v3.0.0 - five phase commands, opt-in gates, and evals

.claude-plugin/marketplace.json

@@ -4,18 +4,18 @@
     "name": "ValeroK"
   },
   "metadata": {
-    "description": "Plan-first, test-first workflow + bootstrap flow for new projects. Two skills: one for daily feature work inside an existing codebase (the-ultimate-workflow-guidelines), one for greenfield project kickoff (project-bootstrap-guidelines). Behavioral principles derived from Andrej Karpathy's observations on LLM coding pitfalls. Ships Cursor rules under rules/ and Agent Skills under skills/.",
-    "version": "2.6.0"
+    "description": "Plan-first, test-first workflow for AI coding assistants. Five phase commands (plan, tests, build, review, harvest) that each run in a fresh context with scoped tools, two skills for feature work and greenfield bootstrap, and opt-in lifecycle gates. Behavioral principles derived from Andrej Karpathy observations on LLM coding pitfalls.",
+    "version": "3.0.0"
   },
   "plugins": [
     {
-      "name": "the-ultimate-workflow-guidelines",
+      "name": "ultimate-workflow",
       "source": {
         "source": "github",
         "repo": "ValeroK/the-ultimate-workflow-guidelines"
       },
-      "description": "Plan-first, test-first workflow + bootstrap flow for new projects. Two skills: one for daily feature work inside an existing codebase (the-ultimate-workflow-guidelines), one for greenfield project kickoff (project-bootstrap-guidelines). Behavioral principles derived from Andrej Karpathy's observations on LLM coding pitfalls. Ships Cursor rules under rules/ and Agent Skills under skills/.",
-      "version": "2.6.0",
+      "description": "Plan-first, test-first workflow for AI coding assistants. Five phase commands (plan, tests, build, review, harvest) that each run in a fresh context with scoped tools, two skills for feature work and greenfield bootstrap, and opt-in lifecycle gates. Behavioral principles derived from Andrej Karpathy observations on LLM coding pitfalls.",
+      "version": "3.0.0",
       "author": {
         "name": "ValeroK"
       },

.claude-plugin/plugin.json

@@ -1,12 +1,22 @@
 {
-  "name": "the-ultimate-workflow-guidelines",
-  "description": "Plan-first, test-first workflow + bootstrap flow for new projects. Two skills: one for daily feature work inside an existing codebase (the-ultimate-workflow-guidelines), one for greenfield project kickoff (project-bootstrap-guidelines). Behavioral principles derived from Andrej Karpathy's observations on LLM coding pitfalls. Ships Cursor rules under rules/ and Agent Skills under skills/.",
-  "version": "2.6.0",
+  "name": "ultimate-workflow",
+  "description": "Plan-first, test-first workflow for AI coding assistants. Five phase commands (plan, tests, build, review, harvest) that each run in a fresh context with scoped tools, two skills for feature work and greenfield bootstrap, and opt-in lifecycle gates. Behavioral principles derived from Andrej Karpathy observations on LLM coding pitfalls.",
+  "version": "3.0.0",
   "author": {
     "name": "ValeroK"
   },
   "homepage": "https://github.com/ValeroK/the-ultimate-workflow-guidelines",
   "repository": "https://github.com/ValeroK/the-ultimate-workflow-guidelines",
   "license": "LicenseRef-MIT-Attribution",
-  "keywords": ["workflow", "planning", "tdd", "bootstrap", "prd", "karpathy", "guidelines", "best-practices", "coding"]
+  "keywords": [
+    "workflow",
+    "planning",
+    "tdd",
+    "bootstrap",
+    "prd",
+    "karpathy",
+    "guidelines",
+    "best-practices",
+    "coding"
+  ]
 }

.cursor-plugin/plugin.json

@@ -1,7 +1,7 @@
 {
-  "name": "the-ultimate-workflow-guidelines",
-  "version": "2.6.0",
-  "description": "Plan-first, test-first workflow + bootstrap flow for new projects. Two skills: one for daily feature work inside an existing codebase (the-ultimate-workflow-guidelines), one for greenfield project kickoff (project-bootstrap-guidelines). Behavioral principles derived from Andrej Karpathy's observations on LLM coding pitfalls. Ships Cursor rules under rules/ and Agent Skills under skills/.",
+  "name": "ultimate-workflow",
+  "version": "3.0.0",
+  "description": "Plan-first, test-first workflow for AI coding assistants. Five phase commands (plan, tests, build, review, harvest) that each run in a fresh context with scoped tools, two skills for feature work and greenfield bootstrap, and opt-in lifecycle gates. Behavioral principles derived from Andrej Karpathy observations on LLM coding pitfalls.",
   "author": {
     "name": "ValeroK"
   },

.github/workflows/release-skills.yml

@@ -51,10 +51,29 @@ jobs:
             -x '.github/*' \
             -x 'dist/*' \
             -x '.gitignore' \
+            -x 'CLAUDE.md' \
             -x 'AGENTS.md' \
-            -x 'CLAUDE.md'
+            -x 'research/*.pdf' \
+            -x 'fixtures/*' \
+            -x 'evals/results/*' \
+            -x 'internal/*'
           ls -la dist/
 
+      - name: Fail if the plugin ZIP has grown past what a plugin should be
+        # research/new-sdlc-with-vibe-coding.pdf is 9.97 MB of a 10.6 MB repo,
+        # and for one release it shipped inside this ZIP -- the file Claude
+        # Desktop and offline Cursor users download -- taking it from under 1 MB
+        # to roughly 10 MB for a third-party binary no skill, workflow, hook or
+        # test reads. Only its 9 KB summary is indexed. Nobody noticed because
+        # nothing looked at the size.
+        run: |
+          BYTES=$(stat -c%s dist/the-ultimate-workflow-guidelines-plugin.zip)
+          echo "Plugin ZIP: $BYTES bytes"
+          if [ "$BYTES" -gt 2000000 ]; then
+            echo "::error::Plugin ZIP is $BYTES bytes, over the 2 MB ceiling. Something large was added to the repo root; exclude it above or keep it out of the repo."
+            exit 1
+          fi
+
       - name: Upload to GitHub Release (create if missing)
         env:
           GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}

.github/workflows/validate.yml

@@ -6,6 +6,21 @@ on:
     branches: [main]
 
 jobs:
+  test:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: actions/setup-node@v4
+        with:
+          node-version: '22'
+
+      - name: Hook gate unit tests
+        # The glob is quoted so Node expands it, not the shell -- identical on
+        # Windows and Linux. `node --test hooks/` does NOT work: a directory
+        # argument is resolved as a module path, not a search root.
+        run: node --test "hooks/**/*.test.js" "evals/**/*.test.js" mirrors.test.js
+
   validate:
     runs-on: ubuntu-latest
     steps:
@@ -49,3 +64,33 @@ jobs:
             exit 1
           fi
           echo "All four version slots agree on $CANON."
+
+      - name: Manifest name consistency
+        # The plugin name namespaces every skill, command, and workflow it
+        # ships, so a mismatch silently changes how users invoke them.
+        # The marketplace's own top-level name is deliberately NOT checked:
+        # it identifies the marketplace, not the plugin.
+        run: |
+          set -e
+          CANON=$(jq -r '.name' .claude-plugin/plugin.json)
+          echo "Canonical plugin name: $CANON"
+
+          CURSOR=$(jq -r '.name' .cursor-plugin/plugin.json)
+          MKT=$(jq -r '.plugins[0].name' .claude-plugin/marketplace.json)
+
+          fail=0
+          for pair in \
+            ".cursor-plugin/plugin.json:$CURSOR" \
+            ".claude-plugin/marketplace.json#plugins[0].name:$MKT"; do
+            label="${pair%%:*}"
+            value="${pair##*:}"
+            if [ "$value" != "$CANON" ]; then
+              echo "::error::Name mismatch at $label: got '$value', expected '$CANON'"
+              fail=1
+            fi
+          done
+
+          if [ "$fail" -ne 0 ]; then
+            exit 1
+          fi
+          echo "All plugin name slots agree on $CANON."

.gitignore

@@ -11,3 +11,7 @@ dist/
 # OS noise
 .DS_Store
 Thumbs.db
+
+
+# Gate heartbeat: per-machine runtime evidence, not source.
+.ultimate-workflow/