@spardutti/claude-skills 2.60.1 → 2.60.3

package/README.md

@@ -149,6 +149,7 @@ The CLI installs three kinds of artifact into your project's `.claude/` director
 - **Skills** → `.claude/skills/` — playbooks Claude loads while coding.
 - **Commands** → `.claude/commands/` — slash commands you invoke directly.
 - **Subagents** → `.claude/agents/` — declared by commands via `requires-agents`, installed for you.
+  A command that needs another names it in `requires-commands`, and that one is installed too.
 
 It then offers to install the hooks. Two guard the **front**: nothing is written
 until your skills have been evaluated *and* applied to the file being written.
@@ -159,6 +160,27 @@ receipt proving the delivery gate passed on exactly that code.
 Full details — every hook, every marker, every config key, and what to do when one
 blocks something it shouldn't — are in **[HOOKS.md](./HOOKS.md)**.
 
+### The registry: systems a plan must reuse
+
+Grep only finds the words a request uses. Ask for "a skill that fires at the nearby
+enemy" and nothing matches the aiming code, so Claude writes new aiming code. A
+registry entry is a small file beside the system that names it once:
+
+```yaml
+# sim/aim.registry.yaml
+name: aim-rules
+summary: how a skill picks which enemy to hit
+location: sim/aim.gd
+extend_by: add a value to `Aim.Rule` and its branch in `Aim.pick`
+never: sort or filter enemies inside a skill
+```
+
+`/plan-feature` reads every entry and must mark each one `reuse` or `skip`. A plan
+fails if it says `reuse` while its stubs never use the names in `extend_by`, and an
+entry whose names left the code is flagged as stale. Nothing else reads the registry.
+`/registry aim` drafts an entry for your yes. Replayed on a real game feature, the
+targeting rule went inside the skill without an entry, and into `Aim` with one, 5 runs of 5.
+
 ### Tracking & Updates
 
 Every install writes a manifest at `.claude/.claude-skills.json` recording what the CLI installed and the catalog version. On the next run it uses the manifest to:
@@ -172,7 +194,7 @@ Every install writes a manifest at `.claude/.claude-skills.json` recording what
 npx @spardutti/claude-skills --sync
 ```
 
-`--sync` refreshes every tracked item, and any hooks the project already has, to the latest catalog and prunes stale ones in one shot — no menu. For a project that predates the manifest, the first normal run offers a one-time cleanup of `.claude/` content no longer in the catalog.
+`--sync` refreshes every tracked item, and any hooks the project already has, to the latest catalog and prunes stale ones in one shot — no menu. It adds any command a tracked command requires, and names each skill or command new to the catalog once, so you can pick it. For a project that predates the manifest, the first normal run offers a one-time cleanup of `.claude/` content no longer in the catalog.
 
 `--sync-all[=dir]` does the same for every project under `dir` (default: your home folder) that has a manifest, fetching the catalog once:
 

package/hooks/registry-lib.mjs

@@ -57,7 +57,8 @@ export function loadRegistry(root, files) {
 // Code-shaped words (AimStrategy, pick_target, aimRegistry) or `backticked` — never "implement".
 export function symbolsOf(extendBy = "") {
   const ticked = [...extendBy.matchAll(/`([\w.]+)`/g)].map((m) => m[1]);
-  const shaped = extendBy.match(/\b(?:[A-Z][a-z0-9]+[A-Z]\w*|[a-z]\w*[A-Z]\w*|\w+_\w+)\b/g) ?? [];
+  const bare = extendBy.replace(/`[^`]*`/g, " ");
+  const shaped = bare.match(/\b(?:[A-Z][a-z0-9]+[A-Z]\w*|[a-z]\w*[A-Z]\w*|\w+_\w+)\b/g) ?? [];
   return [...new Set([...ticked, ...shaped])];
 }
 

package/hooks/registry.mjs

@@ -9,6 +9,12 @@ const root = projectRoot();
 const files = projectFiles(root);
 const entries = loadRegistry(root, files);
 const uses = (text, word) => new RegExp(`\\b${escape(word)}\\b`).test(text);
+// Code calls a method on a value (`e.chill(2)`), so `EnemyState.chill` is the type named plus `.chill` used.
+const usesIn = (text, name) => {
+  const parts = name.split(".");
+  if (parts.length === 1 || uses(text, name)) return uses(text, name);
+  return uses(text, parts[0]) && new RegExp(`\\.${escape(parts.at(-1))}\\b`).test(text);
+};
 const fromCwd = (f) => relative(root, resolve(f));
 
 function list() {
@@ -38,7 +44,7 @@ function checkPlan(planPath, stubs) {
   for (const e of entries) {
     const mark = marks.get(e.name);
     if (!mark) fails.push(`registry FAIL ${e.name}: missing from ## Registry — mark it reuse or skip`);
-    else if (mark === "reuse" && !symbolsOf(e.extend_by).some((s) => uses(added, s))) {
+    else if (mark === "reuse" && !symbolsOf(e.extend_by).some((s) => usesIn(added, s))) {
       fails.push(`registry FAIL ${e.name}: marked reuse, but no stub uses ${symbolsOf(e.extend_by).join(", ") || "a name from extend_by"}`);
     } else if (mark === "skip") {
       const hits = triggerHits(e, added);

package/hooks/ship-prep.sh

@@ -17,6 +17,8 @@ start() {
   local branch base dirty behind ahead
   branch=$(git branch --show-current)
   [ -n "$branch" ] || stop "Detached HEAD — check out a branch first."
+  # Before any output: `start | head -3` closed the pipe and killed the script before a last-line mark.
+  bash "$HERE/ship-mark.sh" begin || stop "ship-mark.sh begin failed."
   if git remote | grep -qx origin; then git fetch -q origin 2>/dev/null; else echo "remote none: pr, merge and release cannot run"; fi
   base=$(base_branch)
   echo "branch $branch"
@@ -35,7 +37,6 @@ start() {
   else
     echo "gh NOT logged in: pr, merge and release cannot run"
   fi
-  bash "$HERE/ship-mark.sh" begin || stop "ship-mark.sh begin failed."
 }
 
 skills_report() {  # skills_report <files...>

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@spardutti/claude-skills",
-  "version": "2.60.1",
+  "version": "2.60.3",
   "description": "Guardrails for Claude Code — best-practice skills, planning commands, and delivery gates.",
   "type": "module",
   "repository": {