cyclecad 3.10.4 → 3.12.0

package/.github/workflows/pages.yml

@@ -0,0 +1,34 @@
+name: Deploy to GitHub Pages
+
+on:
+  push:
+    branches: [main]
+  workflow_dispatch:
+
+permissions:
+  contents: read
+  pages: write
+  id-token: write
+
+concurrency:
+  group: pages
+  cancel-in-progress: true
+
+jobs:
+  deploy:
+    runs-on: ubuntu-latest
+    environment:
+      name: github-pages
+      url: ${{ steps.deployment.outputs.page_url }}
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+      - name: Setup Pages
+        uses: actions/configure-pages@v5
+      - name: Upload artifact
+        uses: actions/upload-pages-artifact@v3
+        with:
+          path: '.'
+      - name: Deploy to GitHub Pages
+        id: deployment
+        uses: actions/deploy-pages@v4

package/CLAUDE.md

@@ -1193,9 +1193,6 @@ Pointer events on `#dialog-title` drag the whole dialog container via `position:
 - Groq Llama 3.3 70B (free, 30 RPM — better for repeat prompts)
 
 ### Pending
-- Retry `npm publish cyclecad@3.10.3` (earlier attempt interrupted; registry still shows 3.10.2)
-- Add gear/pulley/shaft templates (Task #7)
-- Verify ExplodeView redirects
 - The `block-only-in-chrome` user report — needs repro with actual prompt
 
 ### Collaboration notes
@@ -1204,3 +1201,351 @@ Pointer events on `#dialog-title` drag the whole dialog container via `position:
 - Clipboard delivery via `write_clipboard` — user pastes in Terminal
 - Verification via Claude-in-Chrome MCP (navigate → execute → screenshot → dom inspect)
 - Terminal is tier-click, so clipboard write is BLOCKED while Terminal is frontmost — switch to Chrome first
+
+## Session 2026-04-24 — GitHub Pages Pipeline Fix + Template Library Expansion
+
+### What shipped
+- **cyclecad@3.10.3** confirmed live on npm (earlier interrupted publish actually went through)
+- **cyclecad@3.10.4 patch prepared** (delivery command sent via clipboard heredoc, user ran it): 3 new AI Copilot templates
+  - `spur gear 20 teeth module 2 bore 10mm` → cylinder blank at OD with bore, correct DIN pitch math (`pitchDia = m*z`, `outsideDia = m*(z+2)`, default face width 10mm, default bore 8mm)
+  - `pulley 80mm bore 12mm` or `v-belt pulley 100mm` or `timing pulley 60mm` → disc with center bore, V-groove noted as next step (true groove needs revolve)
+  - `shaft 20mm dia 150mm long`, `stepped shaft 25mm 120mm`, `axle 16mm 200mm`, `spindle Ø10 80mm long` → simple or 2-step cylinder with fallback heuristic for bare `Nmm` values (smaller=dia, larger=length)
+- **GitHub Pages deploy pipeline migrated** — from auto-generated "Deploy from a branch" to explicit GitHub Actions workflow with `cancel-in-progress: true`
+
+### Pages deploy fix details
+- **Root cause of failed runs #151/#152**: concurrent deployment conflict. Commit `ccacf45` pushed while `746008a` was mid-deploy → HTTP 400 "in progress deployment". Build succeeded (31s) but deploy step failed in 3s with 3 annotations.
+- Self-resolved for subsequent runs (#153-#166 all green) but would recur on future rapid pushes.
+- **Fix shipped**: Added `.github/workflows/pages.yml` with:
+  ```yaml
+  permissions: { contents: read, pages: write, id-token: write }
+  concurrency: { group: pages, cancel-in-progress: true }
+  ```
+  Uses `actions/checkout@v4`, `actions/configure-pages@v5`, `actions/upload-pages-artifact@v3`, `actions/deploy-pages@v4`.
+- Added `.nojekyll` at repo root (harmless, prevents future Jekyll issues).
+- **Pages source setting flipped manually** via GitHub UI (via Chrome MCP): Settings → Pages → Source → "GitHub Actions" (was "Deploy from a branch"). Confirmed with "GitHub Pages source saved." banner.
+- Triggered manual `workflow_dispatch` run → Deploy #2 succeeded in 20s (13s deploy step). Deploy step output URL = `https://cyclecad.com/` confirming new pipeline is authoritative.
+- Old `pages build and deployment` workflow automatically stops firing once source is "GitHub Actions".
+- One remaining harmless warning: "Node.js 20 actions are deprecated" — affects the whole actions ecosystem, will resolve when major versions bump.
+
+### Template patching pattern (for future template additions)
+Templates go in `matchTemplate(prompt)` in `app/js/modules/ai-copilot.js`, inserted before `return null;` (currently ~line 594, was line 514 before gear/pulley/shaft block). Each template:
+1. Matches user prompt with regex
+2. Parses params with sensible defaults
+3. Returns array of plan steps: `sketch.start` → `sketch.circle/rect` → `ops.extrude` → optional `ops.hole` → `view.set iso` → `view.fit`
+4. All solids centered at origin, Y is extrude axis
+
+### Tested template prompts (all passing via Node eval)
+- `spur gear 20 teeth module 2` → m=2, Z=20, pitch Ø40, OD Ø44 ✓
+- `gear m=3 z=40 bore 10mm` → m=3, Z=40, pitch Ø120, OD Ø126, bore Ø10 ✓
+- `v-belt pulley 80mm bore 12` → Ø80x20 with bore Ø12 ✓
+- `shaft 20mm dia 150mm long` → Ø20 x 150mm ✓
+- `stepped shaft 25mm dia 120mm long` → main Ø25x72mm + reduced Ø21x48mm ✓ (60/40 split)
+- `axle 16mm 200mm` (bare Nmm heuristic) → dia=16 (min), len=200 (max) ✓
+- `spindle Ø10 80mm long` → Ø10 x 80mm ✓
+
+### ExplodeView AI Render diagnostic (cross-reference)
+User reported Nano Banana v2 renders ignoring source CAD object. Three stacked problems in `docs/demo/app.js` `buildPrompt()` (line 6595):
+1. Prompt wrapper `Photorealistic render of an industrial/mechanical product in ${scene}...` nests user's preservation intent grammatically, weakening it
+2. Nano Banana v2 is generative not inpainting — reference image is style guidance, not pixel-level preservation
+3. Product Photo preset suffix "studio lighting, minimal shadows" fights outdoor scene prompts
+Fix delivered as prompt rewrite (preservation-first pattern). Code fix as v304 prepared but NOT YET SHIPPED. See explodeview CLAUDE.md for full details.
+
+### Pending (next session)
+- Confirm cyclecad@3.10.4 publish went through (user ran command; verify with `curl -s https://registry.npmjs.org/cyclecad/latest | python3 -c 'import json,sys; print(json.load(sys.stdin)["version"])'`)
+- Ship ExplodeView v304 (buildPrompt fix for AI Render prompt hijacking)
+- Add more AI Copilot templates: mounting bracket variations, T-slot extrusions, bearing cutouts, ball/roller bearings
+- Eventually: true involute gear teeth via `sketch.polyline` (would require mini-executor support for polyline → geometry, currently polyline is a no-op)
+- Post viral LinkedIn draft (user approval pending) — see `~/explodeview/linkedin-post-viral.md`
+- Push ExplodeView route redirects (explodeview.com/docs/demo/ currently 404)
+- The `block-only-in-chrome` user report — still needs repro
+
+### Collaboration lessons learned this session
+- **GitHub Pages 3-annotation failures** ("Failed to create deployment status: 400" + "Deployment request failed...due to in progress deployment" + "Creating Pages deployment failed") = concurrent deploy conflict, NOT build failure. Fix: explicit workflow with `concurrency: group: pages, cancel-in-progress: true`.
+- **GitHub API path for Pages source**: `PUT /repos/{owner}/{repo}/pages` with `-f build_type=workflow` (workflow) or `legacy` (branch). Good fallback when Chrome MCP auth is unavailable — user just runs `gh api --method PUT /repos/vvlars-cmd/cyclecad/pages -f build_type=workflow` on Mac where `gh` is authenticated.
+- **Password entry via Chrome MCP is prohibited** — Claude will not sign in on user's behalf under any circumstance. User must sign in themselves; Claude can then operate in authenticated sessions.
+- **Chrome tier "read"** blocks `computer_use` clicks in the browser itself, but Claude-in-Chrome MCP tools (`mcp__Claude_in_Chrome__*`) DO work — they go through the extension.
+- **Use `find` tool with aria role** (e.g., `menuitemradio "GitHub Actions"`) rather than guessing pixel coordinates — much more reliable. First click attempt at (585, 452) missed the dropdown option, second via `ref_311` landed perfectly.
+- **After Pages source flip**, trigger `workflow_dispatch` immediately to prove the new pipeline actually serves traffic (the existing run only built the artifact — it didn't serve until source was flipped).
+- **Successful Pages deploy step** shows the deploy environment URL (e.g. `https://cyclecad.com/`) as a link under the job name — that's the tell that it worked end-to-end, not just a green check.
+
+## Session 2026-04-24 (continued) — ExplodeView v304 shipped + new roadmap item
+
+### What shipped
+- **ExplodeView v1.0.22 / v304** published to npm and deployed (commit `b47d6a9`, then `41471f3` for the version bump). `buildPrompt()` in `docs/demo/app.js` rewritten with preservation-first logic:
+  - Detects explicit preservation cues ("do not change", "preserve", "keep exactly", "same shape", etc.) → leads with strong reference-image anchor
+  - Detects scene-descriptive prompts (>40 chars + spatial language) → passes them to the model directly with only the preamble in front, drops the rigid "render of a product in X" wrapper
+  - Suppresses Product Photo / Technical preset suffixes when the scene has outdoor cues (forest, trail, sunset, etc.) so the studio suffix doesn't fight the scene
+  - Adds a "No Style" preset (`presetSuffixes.none = ''`) that skips the suffix entirely; new dashed-border button in `docs/demo/index.html` at line ~2145
+  - Version badge updated: "ExplodeView v304 — preservation-first AI Render"
+
+### New roadmap item — AI Engineering Analyst (MecAgent-parity)
+
+**Source**: user shared MecAgent demo screenshots in `/Users/sachin/Desktop/mec` (12 screenshots, 2026-04-24 at 11:49-11:51). The demo runs inside Inventor via MecAgent's Copilot panel and solves a complete bolted joint problem:
+
+Problem statement (exact wording from screenshots):
+> A steel plate is secured using 4 × M12 bolts (property class 10.9) in a preloaded configuration. The joint is subjected to: a shear force of 18 kN, an axial separating force of 18 kN, an in plane moment of 420 Nm. Given: friction coefficient between contact surfaces μ = 0.16, preload per bolt 39 kN, bolt circle diameter: 96mm, required safety factor against slipping: 1.5. Question: Verify whether the joint is safe with respect to: 1. slip resistance, 2. maximum tensile load in the most loaded bolt, 3. combined tension and shear in the bolts.
+
+MecAgent's response walks through three checks with full equations:
+- **Slip resistance**: `F_friction = z·Q_F·μ = 4·39000·0.16 = 24960 N` vs required `F_required = K_s·F_shear = 1.5·26000 = 39000 N`
+- **Tension with moment**: `F_axial,bolt = F_axial,total/z = 18000/4 = 4500 N`; moment contribution `F_moment = M·r/Σr² = 420·1000·48/(4·48²) = 2187.5 N`; `F_max,tensile = 6687.5 N`
+- **Combined**: preload adds on top → `F_tensile,total = 45687.5 N`; tensile stress `σ = F/A = 45687.5/84.3 = 542 MPa`; shear stress `τ = 6500/84.3 = 77 MPa`; Von Mises `σ_vm = √(σ²+3τ²) = 558 MPa`
+- Compared against 10.9 proof strength (~936 MPa) → verdict "safe"
+- **Critical UX detail**: cites "Analysis and Design of Machine Elements" by Wei Jiang (Wiley), page 85, with clickable "Open Document" buttons showing the exact textbook page
+
+**Why this matters**: cycleCAD's existing `ai-copilot.js` generates geometry (templates for gears, bolts, flanges), `fusion-simulation.js` has stubbed FEA, and `validate.designReview` returns A-F scores from geometry heuristics. None of them answer a natural-language engineering question with analytical methods and cited sources. This is the single biggest competitive gap vs MecAgent's pitch, and closing it makes cycleCAD's "agent-first" positioning real.
+
+**Scope v1 — bolted joint analysis (VDI 2230 / Shigley)**:
+- Module: `app/js/modules/ai-engineer.js` (~800-1200 lines)
+- Inputs: bolt grade (4.6–12.9), bolt count, thread size, preload, external loads (shear/tension/moment + BCD), friction coefficient, safety factor target
+- Outputs: slip-resistance check, per-bolt tension with moment redistribution, combined Von Mises stress, proof-strength comparison, pass/fail verdict
+- Rendering: LaTeX equations via KaTeX (faster than MathJax), stepwise narrative, green/red badges per check
+
+**Scope v2 — other machine elements**: gear pair (AGMA bending + pitting), shaft fatigue (Goodman/Soderberg), rolling bearing L10 life, fillet weld sizing
+
+**RAG / citations**:
+- User uploads machine-element PDF → chunk + embed in IndexedDB (no server, runs locally with @xenova/transformers MiniLM)
+- Bundled fallback: public-domain machine-element references + excerpts from DIN/ISO standards we can legally redistribute
+- Every equation in the response cites page + source
+
+**LLM layer**:
+- Same provider stack as ai-copilot.js (Claude Sonnet 4.6 / Gemini 2.0 Flash / Groq Llama 3.3 70B)
+- Tool-use pattern: model writes narrative, calls deterministic analytical JS functions for every number, never fabricates results. Analytical core is pure math (no LLM) — LLM is only for problem parsing + narrative.
+
+**UI**:
+- New "Engineer" tab in right panel alongside Properties / Chat / Guide
+- Reuses selected-assembly context, materials, and bolt patterns from Fastener Wizard as implicit inputs (so user doesn't have to re-enter them)
+
+### Pending (next session, updated)
+- **AI Engineering Analyst module** (above) — the biggest pending feature
+- More AI Copilot templates: mounting bracket variations, T-slot extrusions, bearing cutouts, ball/roller bearings
+- Eventually: true involute gear teeth via `sketch.polyline` (requires mini-executor support for polyline → geometry, currently polyline is a no-op)
+- ExplodeView compositing render (true pixel preservation — render 3D with transparent bg, send only background region to Nano Banana, composite in canvas)
+- Run cyclecad test agent in Chrome (`app/test-agent.html`, 113 tests) and fix failures
+- Dynamic version badge in cyclecad status bar (currently hardcoded v0.9.0 but npm is 3.10.4)
+- Wire cyclecad splash buttons (New Sketch / Open-Import / Text-to-CAD / Inventor Project) to actually trigger their actions
+- 138MB STEP import — server-side conversion path (server/converter.py) is the safer route than opencascade.js v293
+- Text-to-CAD with live preview
+- Photo-to-CAD reverse engineering
+- Docker compose local test
+- The `block-only-in-chrome` user report — still needs repro
+
+## Session 2026-04-24 (session 2) — Pentamachine Collaboration + Suite Positioning + Pentacad Extension
+
+### Big strategic updates
+
+**Suite positioning locked in:**
+cyclecad.com is now an umbrella brand — **cycleCAD Suite** — covering three products:
+- **cycleCAD** (parametric CAD modeller, existing, MIT)
+- **ExplodeView** (3D viewer / AR / AI render, existing, MIT)
+- **Pentacad** (CAM + 5-axis simulator + Kinetic Control bridge, NEW, AGPL-3 / commercial dual)
+
+Pentacad is a **cycleCAD module**, NOT a separate repo. Lives at `app/js/modules/pentacad*.js` following the existing `window.CycleCAD.Pentacad` IIFE pattern.
+
+**Matt / Pentamachine status — CRITICAL:**
+Matt agreed to send files without NDA, but he explicitly stated: *"We do provide feedback for 4+ companies that are working on similar CAM concepts to yours."*
+
+Implication: Matt is a **neutral ecosystem supplier**, not a partner. JV ask is dead. Strategy is pilot-first: build in silence, ship polished demos (not source, not architecture), public launch before he asks for exclusivity. Commercial progression: Phase 0-1 silent, Phase 2 license discussion, Phase 3 exclusive EU resale if warranted, JV only if Matt raises equity first. Full tactical playbook at `~/Documents/Penta/pentacad-notes/pilot-intelligence-brief.md` (private).
+
+### Pentacad extension shipped (scaffold)
+
+Files added to cycleCAD repo:
+| File | Purpose |
+|---|---|
+| `app/js/modules/pentacad.js` | Coordinator — registers sub-modules, machine picker UI, loadMachine(id), public API |
+| `app/js/modules/pentacad-cam.js` | 12 CAM strategies (stubs), post-processor skeleton, emitGCode() matching Pentamachine dialect |
+| `app/js/modules/pentacad-sim.js` | G-code parser (works for G20/G90/G94/G93/G54, tracks modal state + A/B axes), forward-kinematics stub, soft-limit check |
+| `app/js/modules/pentacad-bridge.js` | WebSocket client for controller bridge — connect/disconnect/stream/jog/pause/abort, DRO event handling, auto-reconnect |
+| `app/pentacad.html` | Dedicated UI entry point with split-screen viewport + machine picker + workspace tabs (Machine / Design / CAM / Simulate / Control) |
+| `machines/v2-50-chb/kinematics.json` | Template values marked `_confirmed: false` pending Matt's real data |
+
+### Suite landing mockup
+
+`mockups/cyclecad-suite-mockup.html` — full animated wireframe (not yet in production).
+
+Page structure:
+1. **Hero** — "From idea to finished part. One browser tab." with staggered fade-up, drifting orb gradients, blinking cursor
+2. **Lifecycle strip** — 5 stages with rainbow gradient line drawing across all chips on scroll-into-view
+3. **End-to-End Journeys** — 3 animated product examples showing the full business lifecycle (idea → design → market → sell → manufacture → feedback):
+   - 🔔 Custom bike bell (consumer, 6 days, €345, 23 pre-orders)
+   - ⚙️ Precision flange (B2B, 3 days, €2,400, 50-unit PO)
+   - 🍺 Branded bottle-opener ring (promo, 2 days, €1,900, 100 units)
+   - 15s loop, 2.5s per step, rows offset by one step each, hover to pause, click step to jump
+4. **5 stage deep-dives** — each with SVG animations (typing terminal, line-draw CAD, exploded parts, toolpath tracing, data-flow arrows)
+5. **Products grid** (3 cards — gold/teal/emerald per product), stats, CTA, founder note, footer
+
+Old `index.html` backed up as `index-agent-first.html.bak`. Production index.html NOT yet written — mockup is the design spec.
+
+### Pentamachine deliverables at `~/Documents/Penta/pentacad-deliverables/`
+- `ARCHITECTURE.md` — full system architecture (repo doc, shareable with Matt)
+- `Pentacad-Complete.pptx` — 25-slide deck (16 public architecture + 8 private intelligence brief, clearly divided)
+- `Pentacad-Complete.pdf` — same, non-editable
+- `pilot-intelligence-brief.md` — full tactical playbook
+
+Plus standalone pentacad repo scaffold at `~/Documents/Penta/pentacad/` (superseded by extension approach, kept as reference).
+
+### What Matt has sent (in `~/Documents/Penta/`)
+Marketing kit + installation guides + ONE real technical artifact: bottle-opener ring Fusion archive + 3 `.ngc` files confirming A/B kinematics, G20 imperial, G93 inverse-time feed, 40000 RPM spindle.
+
+**Still blocking Phase 0**: native 3D models of V2-10/V2-50CHB/V2-50CHK + kinematics JSON per machine.
+
+### Pending task inbox (tasks #10-24)
+
+**AI Engineering Analyst** (biggest competitive gap vs MecAgent):
+- #10 v1 — bolted-joint analysis (VDI 2230 / Shigley), pure-math core tested against MecAgent numbers, KaTeX rendering, LLM tool-use, RAG
+- #11 v2 — gears (AGMA), shafts (Goodman/Soderberg), bearings (L10), welds (throat stress)
+- #12 — RAG + citations (@xenova/transformers MiniLM, IndexedDB, "Open Document" footnotes)
+
+**cycleCAD quick wins** (ship 3.10.5 next):
+- #13 — more AI Copilot templates (brackets, T-slots 40/40 + 80/40, bearings, cutouts)
+- #14 — polyline → geometry in mini-executor (unblocks involute gear teeth)
+
+**cycleCAD other**:
+- #15 dynamic version badge · #16 wire splash buttons · #17 run test-agent.html (113 tests) · #18 text-to-CAD live preview · #19 photo-to-CAD · #20 server-side 138MB STEP · #21 docker compose test
+
+**ExplodeView**:
+- #22 compositing render (transparent background + Nano Banana + canvas composite, bypasses "generative ≠ inpainting") · #23 killer-features-test · #24 block-only-in-chrome repro
+
+**Suite website**:
+- Port mockup into production index.html once motion is approved
+- Build dedicated /pentacad.html marketing page in same style
+
+### Recommended next sprint
+#13 + #14 as warm-ups (ship 3.10.5 fast), then #10 as the big strategic work.
+
+### Handoff files for next chat
+- `~/cyclecad/HANDOFF-2026-04-24.md` (earlier session — AI Render v304 + Pages workflow)
+- `~/cyclecad/HANDOFF-2026-04-24-session-2.md` (this session — Pentacad + Matt + mockup)
+
+Both should be read together when resuming.
+
+## Session 2026-04-24 (session 3) — cyclecad 3.10.5 ship + AI Engineering Analyst v1
+
+### Sprint executed (per session-2 handoff recommendation)
+
+**Task #14 — Polyline support in mini-executor** ✅
+- `app/js/modules/ai-copilot.js` `sketch.polyline` method: accepts `{points:[[x,z],[x,z],...]}`, min 3 points
+- `ops.extrude` now handles `shape==='polyline'` via `THREE.Shape` + `THREE.ExtrudeGeometry`
+- Coordinate math: `Shape.moveTo(x, -z)` compensates for `rotateX(-π/2)` so user input `[x,z]` maps to world X,Z correctly
+- `g.translate(0, 0, -d/2)` before rotate to center along extrude axis, matching `BoxGeometry` / `CylinderGeometry` convention
+- `SYSTEM_PROMPT` updated so LLM knows about the method
+- Unblocks true involute gear teeth, real T-slot cross-sections, custom 2D cams/profiles
+
+**Task #13 — More AI Copilot templates** ✅
+Added 4 new templates to `matchTemplate()` in `app/js/modules/ai-copilot.js`:
+1. **Ball bearing (ISO 15 / DIN 625)** — lookup table of 20 standard deep-groove designations (608, 625, 6000-6006, 6200-6206, 6300-6304). Matches `"6200 bearing"`, `"6203 deep-groove bearing"`, `"608 bearing"`, etc.
+2. **Generic ball bearing with bore** — matches `"ball bearing 15mm bore"`. Uses approximation `OD ≈ 2.5·bore + 10, B ≈ 0.4·bore + 4` for 62xx-series sizing.
+3. **Bearing housing / pocket** — matches `"bearing housing"`, `"bearing pocket for 6204"`, `"bearing seat"`. Block with recess, shaft clearance bore, 4 mounting holes at corners. If a bearing designation is present, looks up its OD automatically.
+4. **T-slot aluminum extrusion** — matches `"2020 t-slot extrusion"`, `"4040 aluminum extrusion 1000mm"`, `"4080 profile"`, `"40x40 extrusion"`. Solid rect extrusion with 4 slot cuts (one per face). Double-wide profiles (4080) get extra slots on the long faces.
+5. **U-bracket / channel bracket** — matches `"U-bracket 60mm wide 40mm high 120mm long"`, `"channel bracket 80mm long"`. Box with rectangular pocket cut from top, leaves base + 2 side walls. Includes configurable mounting holes through base.
+
+Priority/ordering details:
+- Housing keyword detection (`bearing housing|pocket|seat|recess|mount`) runs BEFORE the bearing code regex so `"bearing pocket for 6204"` routes to the housing template with OD=47 looked up from the 6204 spec, not built as a bearing body.
+- T-slot length fallback: if no `"Nmm long"` keyword, take the largest `\d+mm` value that isn't the profile code (handles `"4040 extrusion 1000mm"`).
+
+### Smoke tests (run via `node /tmp/smoke.js` with stubbed globals)
+All 15 prompts match/reject correctly:
+- Specific bearing designations: `6200`, `6203`, `608` → bearing body w/ looked-up dims
+- Generic bearing w/ bore: `"ball bearing 15mm bore"` → approximated Ø48×10
+- Housing w/ explicit OD: `"bearing housing for a 32mm OD bearing"` → 48×48×15 block
+- Housing w/ designation: `"bearing pocket for 6204"` → 63×63×19 block (OD=47 from lookup)
+- T-slot variants: `2020`, `4040` (with 500mm, 1000mm, 600mm lengths)
+- U-bracket with/without dims and hole counts
+- Regression: spur gear + M8 nut still work
+- Unrelated prompt correctly returns `null`
+
+### Key files changed in 3.10.5
+| File | Delta | Notes |
+|------|-------|-------|
+| `app/js/modules/ai-copilot.js` | +190 lines | 4 templates + polyline support + docs |
+| `package.json` | version 3.10.4 → 3.10.5 | |
+| `CLAUDE.md` | This block | Session 3 notes |
+
+### Task #10 — AI Engineering Analyst v1 ✅ SHIPPED (3.11.0)
+
+**File**: `app/js/modules/ai-engineer.js` (~570 lines).
+
+**Architecture delivered:**
+- **Analytical core** (pure JS math, 0 deps) — `boltedJointAnalysis(params)` returns `{inputs, slipResistance, tensionCheck, combinedStress, verdict, verdictClass, notes}`. All numbers are computed deterministically — no LLM fabrication possible.
+- **Reference tables** (ISO/DIN data, frozen objects):
+  - `STEEL_GRADES` — 8 ISO 898-1 property classes (4.6 through 12.9) with R_m, R_p0.2, R_el
+  - `BOLT_STRESS_AREA` — DIN 13 A_s for 26 thread sizes M3–M56
+  - `BOLT_MAJOR_DIA` — nominal dia per thread
+  - `FRICTION_PRESETS` — VDI 2230 Table A6 surface conditions
+- **Natural-language parser** — `parseBoltedJointPrompt(text)` extracts boltCount, thread, grade, forces (kN/N), moment (Nm/kNm), preload, bcd, friction, safetyFactor. Tested against MecAgent's verbatim problem statement → all 10 params extracted correctly.
+- **UI** — dialog-mounted form with:
+  - Free-text prompt → "Parse" button that auto-fills all fields
+  - 11-field input grid with live recompute on any change
+  - 3 preset buttons (MecAgent demo, flange M8 light, heavy M20)
+  - Live report with verdict banner (green/yellow/red), 3 check sections (slip / tension / stress), KaTeX-rendered formulas, pass/fail messages per check
+  - Collapsible self-test panel showing the MecAgent reference comparison
+- **KaTeX integration** — loaded on demand from CDN (jsdelivr), 0.16.21. Non-blocking, falls back to LaTeX source text if CDN fails.
+- **Menu wiring** — Tools → `🔩 AI Engineering Analyst` opens the dialog.
+
+**Analytical verification (self-tests in module):**
+All 7 tests pass against MecAgent screenshot numbers within tolerance:
+```
+PASS: F_friction            actual=24960.00  expected≈24960
+PASS: F_moment/bolt         actual=2187.50   expected≈2187.5
+PASS: F_bolt_max_tang       actual=6687.50   expected≈6687.5
+PASS: F_bolt_total          actual=45687.50  expected≈45687.5
+PASS: σ (tensile)           actual=541.96    expected≈542
+PASS: τ (shear)             actual=79.33     expected≈79
+PASS: σ_vm                  actual=559.11    expected≈558
+overall: PASS
+```
+
+Self-tests run automatically on `init()` — warnings logged if any fail. Also available as `window.CycleCAD.AIEngineer.runSelfTests()`.
+
+**Verdict classification:**
+- `SAFE` — both slip and stress checks pass.
+- `SAFE (bearing-type)` — stress OK but slip fails. Bolts carry shear directly; common in practice.
+- `UNSAFE` — stress check fails. Bolt will yield.
+
+For the MecAgent problem with defaults it correctly returns "SAFE (bearing-type)" because σ_vm=559 < R_p0.2=830 but F_friction=24960 < F_required=39000.
+
+**Public API:**
+```js
+window.CycleCAD.AIEngineer = {
+  analyze(params),        // run analysis, returns structured result
+  parsePrompt(text),      // NL → params
+  runSelfTests(),         // {results[], allPass}
+  STEEL_GRADES, BOLT_STRESS_AREA, BOLT_MAJOR_DIA, FRICTION_PRESETS,
+  init(), getUI(),
+  execute(cmd, params)    // 'analyze' | 'parse' | 'show'
+};
+```
+
+**Scope v1 limits (to address in task #11 / #12):**
+- Load factor Φ simplified to 1 (conservative — real VDI 2230 uses 0.1–0.3 for gasketed joints).
+- Bolt-circle geometry assumes uniform spacing (moment redistribution formula M / (z·r) is exact for that case).
+- No fatigue check.
+- No ball-bearing / shaft / gear / weld analysis — v2 will add these.
+- No RAG / textbook citations — v3.
+
+### 3.11.0 files changed
+| File | Delta |
+|------|-------|
+| `app/js/modules/ai-engineer.js` | new, 570 lines |
+| `app/index.html` | +3 lines (menu entry, dispatch case, script tag) |
+| `package.json` | 3.10.5 → 3.11.0 |
+
+### Ship commands (both 3.10.5 + 3.11.0 in a single push)
+
+Commits landed in the VM:
+- `41d4862` — v3.10.5 (AI Copilot polyline + bearing / housing / T-slot / U-bracket templates)
+- HEAD (pending commit) — v3.11.0 (AI Engineering Analyst v1)
+
+Push + publish in one combined command (lock-file prefix handles the VM-generated `.git/*.lock` cruft):
+```bash
+rm -f ~/cyclecad/.git/HEAD.lock ~/cyclecad/.git/index.lock && \
+  cd ~/cyclecad && \
+  git push origin main && \
+  npm publish
+```
+
+If the 3.10.5 commit was also pending as of this handoff, `git push` covers both in one shot. The `npm publish` publishes 3.11.0 (latest version in `package.json`). 3.10.5 is not separately published to npm since 3.11.0 supersedes it — users bumping from 3.10.4 get both feature sets at once.
+
+### Pending after this session
+- #11 AI Engineering Analyst v2 — gears (AGMA bending + pitting), shafts (Goodman/Soderberg), bearings (L10), welds (throat stress)
+- #12 AI Engineering Analyst — RAG citations via `@xenova/transformers` MiniLM-L6-v2 + IndexedDB + "Open Document" footnote UI
+- All other pending tasks from session 2 handoff (#15-24)

package/HANDOFF-2026-04-24-session-2.md

@@ -0,0 +1,239 @@
+# Handoff — 2026-04-24, session 2 → next chat
+
+Drop this file into the next chat (or tell Claude to read it). Continues the earlier handoff from the same day; read both.
+
+## Who I am
+- **SACHIN** (vvlars@googlemail.com, GitHub vvlars-cmd)
+- Building the **cycleCAD Suite** — now three products: cycleCAD + ExplodeView + Pentacad
+- Working style: fast iteration, minimal clarifying questions, clipboard-paste commands via terminal, Safari private window for testing
+- Full persistent memory: `~/cyclecad/CLAUDE.md` and `~/explodeview/CLAUDE.md`
+
+---
+
+## THE BIG STRATEGIC UPDATES THIS SESSION
+
+### 1. Suite positioning crystallised
+**cycleCAD Suite** is the umbrella brand at cyclecad.com. Three products under it:
+
+| Product | What | License | Status |
+|---|---|---|---|
+| **cycleCAD** | Parametric 3D CAD modeller (the core) | MIT | Live |
+| **ExplodeView** | 3D viewer, AI render, AR, analysis | MIT | Live |
+| **Pentacad** | CAM + 5-axis simulator + Kinetic Control bridge | AGPL-3 / commercial dual | Phase 0 scaffolded |
+
+**Key decision**: Pentacad is built as a **cycleCAD module**, NOT a separate repo. Lives at `~/cyclecad/app/js/modules/pentacad*.js`. Follows the existing `window.CycleCAD.ModuleName` IIFE pattern used by every other cycleCAD module.
+
+### 2. Pentamachine / Matt status — CRITICAL
+**Matt agreed to send files without NDA**, but his reply included a red flag:
+
+> "We do provide feedback for 4+ companies that are working on similar CAM concepts to yours."
+
+**What this means:**
+- Matt is a **neutral ecosystem supplier**, not a partner candidate
+- He legally has no confidentiality obligation toward us
+- Ideas we share with him will influence his feedback to other teams (whether he means to or not)
+
+**Revised strategy:**
+- **Do not lead with JV** — the equity/partnership proposal is dead until Pentacad is a proven product
+- **Pilot-first**: build cycleCAD Suite Pentacad extension in silence, ship polished demos (not source, not architecture docs), public launch before Matt asks for exclusivity
+- Commercial conversation sequence: Phase 0-1 = silent, Phase 2 = open license discussion, Phase 3 = exclusive EU resale if warranted, JV only if Matt proactively offers equity
+
+**Full tactical playbook**: `~/Documents/Penta/pentacad-notes/pilot-intelligence-brief.md` (private, DO NOT share with Matt)
+
+---
+
+## What shipped this session
+
+### Pentacad extension inside cycleCAD (`~/cyclecad/`)
+New files, follow the existing cycleCAD IIFE module pattern (`window.CycleCAD.Pentacad*`):
+
+| File | Purpose |
+|---|---|
+| `app/js/modules/pentacad.js` | Coordinator module — registers sub-modules, machine picker UI, public API |
+| `app/js/modules/pentacad-cam.js` | 12 CAM strategies (stubs), post-processor skeleton for Pentamachine .ngc dialect |
+| `app/js/modules/pentacad-sim.js` | G-code parser skeleton, forward-kinematics stub for 5-axis A/B tables |
+| `app/js/modules/pentacad-bridge.js` | WebSocket client for controller bridge — jog/stream/DRO/pause/abort scaffolding |
+| `app/pentacad.html` | Dedicated UI entry point — split-screen viewport + machine picker + workspace tabs |
+| `machines/v2-50-chb/kinematics.json` | Template values marked `_confirmed: false` until Matt sends real data |
+
+Load via `<script src="./js/modules/pentacad-cam.js"></script>` etc. in `app/pentacad.html`. Coordinator auto-initialises on DOMContentLoaded unless `window.CycleCAD.SkipPentacadInit` is set.
+
+### cyclecad.com suite landing mockup (`~/cyclecad/mockups/`)
+- `cyclecad-suite-mockup.html` — full animated wireframe for the new suite landing page
+- `index-agent-first.html.bak` — backup of the previous index.html (agent-first OS-of-manufacturing framing)
+
+**Page structure** (top to bottom):
+1. **Hero** — "From idea to finished part. One browser tab." Staggered fade-up, animated kicker dot, drifting orbs, spinning logo
+2. **Lifecycle strip** — 5 stages (Ideate / Design / Present / CAM / Produce). Rainbow gradient line draws across all 5 chips when it scrolls into view
+3. **End-to-End Journeys section** — 3 animated product examples:
+   - 🔔 Custom bike bell (consumer, 6 days, €345, 23 pre-orders)
+   - ⚙️ Precision flange (B2B, 3 days, €2,400, 50-unit PO)
+   - 🍺 Branded bottle-opener ring (promo, 2 days, €1,900, 100 units)
+   - Each cycles through 6 lifecycle steps (idea → design → market → sell → manufacture → feedback) on a 15s loop with 2.5s per step, rows offset by one step, hover to pause, click step to jump
+4. **5 stage deep-dive sections** — each with SVG animations (typing terminal, line-draw CAD, exploded parts, toolpath tracing, data flow arrows)
+5. **Products grid** — 3 cards (cycleCAD gold / ExplodeView teal / Pentacad emerald)
+6. **Stats, CTA, founder note, footer**
+
+Production version **not yet written** — the mockup is the design spec. When approved, port into real `/index.html`.
+
+### Pentamachine deliverables (`~/Documents/Penta/pentacad-deliverables/`)
+- `ARCHITECTURE.md` — full Pentacad system architecture (repo doc, shareable)
+- `Pentacad-Complete.pptx` — 25-slide deck (16 architecture + 8 private intelligence brief + dividers)
+- `Pentacad-Complete.pdf` — same content, non-editable
+- `pilot-intelligence-brief.md` — full tactical playbook for managing Matt
+
+**Separate** private notes at `~/Documents/Penta/pentacad-notes/pilot-intelligence-brief.md`.
+
+### Standalone pentacad repo (SUPERSEDED but preserved)
+- Located at `~/Documents/Penta/pentacad/`
+- Full monorepo scaffold I built before the user pivoted to the "cycleCAD module" approach
+- Keep as reference — useful pieces (LICENSE, .gitignore, README, CLAUDE.md, web-app main.js, CAM module index, strategies registry) can be copied if ever needed
+- Do not commit as a separate repo
+
+### Emails drafted (for Matt, Penta Machine Company)
+1. **Short JV proposal** — not used; user went pilot-first instead
+2. **Long-form detailed proposal** — kept in reserve
+3. **Follow-up acknowledging 4+ competitors** — softer tone, pilot framing
+4. **Phase 0-3 file request** — the canonical list of what Matt needs to send
+
+---
+
+## What Matt already sent vs still missing
+
+**In `~/Documents/Penta/` right now:**
+- Marketing kit (logos, spec PDFs, product videos, photos)
+- Installation guides (.docx) for V2-10 and V2-50
+- **One real technical artifact**: `Sample Part for Demonstration/Ring Aluminum V2-50/`
+  - Fusion 360 archive (`.f3z`) with CAM setup
+  - 3 × `.ngc` files — 5-axis G-code confirming A/B kinematics, G20 imperial, G93 inverse-time feed, 40000 RPM spindle
+
+**Still blocking Phase 0 (the foundation work):**
+1. Native 3D models of V2-10, V2-50CHB, V2-50CHK (STEP or Fusion/Inventor) — need for digital twin
+2. Machine kinematics JSON (A/B offsets, travel, home, tool-length zero, soft limits)
+
+**Needed for later phases:**
+3. LinuxCNC version / Kinetic Control fork reference + HAL / INI files
+4. Fusion 360 post-processor (`.cps`) — to match the dialect exactly
+5. Tool library (ER11/ER40 holders, tool numbers, max RPM)
+6. Feeds & speeds defaults per material
+7. Kinetic Control IPC/WebSocket API surface
+8. Simulator source (if separate from control)
+
+**Status**: Matt agreed to send items 1-8, awaiting delivery. Safe to continue cycleCAD and ExplodeView work in parallel.
+
+---
+
+## Pending task inbox (15 items, from tracker #10-24)
+
+### Priority strategic work
+- **#10** · AI Engineering Analyst v1 — bolted-joint analysis. Biggest competitive gap vs MecAgent. Target: `app/js/modules/ai-engineer.js` (~800-1200 lines). Analytical-core-first (pure JS math, unit-tested against MecAgent screenshot numbers: F_friction=24960N, F_max_tensile=6687.5N, σ_vm=558MPa), then KaTeX rendering, then LLM tool-use layer, then RAG.
+- **#11** · AI Engineering Analyst v2 — gears (AGMA), shafts (Goodman/Soderberg), bearings (L10), welds (throat stress)
+- **#12** · AI Engineering Analyst — RAG + citations (@xenova/transformers MiniLM-L6-v2, IndexedDB, MecAgent-style "Open Document" footnotes)
+
+### cycleCAD quick wins
+- **#13** · Ship 3.10.5 — more AI Copilot templates (mounting brackets, T-slot extrusions 40/40 + 80/40, ball/roller bearings, bearing cutouts). Same pattern as gear/pulley/shaft block in 3.10.4.
+- **#14** · Polyline → geometry support in mini-executor. Unblocks true involute gear teeth.
+
+### cycleCAD
+- **#15** · Dynamic version badge in status bar (currently hardcoded v0.9.0)
+- **#16** · Wire splash screen buttons (New Sketch / Open-Import / Text-to-CAD / Inventor Project)
+- **#17** · Run `app/test-agent.html` (113 tests, 15 categories), fix failures
+- **#18** · Text-to-CAD with live preview
+- **#19** · Photo-to-CAD reverse engineering
+- **#20** · Server-side 138MB STEP import (FastAPI converter path, safer than opencascade.js v293)
+- **#21** · Docker compose local test
+
+### ExplodeView
+- **#22** · Compositing render for guaranteed CAD preservation. Post-v304 follow-up. Render 3D with transparent background, send only cropped background region to Nano Banana, composite 3D foreground over generated background in canvas. Bypasses "generative ≠ inpainting" limitation entirely.
+- **#23** · Run `docs/demo/killer-features-test.html`, fix failures
+- **#24** · Repro `block-only-in-chrome` ExplodeView bug (user report, unclear trigger)
+
+### Suite website
+- Port the `mockups/cyclecad-suite-mockup.html` design into production `/index.html` once motion is approved
+- Build dedicated `/pentacad.html` product marketing page in the same style
+
+---
+
+## Recommended next sprint
+
+**Cheap high-impact**: #13 + #14. Ships 3.10.5, unlocks real gear teeth.
+**Strategic**: #10 — AI Engineering Analyst v1. Biggest competitive moat, direct MecAgent parity.
+**While waiting on Matt**: #22 (compositing render for ExplodeView) and #17 (run cycleCAD test agent).
+
+Do #13 + #14 as warm-up, then go big on #10.
+
+---
+
+## Critical operational context (unchanged from earlier handoff)
+
+- **Git lock dance**: VM commits leave stale `.git/*.lock` files. Always give ONE combined command:
+  ```bash
+  rm -f ~/REPO/.git/index.lock ~/REPO/.git/HEAD.lock && cd ~/REPO && git add ... && git commit -m "..." && git push origin main
+  ```
+- **Clipboard delivery**: Commands via `mcp__computer-use__write_clipboard` → user pastes in Terminal. Must switch OFF Terminal (make Chrome or another app frontmost) before writing — Terminal is tier-"click" and blocks clipboard writes when frontmost.
+- **Egress allowlist**: `explodeview.com` and `cyclecad.com` are NOT on the allowlist. Use `npm registry` (allowed) for version verification. For live-site checks, ask user to add domain or use Claude-in-Chrome MCP.
+- **Deferred tools**: Schemas load on demand. Load computer-use in bulk `{query: "computer-use", max_results: 30}`, chrome `{query: "chrome", max_results: 20}`.
+- **GitHub Pages concurrent-deploy fix**: shipped 2026-04-24 in `.github/workflows/pages.yml` with `concurrency: group: pages, cancel-in-progress: true`.
+
+---
+
+## Current npm versions
+
+- `cyclecad`: **3.10.4** (next: 3.10.5 with AI Copilot templates, OR 3.11.0 with AI Engineering Analyst)
+- `explodeview`: **1.0.22** / v304 (preservation-first `buildPrompt()` rewrite shipped earlier this session)
+
+## Latest commits
+
+- ExplodeView: `41471f3` v1.0.22 tag, `b47d6a9` v304 code
+- cycleCAD: HEAD clean before session; **unstaged changes this session** in:
+  - `app/js/modules/pentacad.js` + `pentacad-cam.js` + `pentacad-sim.js` + `pentacad-bridge.js` (new)
+  - `app/pentacad.html` (new)
+  - `machines/v2-50-chb/kinematics.json` (new)
+  - `mockups/cyclecad-suite-mockup.html` (new)
+  - `index-agent-first.html.bak` (new)
+  - `HANDOFF-2026-04-24-session-2.md` (this file, new)
+
+Nothing pushed yet — waiting on user to commit.
+
+---
+
+## Suggested commit + push
+
+```bash
+rm -f ~/cyclecad/.git/index.lock ~/cyclecad/.git/HEAD.lock && \
+  cd ~/cyclecad && \
+  git add app/js/modules/pentacad.js \
+          app/js/modules/pentacad-cam.js \
+          app/js/modules/pentacad-sim.js \
+          app/js/modules/pentacad-bridge.js \
+          app/pentacad.html \
+          machines/v2-50-chb/kinematics.json \
+          mockups/cyclecad-suite-mockup.html \
+          HANDOFF-2026-04-24-session-2.md \
+          index-agent-first.html.bak && \
+  git commit -m "Pentacad extension scaffold + suite landing mockup
+
+- app/js/modules/pentacad*.js: 4 new modules (coordinator, CAM, sim, bridge)
+- app/pentacad.html: dedicated Pentacad UI entry point
+- machines/v2-50-chb/: template kinematics JSON (awaiting real data from Matt)
+- mockups/cyclecad-suite-mockup.html: animated suite landing page with 3 end-to-end journey examples
+- HANDOFF-2026-04-24-session-2.md: session handoff" && \
+  git push origin main
+```
+
+---
+
+## How to resume in the next chat
+
+1. Read `~/cyclecad/CLAUDE.md` (has been updated with suite positioning + Matt status + Pentacad extension)
+2. Read `~/explodeview/CLAUDE.md` (not modified this session, earlier context still current)
+3. Read `~/cyclecad/HANDOFF-2026-04-24.md` (the earlier session today — AI Render fix, GitHub Pages workflow)
+4. Read **this file** for session-2 context
+5. Read `~/Documents/Penta/pentacad-notes/pilot-intelligence-brief.md` if anything Matt-related comes up
+6. Open `~/cyclecad/mockups/cyclecad-suite-mockup.html` in Chrome to see the suite landing design
+7. Review deliverables at `~/Documents/Penta/pentacad-deliverables/`
+8. Ask me which of tasks #10-24 to start with, or just kick off #13 + #14 + #10 as the recommended sprint
+
+---
+
+*File generated at end of session 2 on 2026-04-24. Next chat picks up from here.*