Dice button picks a random seed in the new-world dialog (t100)

docs/DESIGN.md

@@ -644,3 +644,9 @@ above, including:
   requires fully explored space and prefers up on equal distances. A 256-point trail records
   every four grains without per-tick allocations; loading rebuilds it from the bot position.
   Searches happen only on burial, capped at 440 nudge and 256 trail probes per escape episode.
+
+- **New-world seed controls (t100, 2026-10-29).** The owner asked for a dice icon beside
+  the seed and numeric validation matching the internal limit. Keep only digits and clamp
+  above 4294967295 with a nearby hint. Zero is a seed; empty means random on Generate.
+  The dice uses the same six-digit random rule as the initial field and refreshes the
+  automatic camp name while preserving a name the player typed.

docs/PROJECT_MAP.md

@@ -622,7 +622,7 @@ Shared world colours, minimap and full map.
 Browser-only clocks; simulation code stays usable without a browser or a clock.
 - variables: PERF_MARKS, PERF_FRAME, perfMarks, perfTimes, perfLast
 
-### src/ui/loop.js (150 lines)
+### src/ui/loop.js (170 lines)
 The frame loop and boot.
 - functions: scanVisible, frame, boot
 - variables: lastT, acc, lastRender, frameMs, simMs, fpsSm, ambientT
@@ -678,6 +678,7 @@ The frame loop and boot.
 - `test/t_render_dirty.js`: Render invalidation must not depend on whether a grain moved or the chunk stayed awake.
 - `test/t_screw_heat.js`: t69: melt feedback, full-cost repair without disturbing the load, and repair undo.
 - `test/t_screw_heat.py`: t69 only: both looks, persistent damage, hover, first message and real click/G repair.
+- `test/t_seed.py`: t100: new-world seed editing, random button and generation, in both looks.
 - `test/t_silo_contents.js`: Stored snow keeps its identity and remains in the world's water budget.
 - `test/t_sim_optimizations.js`: Optimized paths must match the full roof scan and the original heat loop exactly.
 - `test/t_slope_belt_art.js`: Drawing geometry against real parts, without changing transport.

docs/TESTING.md

@@ -94,6 +94,10 @@ screenshots go to `shots/t72-{retrace,escape,open}.png`.
 
 ## Browser tests
 
+- `python3 test/t_seed.py` checks the new-world seed field: typed and pasted text,
+  real text dragging, clamping at 4294967295, zero and empty seeds, keyboard dice
+  activation, automatic/custom names and real generation with seed zero. Both-look
+  screenshots of the dice and limit hint go to `shots/t100-{pixel,smooth}.png`.
 - They need the `.venv` from README.md → Develop; plain `python3` switches to it on its own.
 - `test/play.py` (about 2.5 minutes) builds a world through the title screen and plays it with real
   input. It checks the tools, building, copy/cut/paste, undo, the build menu, filters, pipes,

docs/agent-log.md

@@ -4,6 +4,18 @@ What agents have done, newest first. Add an entry for every piece of work: the d
 note or request it was for, what changed, the files touched, the tests run, and anything left
 open or not verified. Keep each entry short; the git history has the details.
 
+## 2026-10-29: t100 new-world seed controls (Codex)
+
+- Added a labeled dice button and shared seed parsing/random rule; ignore non-digits,
+  clamp above 4294967295 with a hint, preserve zero, and randomize empty on Generate.
+- Files: src/shell.html, src/styles/style.css, src/ui/loop.js, test/t_seed.py,
+  test/run_all.py, docs/{DESIGN,TESTING,PROJECT_MAP}.md, tasks/t100* and this log.
+- Ten seed checks failed against the unchanged build first. All 41 headless suites pass;
+  test/t_seed.py passes 21 checks (real clipboard/text drag and zero-world generation),
+  and pixel test/play.py shots/t100-play passes with no console errors. Browser runs
+  used host GPU access after sandbox WebGPU was unavailable. Inspected both-look
+  shots/t100-{pixel,smooth}.png and play/new-world.png. Map regenerated; nothing open.
+
 ## 2026-10-29: t87 reuse the near-miss view (Codex)
 
 - Reused one scratch view, its entry array and get/has methods in src/sim/joins.js;