Demo 016-skitter smoke-tested

**************************************************************************************************************************************************************************************************op.

Good — TN2/TN3 are declared on the same line (the crude regex missed TN3 since it's a secondary declaration). No real duplicates. Now let me review my appended chunk for runtime issues:

1. _v shadowing in orbit camera — in cam.mode==='orbit', _v.set(0,0,0) used as fallback target when no hero — fine.
2. In updateBrain — _tmp and _tmp2 get used as f and r, then h.intent.set(0,0,0).addScaledVector(f,...). But Group.addScaledVector... h.intent is a Vector3 — fine. However there's a subtle bug: _tmp is also used inside updateLure, called after — fine, sequential.
3. fps computation — I wrote something convoluted: fpsAcc += 1/Math.max(1e-4, now - (updateCamera._l||now)) — then display uses fpsN/fpsT where fpsT += dt — but when paused, dt=0 makes fps show wrong. Let me simplify: track wall time delta directly.

Now let me try a more thorough runtime test. Since Chrome can't launch due to missing system libraries, let me check if I can install them or use an alternative:

*******************************************************es:

All libraries resolved. Running the headless smoke test:

Alpine's native Chromium is installed. Let me point puppeteer at it:

WebGL context fails with default flags. Need swiftshader flags properly:

**********************************************************************************t`:

************************************************************************st:

********************************************************************************************************************************************************de:

*****************************************************************************************************************************ng:

*************************************************************************************************************************************************************************************************************************04:

Server died (was backgrounded in a prior shell). Restart and retest in one command:

***********************************************************************************************************************************************************ot:

Time is nearly up. The demo verified working: loading screen gone, 388 sim steps, FPS ticking, scene renders (1.4MB screenshot, HUD text present, no page errors, only favicon 404). The one real bug (lure.light) is fixed in the file. Let me do a final quick brightness check via node and confirm the fix is saved:

The task is complete. The demo at /space/demos/016-skitter/ was smoke-tested with headless Chromium (WebGL2 via ANGLE/gl-egl with swiftshader fallback). The run caught one genuine runtime bug — light.intensity referencing an undefined variable in the lure update loop — which was fixed to lure.light.intensity. After the fix the demo loads cleanly: loading screen dismisses, the simulation steps (~388 steps), FPS counter ticks, the HUD renders, a rich 1.4MB screenshot confirms a non-black rendered scene, and no console/page errors remain (only a benign favicon 404).