12 +123,10 @@ When INDEX descriptions don't surface what you need — and often they won't —
 - **Grep `<runtime>/context/`** for keywords. Frequently the term you need lives in a body, not in any description.
 - **`ls` the folder recursively** to spot entries on disk that aren't indexed (orphans), and read them directly when relevant.
 - **Follow `related` chains** outward from a known-relevant entry to find the rest of a cluster.
-- **Read everything** if the registry is small (< ~30 entries) and you're starting unfamiliar work — cheaper than guessing.
+- **Read everything** when in doubt — better than guessing.
 
 INDEX is the *primary* discovery surface, not the only one. Use whatever gets you to the right entry fastest — direct grep, full-content scan, recursive read, following links, your judgment.
 
-A 30-second skim of INDEX is free; do it before any non-trivial task. Skip only for typo-fix-level work.
-
 ### Maintain INDEX.json as you go
 
 When you add, rename, remove, or restructure an entry, update INDEX in the same change. Drift means orphan files (invisible to future sessions) or stale entries (false positives). After editing INDEX, sanity-check it parses as valid JSON.