Skip to content

Official Codebase of 3DCodeVerse

Python

Standardize CLI as 3dcodeverse and streamline regression tests

gaoypengfc165dfe3883
diff --git a/CLAUDE.md b/CLAUDE.md
index fe964661..b7a88138 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -6,7 +6,7 @@ The harness lives in **`harness/`** — read `harness/CLAUDE.md` +
`cd harness && python -m pytest tests -q -m "not live"`.
**`eval/`** evaluates things and is never imported by the harness: `eval/bench` evaluates the HARNESS
-(batteries — `cd eval && python -m bench.run_bench <battery.yaml>`, which replaced `3dcode bench` —
+(batteries — `cd eval && python -m bench.run_bench <battery.yaml>` —
A/B rigs, reports; its tests: `cd eval && python -m pytest`), `eval/llm` evaluates a bare LLM/VLM
(was `finetune/3dcodeverse_eval`). Read `eval/README.md` first.
diff --git a/README.md b/README.md
index e0bc89c1..651841aa 100644
--- a/README.md
+++ b/README.md
@@ -84,11 +84,11 @@ languages). Full guide: [harness/docs/INSTALL.md](harness/docs/INSTALL.md).
```bash
cd harness
-bash setup.sh # pip install -e . + runtime_js + headless Chrome + the agy / claude-code / codex CLIs at their latest, then `3dcode doctor`
+bash setup.sh # pip install -e . + runtime_js + headless Chrome + the agy / claude-code / codex CLIs at their latest, then `3dcodeverse doctor`
export GEMINI_API_KEYS="key1,key2" # agy runs in Gemini-API mode on this key pool
-3dcode doctor # every row OK or WARN; a FAIL row says what to fix
-3dcode make "a classic park bench" --track static_object --language threejs
-3dcode gallery serve # browse every run at http://127.0.0.1:8765/
+3dcodeverse doctor # every row OK or WARN; a FAIL row says what to fix
+3dcodeverse make "a classic park bench" --track static_object --language threejs
+3dcodeverse gallery serve # browse every run at http://127.0.0.1:8765/
```
The optional generators claude-code and codex use your local subscriptions: `--generator claude-code:<model>`
diff --git a/eval/bench/compare_backends.py b/eval/bench/compare_backends.py
index 96cc185d..f18328e3 100644
--- a/eval/bench/compare_backends.py
+++ b/eval/bench/compare_backends.py
@@ -316,7 +316,7 @@ def _run_harness(spec: Spec, cell: Path, eval_ws: Workspace, opts: CompareOption
with (exclusive(run_ws.root, what=f"compare {spec.id}:{run_ws.root.parent.name}"),
run_ledger(run_ws.root, run=f"{spec.id}:{run_ws.root.parent.name}")):
rec = deps.run_track(spec, run_ws, resume)
- hand_over(run_ws, rec) # what `3dcode make` delivers: the picked round, packaged
+ hand_over(run_ws, rec) # what `3dcodeverse make` delivers: the picked round, packaged
summary = select.summarise(run_ws.root, record=rec)
res.gen_cost_usd = rec.total_usage.cost_usd
res.tool_calls = rec.total_usage.tool_calls
diff --git a/eval/bench/judge_calib_graphics.py b/eval/bench/judge_calib_graphics.py
index c06fb279..f4d2f05d 100644
--- a/eval/bench/judge_calib_graphics.py
+++ b/eval/bench/judge_calib_graphics.py
@@ -13,7 +13,7 @@ person sees → revise) never ran for graphics. This script is that loop's meas
curator screenshot it / does it look like the thing"). The ground truth this calibrates to.
For every run and every rubric it rebuilds the JudgeInput the loop used — ``build_judge_input``,
-the ``3dcode judge`` payload (the picked round's judged views, gates, plan digest and acceptance
+the ``3dcodeverse judge`` payload (the picked round's judged views, gates, plan digest and acceptance
list, previous verdict, frame metrics) — and judges it with a fresh ``VlmJudge`` (n samples, pro). Output per rubric: ``<rubric>.jsonl`` (one
row per run: overall, per-criterion, defects present, caps) and one ``summary.md`` with
Spearman(judge, eye), mean judge vs mean eye, the biggest disagreements, and defect firing rates.
@@ -40,7 +40,7 @@ OUT_ROOT = Path(__file__).resolve().parent / "out"
def judge_one(run: Path, rubric: str, model: str, n: int) -> dict[str, Any]:
- """Re-judge the round ``codeverse3d.addons.select`` picks on the payload ``3dcode judge`` builds
+ """Re-judge the round ``codeverse3d.addons.select`` picks on the payload ``3dcodeverse judge`` builds
(``build_judge_input``: the harness's ``round_input`` from disk), with ``rubric`` swapped in."""
ws = Workspace(run)
try:
diff --git a/eval/bench/report.py b/eval/bench/report.py
index b4792be2..277296ac 100644
--- a/eval/bench/report.py
+++ b/eval/bench/report.py
@@ -1,11 +1,11 @@
"""Aggregate a bench run: per-tier / per-category stats → markdown + HTML gallery.
- cd eval && python -m bench.report bench/out/static_objects_v1 (was ``3dcode bench report``)
+ cd eval && python -m bench.report bench/out/static_objects_v1 (was ``3dcodeverse bench report``)
``build_report(out_dir)`` reads ``results.jsonl`` (or results.json), aggregates
baseline / picked-round mean + median, the delta and cost per tier and category (no pass
rate: a run is not passed or failed since 2026-09-22), writes ``report.md`` and
-``report.html`` — the same self-contained page as ``3dcode gallery build --embed``
+``report.html`` — the same self-contained page as ``3dcodeverse gallery build --embed``
(``codeverse3d.addons.gallery``), one section per tier, with the stats tables under the
summary strip.
"""
diff --git a/eval/bench/run_bench.py b/eval/bench/run_bench.py
index 2d3e5df3..2e72271f 100644
--- a/eval/bench/run_bench.py
+++ b/eval/bench/run_bench.py
@@ -2,7 +2,7 @@
cd eval && python -m bench.run_bench bench/prompts/<battery>.yaml --generator ... --judge ...
-(``3dcode bench run`` until 2026-09-22. Like every bench script it imports THIS tree's
+(``3dcodeverse bench run`` until 2026-09-22. Like every bench script it imports THIS tree's
``codeverse3d``; the console script resolved the editable install's, main's from a worktree.)
Layout of ``out_dir``::
@@ -16,11 +16,11 @@ Every prompt becomes a ``Spec`` with the battery's track/language, a FIXED judge
model (methodology: paired runs share the judge), and the generator under test.
Each prompt runs inside its own ``codeverse3d.cost.run_ledger``, exactly like a
-``3dcode make``: the batteries are where most runs come from, so without it the
+``3dcodeverse make``: the batteries are where most runs come from, so without it the
priced per-call rows of a whole battery went to the per-process fallback log and
-``3dcode cost --runs-dir <out>/runs`` had to reconstruct them from trajectories.
+``3dcodeverse cost --runs-dir <out>/runs`` had to reconstruct them from trajectories.
The binding is context-local, so ``--parallel N`` keeps N ledgers apart. And like
-``3dcode make``, the round ``codeverse3d.addons.select`` picks is packaged after the run
+``3dcodeverse make``, the round ``codeverse3d.addons.select`` picks is packaged after the run
(``deliverable/`` + ``selection.json``); a result row is baseline → picked, no pass/fail.
"""
@@ -142,7 +142,7 @@ class BenchOptions(BaseModel):
description="re-run prompts already recorded with one of these statuses "
"(the point of `infra_failed`: retry what the weather lost)")
variant: str = Field(default="", description="the per-model prompt/skill variant every run is created with "
- "(RunOptions.variant, D105); '' = resolve from Settings like `3dcode make`")
+ "(RunOptions.variant, D105); '' = resolve from Settings like `3dcodeverse make`")
REFS_DIR = Path(__file__).resolve().parent / "refs"
@@ -183,7 +183,7 @@ def build_spec(
) -> Spec:
"""Shared Spec core for bench drivers (``run_bench`` / ``compare_backends``).
Drivers resolve their own ``Backends`` and pass their leading tag; ``variant`` is frozen on
- ``RunOptions.variant`` the way ``3dcode make --variant`` freezes it (the track activates it)."""
+ ``RunOptions.variant`` the way ``3dcodeverse make --variant`` freezes it (the track activates it)."""
return Spec(
id=f"{battery.name}/{item.id}", track=battery.track, language=item.language or battery.language, prompt=item.prompt,
constraints=Constraints(must_have=list(item.must_have), dimensions_m=item.dimensions_m),
@@ -197,7 +197,7 @@ def build_spec(
def resolve_variant(name: str, generator_id: str) -> str:
"""The variant a bench run is created with: ``name`` when stated (``--variant``), else what the
settings resolve for the generator (``prompts.variants.resolve_variant`` — the same rule as
- ``3dcode make``), else the base. A tree without the variant layer (pre-D105) resolves to the base
+ ``3dcodeverse make``), else the base. A tree without the variant layer (pre-D105) resolves to the base
unless a name was stated, which it cannot honour."""
if name:
return name
@@ -302,7 +302,7 @@ def default_run_track(spec: Spec, ws: Workspace, resume: bool) -> RunRecord:
def hand_over(ws: Workspace, rec: RunRecord) -> None:
- """What ``3dcode make`` does after a run: package the round ``select`` picks (by score).
+ """What ``3dcodeverse make`` does after a run: package the round ``select`` picks (by score).
A packaging failure costs the hand-over, never the scored row."""
try:
idx = select.pick(ws.root, record=rec)
@@ -435,7 +435,7 @@ def main(argv: Sequence[str] | None = None) -> int:
help="write report.md + report.html when the battery ends")
ap.add_argument("--variant", default="",
help="per-model prompt/skill variant for every run (codeverse3d/prompts/variants/<name>, D105); "
- "default: what the settings resolve for the generator, like `3dcode make`")
+ "default: what the settings resolve for the generator, like `3dcodeverse make`")
ns = ap.parse_args(argv)
if not ns.battery.is_file():
ap.error(f"battery not found: {ns.battery}")
diff --git a/eval/docs/EVAL.md b/eval/docs/EVAL.md
index 9bac3588..41202a57 100644
--- a/eval/docs/EVAL.md
+++ b/eval/docs/EVAL.md
@@ -69,7 +69,7 @@ rung's prompts are generated ONCE and cached (`prompts/rung_R_<lang>.yaml`) with
## 3. Harness battery runs (`python -m bench.run_bench`)
```bash
-cd eval # was `3dcode bench run|report` until 2026-09-22: the launcher lives with the evaluation now
+cd eval # the battery launcher lives with the evaluation
python -m bench.run_bench bench/prompts/static_objects_v1.yaml \
--generator agy:gemini-4-argon --judge gemini:gemini-3.1-pro-preview \
--rounds 2 --parallel 4 --out bench/out/static_v1_apiagent
@@ -77,9 +77,9 @@ python -m bench.run_bench bench/prompts/static_objects_v1.yaml --generator agy:g
python -m bench.report bench/out/static_v1_apiagent
```
Like every bench script it imports THIS tree's `codeverse3d` (`../harness`), so a battery
-launched from a worktree measures the worktree — the `3dcode` console script resolved the
+launched from a worktree measures the worktree — the `3dcodeverse` console script resolves the
editable install instead.
-Each item = one full `3dcode make` run (its own workspace under `--out`) followed by the
+Each item = one full `3dcodeverse make` run (its own workspace under `--out`) followed by the
hand-over `make` does — `addons/select.pick` by score, then `package` (deliverable/ +
selection.json); `results.jsonl` rows carry `score_baseline`, `score_picked` (rows written
before 2026-09-22 say `score_final`), `picked_round`, `rounds`, `cost_usd`, `minutes`,
@@ -226,7 +226,7 @@ Notes: the earlier criteria-first schema compressed flash to 0.6–0.7 (std 0.01
"images beat gate text for visible facts" in the prompt fixed pro hallucinating
"nothing moves" from contract-gate text.
-* Variance on your own runs: `3dcode judge <slug> --n k` and read `score_std`.
+* Variance on your own runs: `3dcodeverse judge <slug> --n k` and read `score_std`.
* Sanity anchors: a skeleton placeholder should score ≈ 0.3–0.5; a deliberately wrong
object should trip the `intent_fidelity` floor / `wrong_object` defect; a floating
part should cap via `connectivity` findings with `data["kind"]="floating"`.
@@ -366,7 +366,7 @@ Notes: the earlier criteria-first schema compressed flash to 0.6–0.7 (std 0.01
tile is not silently dropped. `judge_prompt_hash` moves for every rubric (the rig rules
are shared), the manifest with it.
* **Thinking** — `VlmJudge(thinking=None)` reads `Settings.judge.thinking`
- (`C3D_JUDGE__THINKING`; argon thinks only from `medium`), `3dcode judge --thinking` and
+ (`C3D_JUDGE__THINKING`; argon thinks only from `medium`), `3dcodeverse judge --thinking` and
`addons.calibration --thinking` default to it, the record's sampling snapshot carries it.
* **The A/B evaluator** (`eval/bench/_fixed_eval.py`) runs `placement_gate` on a scene cell
before the judge, so the deterministic caps fire in a compare cell as in the loop.
@@ -822,7 +822,7 @@ python -m bench.loop status --out bench/out/loop_argon
Pearson / Spearman of gate ERRORs against the score, the defect histogram, cost and latency. The
judge is a measuring instrument; this is its calibration certificate per iteration.
* **accept** first refuses an overlay whose `routes.toml` does not type-check or whose bundles fail
- `validate_bundle` (the standards of `3dcode skills validate`), then bumps `variant.toml`'s `version`, snapshots both overlay dirs to
+ `validate_bundle` (the standards of `3dcodeverse skills validate`), then bumps `variant.toml`'s `version`, snapshots both overlay dirs to
`<out>/snapshots/<v>_v<N>/`, and appends a row to `bench/data/loop/<name>/ledger.jsonl` (format:
`bench/data/loop/README.md`); **rollback** restores the last accepted snapshot (an iteration that
was itself the latest accept is un-accepted, version and all); **status** prints the ledger and the
diff --git a/eval/docs/PAPER_WRITING.md b/eval/docs/PAPER_WRITING.md
index 82144185..0d929849 100644
--- a/eval/docs/PAPER_WRITING.md
+++ b/eval/docs/PAPER_WRITING.md
@@ -19,7 +19,7 @@ that owns it — §5.2 was exactly that until 2026-09-03 (review) and is now a p
A/B with an A/A mode), `bench/compare_backends.py` (harness vs one-shot),
`bench/paired_compare.py` (paired statistics), `bench/plan_stage_bench.py` +
`bench/plan_stage_report.py` (the loss-event channel, data in `bench/data/plan_stage/`),
-`3dcode flywheel refine` (the transition corpus) and `finetune/llamafactory/` (its training
+`3dcodeverse flywheel refine` (the transition corpus) and `finetune/llamafactory/` (its training
format).
## 0. The system, in one paragraph
@@ -410,7 +410,7 @@ that moved the bill.
## 7. The corpus the loop produces
-`3dcode flywheel refine` exports every refine round as the transition it was: the previous
+`3dcodeverse flywheel refine` exports every refine round as the transition it was: the previous
round's judge issues and improvement plan, the gate ERROR findings with their fix hints, the
instruction lines `build_refine_instructions` compiled for the round (the task list, not the
per-group prompt text rendered from it), the unified `src/` diff between the two
@@ -511,7 +511,7 @@ data the repo does not carry; they are marked in the caveat column.
| base64 amplification | 261 k tokens vs 516 for one 275 kB sheet | measured on a recorded blob + vendor bundle | arithmetic + matching recorded prompt jumps; **no repo script** |
| token growth in turns | input(t) = 12 364 + 1 822·t; total ∝ n^1.60; dollars ∝ n^1.19–1.34 | 1 063 in-process sessions, 30 401 turns | **no repo script** — the in-process agent, since gemini-cli emits no per-turn usage |
| coupled mechanisms declared | 8 of 14 prompts; driven joints per pose 6 → 3 (1–2 on the one-input mechanisms) | `bench/coupling_stats.py bench/out/wave2_lean --per-prompt` | recomputed from the recorded URDFs |
-| refine corpus | 254 transitions, 112 SFT samples (2026-09-03) | `3dcode flywheel refine`, `finetune/llamafactory/build_refine_sft.py` over `bench/out` | corpus grows with every battery |
+| refine corpus | 254 transitions, 112 SFT samples (2026-09-03) | `3dcodeverse flywheel refine`, `finetune/llamafactory/build_refine_sft.py` over `bench/out` | corpus grows with every battery |
## 9.2 Open questions
@@ -605,7 +605,7 @@ python bench/plan_stage_bench.py --tree ../harness --label restart_on --reps 20
--out bench/data/plan_stage/restart_on.jsonl --env C3D_PLAN_RESTART=1
python bench/plan_stage_report.py bench/data/plan_stage/*.jsonl
# the corpus the loop produces
-3dcode flywheel refine bench/out refine.jsonl --with-code
+3dcodeverse flywheel refine bench/out refine.jsonl --with-code
python ../finetune/llamafactory/build_refine_sft.py refine.jsonl --out refine_sft.jsonl # finetune/ is at the repo root
```
diff --git a/eval/tests/test_bench.py b/eval/tests/test_bench.py
index ea2558ac..bb14970d 100644
--- a/eval/tests/test_bench.py
+++ b/eval/tests/test_bench.py
@@ -58,7 +58,7 @@ def test_run_battery_resume_and_report(tmp_path: Path):
assert len(res) == 4
by_id = {r.id: r for r in res}
assert by_id["furn_easy_stool"].score_picked == 0.85 and by_id["furn_easy_stool"].picked_round == 1
- # like `3dcode make`, the battery hands over the picked round
+ # like `3dcodeverse make`, the battery hands over the picked round
assert (out / "runs" / "furn_easy_stool" / "deliverable" / "manifest.json").is_file()
assert json.loads((out / "runs" / "furn_easy_stool" / "selection.json").read_text())["round"] == 1
assert by_id["veh_easy_toy_car"].status == "error" and "boom" in by_id["veh_easy_toy_car"].errors
@@ -236,7 +236,7 @@ def test_a_redo_starts_from_a_fresh_workspace(tmp_path):
def test_the_launcher_hands_its_flags_to_run_battery(tmp_path: Path, monkeypatch, capsys):
- """``python -m bench.run_bench`` replaced ``3dcode bench run`` (2026-09-22) with the same flags."""
+ """``python -m bench.run_bench`` replaced ``3dcodeverse bench run`` (2026-09-22) with the same flags."""
import bench.run_bench as rb
seen: dict = {}
@@ -324,7 +324,7 @@ def test_a_variant_is_frozen_on_every_spec_and_active_for_the_battery(tmp_path:
assert fake.seen == ["argon/argon"] * 2 and fake.active == "", "frozen on the spec, active for the block, released"
spec = json.loads((tmp_path / "a" / "runs" / "furn_easy_stool" / "spec.json").read_text())
assert spec["options"]["variant"] == "argon"
- # no --variant: the settings resolve it for the generator, as `3dcode make` does — frozen, not activated
+ # no --variant: the settings resolve it for the generator, as `3dcodeverse make` does — frozen, not activated
fake.seen.clear()
run_battery(battery, tmp_path / "b", BenchOptions(parallel=1, limit=1, generator="agy:gemini-4-argon"), run_fn=run)
assert fake.seen == ["argon/"]
@@ -344,7 +344,7 @@ def test_a_variant_is_frozen_on_every_spec_and_active_for_the_battery(tmp_path:
def test_the_report_command_rebuilds_the_report(tmp_path: Path, capsys):
- """``python -m bench.report <out>`` replaced ``3dcode bench report``."""
+ """``python -m bench.report <out>`` replaced ``3dcodeverse bench report``."""
from bench.report import main
battery = REPO / "bench" / "prompts" / "static_objects_v1.yaml"
diff --git a/harness/CLAUDE.md b/harness/CLAUDE.md
index 7f70e283..e8a68353 100644
--- a/harness/CLAUDE.md
+++ b/harness/CLAUDE.md
@@ -1,6 +1,6 @@
# 3dcodeverse — standing context
-Backend harness (python package `codeverse3d`, CLI `3dcodeverse` / short alias `3dcode`,
+Backend harness (python package `codeverse3d`, CLI `3dcodeverse`,
repo path `harness/` under the 3dcodeverse checkout) for LLM-written **raw** 3D code:
Blender bpy · CadQuery · Three.js · URDF · GLSL / OpenGL, across four tracks
(`static_object`, `articulated_object`, `scene`, `graphics`), with pluggable
@@ -82,11 +82,11 @@ push, with `set -o pipefail` so a `| tail` cannot swallow a red exit. The versi
- Live runs under `runs/` (`e2e_*`): read-only reference material; never modify.
- Every model call and vendor session is priced into the run's `telemetry/cost.jsonl` — the ONLY record of
money: `record.total_usage` is its sum at list price (D84). A run's minutes are its steps' time minus
- the time lost to provider errors (`RunRecord.minutes`). Use `3dcode cost <slug>` for actual costs.
+ the time lost to provider errors (`RunRecord.minutes`). Use `3dcodeverse cost <slug>` for actual costs.
## Commands
```
-pip install -e harness # once, from the 3dcodeverse checkout (entry points: 3dcodeverse, 3dcode)
+pip install -e harness # once, from the 3dcodeverse checkout (entry point: 3dcodeverse)
cd harness
3dcodeverse doctor [--live] [--no-gpu] [--json]
3dcodeverse make "a mid-century wooden dining chair" --track static_object --language blender
@@ -101,14 +101,14 @@ cd harness
3dcodeverse make "neon rain on a window" --track graphics --language glsl_shader
3dcodeverse make "..." --image ref.png --candidates 3 --rounds 2 --dim height=0.45 --must "three legs" --no-run
3dcodeverse make "..." --no-pick # a run = baseline + --rounds refine rounds (fixed); make then picks a round → deliverable/ unless --no-pick
-3dcode pick <slug> [--by score|pairwise] [--round N] [--judge MODEL] # hand over a round (codeverse3d/addons/select.py)
-3dcode resume <slug> · 3dcode show <slug> [--section status] (`status` = its STATUS section) · 3dcode render <slug> [--mode wire] · 3dcode judge <slug> [--model ... --n 3 --thinking medium]
-3dcode tools list · 3dcode tools measure --workspace runs/<slug> · 3dcode mcp --workspace runs/<slug>
-3dcode cost <slug> · 3dcode cost --runs-dir ../eval/bench/out/<battery> · 3dcode cost cache <slug>
-3dcode cost prices [--stale] · 3dcode cost estimate gemini:gemini-3.1-pro-preview --in 12000
-3dcode flywheel export runs/ dataset/ [--pack --drop-duplicates --captions-dir d/] · 3dcode flywheel pairs runs/ pairs.jsonl
-3dcode flywheel caption <slug> [--out dir] # the gallery is `3dcode gallery build --embed`
-3dcode gallery serve [ROOTS...] [--port 8765] [--reload] · 3dcode gallery build --out gallery.html [--embed]
+3dcodeverse pick <slug> [--by score|pairwise] [--round N] [--judge MODEL] # hand over a round (codeverse3d/addons/select.py)
+3dcodeverse resume <slug> · 3dcodeverse show <slug> [--section status] (`status` = its STATUS section) · 3dcodeverse render <slug> [--mode wire] · 3dcodeverse judge <slug> [--model ... --n 3 --thinking medium]
+3dcodeverse tools list · 3dcodeverse tools measure --workspace runs/<slug> · 3dcodeverse mcp --workspace runs/<slug>
+3dcodeverse cost <slug> · 3dcodeverse cost --runs-dir ../eval/bench/out/<battery> · 3dcodeverse cost cache <slug>
+3dcodeverse cost prices [--stale] · 3dcodeverse cost estimate gemini:gemini-3.1-pro-preview --in 12000
+3dcodeverse flywheel export runs/ dataset/ [--pack --drop-duplicates --captions-dir d/] · 3dcodeverse flywheel pairs runs/ pairs.jsonl
+3dcodeverse flywheel caption <slug> [--out dir] # the gallery is `3dcodeverse gallery build --embed`
+3dcodeverse gallery serve [ROOTS...] [--port 8765] [--reload] · 3dcodeverse gallery build --out gallery.html [--embed]
(cd ../eval && python -m bench.run_bench bench/prompts/static_objects_v1.yaml --generator ... --judge gemini:gemini-3.1-pro-preview) # report: python -m bench.report bench/out/<battery>
python -m codeverse3d.addons.calibration runs/<slug>... --model gemini:gemini-3.1-pro-preview --n 3 --out out/
(cd ../eval && python -m bench.loop init --name argon_scene --out bench/out/loop_argon --variant argon --languages scene_threejs,scene_blender --judge gemini:gemini-4-argon --judge-thinking medium --parallel 15 && python -m bench.loop generate --iter 0 && python -m bench.loop run --iter 0 && python -m bench.loop diagnose --iter 0) # the argon loop (eval/docs/EVAL.md §10): then compare --iter N --against M · judge-calib --iter N · accept|rollback --iter N · status
diff --git a/harness/README.md b/harness/README.md
index b5ad0e40..9d809157 100644
--- a/harness/README.md
+++ b/harness/README.md
@@ -20,10 +20,10 @@ renders, judges, refines and records every run as data-flywheel material.
* **Flywheel:** git-versioned `src/` per round, `record.json`, sample export with
quality tiers + dedupe (+ parquet, tar locators), preference/repair/cross-backend
pairs, captions, HTML gallery.
-* **Looking at results:** `3dcode gallery serve` — a local page over `runs/` +
+* **Looking at results:** `3dcodeverse gallery serve` — a local page over `runs/` +
`eval/bench/out/*/runs` with filters, a per-filter summary strip and a detail page per run;
it serves the run directories too, so every link (sheet, renders, `src/`, `object.glb`
- in an orbit viewer, `record.json`) actually opens. `3dcode gallery build --embed` writes
+ in an orbit viewer, `record.json`) actually opens. `3dcodeverse gallery build --embed` writes
the same page as one shareable file.
## Architecture
@@ -55,8 +55,8 @@ codeverse3d/
record/ what every run writes: record.json, deliverable/, telemetry
addons/ optional tools that READ finished runs — nothing a run needs:
gallery (the runs browser), dataset (export, tiers, preference/repair pairs,
- captions), costreport (`3dcode cost`), calibration, skill_targets
- cli/ the typer CLI · doctor.py the environment checks behind `3dcode doctor`
+ captions), costreport (`3dcodeverse cost`), calibration, skill_targets
+ cli/ the typer CLI · doctor.py the environment checks behind `3dcodeverse doctor`
../eval/bench/ the evaluation harness around the harness: run_bench, compare_backends
(A/B matrix), ab_plan, infra-failure classification
runtime_js/ node side: [email protected] + headless-Chrome render/probe hosts
@@ -71,22 +71,22 @@ Design laws, the full package map and what each stage does: `docs/ARCHITECTURE.m
```bash
bash setup.sh # python package + runtime_js + headless Chrome + the agy / claude / codex CLIs at their latest + doctor
# ...or by hand:
-pip install -e '.[all,dev]' # entry points: 3dcodeverse, 3dcode
+pip install -e '.[all,dev]' # entry point: 3dcodeverse
(cd runtime_js && npm ci) # [email protected] + puppeteer 24 (chrome → ~/.cache/puppeteer)
curl -fsSL https://antigravity.google/cli/install.sh | bash # agy, the default generator's CLI (`agy update` later)
npm i -g @anthropic-ai/claude-code@latest @openai/codex@latest # optional generators claude-code / codex (local subscriptions)
export GEMINI_API_KEYS="key1,key2" # agy runs in Gemini-API mode on this key pool (docs/INSTALL.md §8)
-3dcode doctor
-3dcode make "a mid-century wooden dining chair" --track static_object --language blender
-3dcode make "a kitchen cabinet with one door and a drawer" --track articulated_object --language urdf_blender
-3dcode make "a small japanese garden at dusk with a koi pond" --track scene --language scene_threejs
-3dcode make "neon cyberpunk rain on a window" --track graphics --language glsl_shader
-3dcode gallery serve # browse every run at http://127.0.0.1:8765/ (docs/RUNBOOK.md §4)
+3dcodeverse doctor
+3dcodeverse make "a mid-century wooden dining chair" --track static_object --language blender
+3dcodeverse make "a kitchen cabinet with one door and a drawer" --track articulated_object --language urdf_blender
+3dcodeverse make "a small japanese garden at dusk with a koi pond" --track scene --language scene_threejs
+3dcodeverse make "neon cyberpunk rain on a window" --track graphics --language glsl_shader
+3dcodeverse gallery serve # browse every run at http://127.0.0.1:8765/ (docs/RUNBOOK.md §4)
python -m pytest tests -q -m "not live"
```
Expected cost and time per track: `docs/RUNBOOK.md` §2. Inspect actual run costs with
-`3dcode cost <slug>` and a call's expected price with `3dcode cost estimate`.
+`3dcodeverse cost <slug>` and a call's expected price with `3dcodeverse cost estimate`.
Docs: `docs/INSTALL.md` (install / prerequisites / doctor troubleshooting) ·
`docs/ARCHITECTURE.md` (design + what a run does) · `docs/RUNBOOK.md` (operate / extend) ·
diff --git a/harness/codeverse3d/__init__.py b/harness/codeverse3d/__init__.py
index 2316d346..7956bc95 100644
--- a/harness/codeverse3d/__init__.py
+++ b/harness/codeverse3d/__init__.py
@@ -11,7 +11,7 @@ Package map:
- ``spatial`` 3D tools on GLB / scene, also an MCP server.
- ``judges`` VLM judges + rubrics. - ``tracks`` the four pipelines.
- ``orchestrator`` stage runner, round loop. - ``record`` record.json, deliverable/, telemetry.
-- ``addons`` tools that read finished runs. - ``cli`` the ``3dcode`` command line.
+- ``addons`` tools that read finished runs. - ``cli`` the ``3dcodeverse`` command line.
"""
__version__ = "0.1.0"
diff --git a/harness/codeverse3d/addons/costreport/__init__.py b/harness/codeverse3d/addons/costreport/__init__.py
index 92f4fa50..cdaafda2 100644
--- a/harness/codeverse3d/addons/costreport/__init__.py
+++ b/harness/codeverse3d/addons/costreport/__init__.py
@@ -1,4 +1,4 @@
-"""The cost of finished runs, read back: ``3dcode cost`` and its markdown report.
+"""The cost of finished runs, read back: ``3dcodeverse cost`` and its markdown report.
* ``audit`` ``audit_runs(paths)`` → ``Audit`` (every run's ledger under the paths, aggregated, + waste)
* ``report`` ``console(audit)`` · ``markdown(audit)``
diff --git a/harness/codeverse3d/addons/costreport/report.py b/harness/codeverse3d/addons/costreport/report.py
index 2e0b9821..d3f37903 100644
--- a/harness/codeverse3d/addons/costreport/report.py
+++ b/harness/codeverse3d/addons/costreport/report.py
@@ -145,7 +145,7 @@ def markdown(audit: Audit, *, title: str = "Cost audit") -> str:
def console(audit: Audit) -> str:
- """Compact plain-text version for ``3dcode cost``."""
+ """Compact plain-text version for ``3dcodeverse cost``."""
lines = [line.replace("**", "") for line in summary_lines(audit)]
lines += ["", "stage:"]
for b in audit.summary.ranked("stage"):
diff --git a/harness/codeverse3d/addons/dataset/__init__.py b/harness/codeverse3d/addons/dataset/__init__.py
index 35bbfb8f..b514d82b 100644
--- a/harness/codeverse3d/addons/dataset/__init__.py
+++ b/harness/codeverse3d/addons/dataset/__init__.py
@@ -1,4 +1,4 @@
-"""Run records → a dataset: what ``3dcode flywheel …`` does AFTER the runs exist.
+"""Run records → a dataset: what ``3dcodeverse flywheel …`` does AFTER the runs exist.
* ``export`` ``export_samples(runs_dir, out_dir, min_score=..)`` → ``ExportReport``
* ``pack`` ``pack_samples(out_dir)`` (plain tars + byte-range locators, optional)
diff --git a/harness/codeverse3d/addons/dataset/export.py b/harness/codeverse3d/addons/dataset/export.py
index 9efd4143..d2315411 100644
--- a/harness/codeverse3d/addons/dataset/export.py
+++ b/harness/codeverse3d/addons/dataset/export.py
@@ -370,7 +370,7 @@ def load_manifest(out_dir: Path | str) -> DatasetGenerationManifest:
path = Path(out_dir) / MANIFEST_NAME
if not path.is_file():
raise ManifestError(
- f"{path} not found — run export first (3dcode flywheel export writes it): "
+ f"{path} not found — run export first (3dcodeverse flywheel export writes it): "
f"pack no longer rescans directories")
try:
return DatasetGenerationManifest.model_validate_json(path.read_text())
diff --git a/harness/codeverse3d/addons/gallery/__init__.py b/harness/codeverse3d/addons/gallery/__init__.py
index 5a8d1ce8..241576a0 100644
--- a/harness/codeverse3d/addons/gallery/__init__.py
+++ b/harness/codeverse3d/addons/gallery/__init__.py
@@ -1,4 +1,4 @@
-"""The local run gallery: ``3dcode gallery serve`` (localhost) and ``3dcode gallery build`` (one file).
+"""The local run gallery: ``3dcodeverse gallery serve`` (localhost) and ``3dcodeverse gallery build`` (one file).
* ``index`` ``build_index(roots)`` · ``default_roots()`` — record.json → typed entries
* ``model`` ``RunEntry`` / ``RootSection`` / ``GalleryIndex`` / ``summarize`` / ``match``,
diff --git a/harness/codeverse3d/addons/gallery/code.py b/harness/codeverse3d/addons/gallery/code.py
index 52771d2b..2332d08f 100644
--- a/harness/codeverse3d/addons/gallery/code.py
+++ b/harness/codeverse3d/addons/gallery/code.py
@@ -118,7 +118,7 @@ def render_dir(entry: RunEntry, urls: UrlMaker, rel: str) -> str:
rows.append(f"<li><a href='{esc(href)}'>{esc(child.name)}</a>"
f"<a class='sz' href='{esc(urls.file(entry, child_rel))}'>{size:,} B · raw</a></li>")
listing = "".join(rows) or "<li class='faint'>empty</li>"
- body = (top_bar("3dcode gallery", entry.slug, crumbs=_crumbs(entry, urls, rel))
+ body = (top_bar("3dcodeverse gallery", entry.slug, crumbs=_crumbs(entry, urls, rel))
+ f"<main class='wrap'><div class='panel'><h2>{esc(rel or '.')}</h2>"
f"<ul class='filelist'>{listing}</ul></div></main>"
+ footer(entry.path))
@@ -133,7 +133,7 @@ def render_file(entry: RunEntry, urls: UrlMaker, rel: str) -> str:
if not target.is_file():
raise FileNotFoundError(f"not a file: {rel}")
if not is_text(target):
- body = (top_bar("3dcode gallery", entry.slug, crumbs=_crumbs(entry, urls, rel))
+ body = (top_bar("3dcodeverse gallery", entry.slug, crumbs=_crumbs(entry, urls, rel))
+ f"<main class='wrap'><div class='panel'><h2>{esc(rel)}</h2>"
f"<p class='muted small'>binary file — "
f"<a href='{esc(urls.file(entry, rel))}'>open the raw bytes</a></p></div></main>"
@@ -143,7 +143,7 @@ def render_file(entry: RunEntry, urls: UrlMaker, rel: str) -> str:
head = (f"<h2>{esc(rel)}</h2><p class='small muted'>{len(text.splitlines())} lines · "
f"<a href='{esc(urls.file(entry, rel))}'>raw</a>"
+ (f" · <span class='pill-warn'>{esc(note)}</span>" if note else "") + "</p>")
- body = (top_bar("3dcode gallery", entry.slug, crumbs=_crumbs(entry, urls, rel))
+ body = (top_bar("3dcodeverse gallery", entry.slug, crumbs=_crumbs(entry, urls, rel))
+ f"<main class='wrap'><div class='panel'>{head}{numbered(text, suffix=target.suffix)}</div></main>"
+ footer(str(target)))
return page_shell(f"{rel} — {entry.slug}", body, extra_css=CODE_CSS)
diff --git a/harness/codeverse3d/addons/gallery/compare.py b/harness/codeverse3d/addons/gallery/compare.py
index e16c844e..758e9cff 100644
--- a/harness/codeverse3d/addons/gallery/compare.py
+++ b/harness/codeverse3d/addons/gallery/compare.py
@@ -80,9 +80,9 @@ def render_compare(entries: list[RunEntry], urls: UrlMaker, *, note: str = "") -
if not entries:
inner = ("<div class='panel'><h2>nothing to compare</h2><p class='muted'>Tick two or more "
"runs on the index and press <b>compare</b>.</p></div>")
- body = (top_bar("3dcode gallery", crumbs="<a href='/'>gallery</a> <span class='faint'>/</span> "
- "<b>compare</b>") + f"<main class='wrap'>{inner}</main>" + footer("3dcode gallery"))
- return page_shell("compare — 3dcode gallery", body, extra_css=COMPARE_CSS)
+ body = (top_bar("3dcodeverse gallery", crumbs="<a href='/'>gallery</a> <span class='faint'>/</span> "
+ "<b>compare</b>") + f"<main class='wrap'>{inner}</main>" + footer("3dcodeverse gallery"))
+ return page_shell("compare — 3dcodeverse gallery", body, extra_css=COMPARE_CSS)
def shot(e: RunEntry) -> str:
image = e.card_image
@@ -125,7 +125,7 @@ def render_compare(entries: list[RunEntry], urls: UrlMaker, *, note: str = "") -
counts = {k: sum(1 for e in entries if e.verdict == k) for k, _ in VERDICT_META.items()}
tally = " · ".join(f"{v} {k}" for k, v in counts.items() if v)
body = (
- top_bar("3dcode gallery", f"{len(entries)} runs · {tally}",
+ top_bar("3dcodeverse gallery", f"{len(entries)} runs · {tally}",
crumbs="<a href='/'>gallery</a> <span class='faint'>/</span> <b>compare</b>",
right=f"<a class='btn' href='/export.csv?runs={quote(','.join(e.key for e in entries))}'>"
f"export csv</a>")
@@ -136,8 +136,8 @@ def render_compare(entries: list[RunEntry], urls: UrlMaker, *, note: str = "") -
+ f"<div class='cmpwrap cmp'><table class='cmp'><tbody>{rows}</tbody></table></div>"
+ "<p class='small faint' style='margin-top:var(--s-3)'>read-only view — the gallery never "
"starts, resumes or deletes a run.</p>"
- + "</main>" + footer("compare · 3dcode gallery"))
- return page_shell("compare — 3dcode gallery", body, extra_css=COMPARE_CSS)
+ + "</main>" + footer("compare · 3dcodeverse gallery"))
+ return page_shell("compare — 3dcodeverse gallery", body, extra_css=COMPARE_CSS)
#: a leading one of these makes a spreadsheet treat the cell as a formula (CSV injection)
diff --git a/harness/codeverse3d/addons/gallery/detail.py b/harness/codeverse3d/addons/gallery/detail.py
index 5681a5ba..821dc3be 100644
--- a/harness/codeverse3d/addons/gallery/detail.py
+++ b/harness/codeverse3d/addons/gallery/detail.py
@@ -338,7 +338,7 @@ def render_detail(entry: RunEntry, urls: UrlMaker, ws: Workspace, rec: RunRecord
nav = "".join(f"<a href='#{a}'>{a}</a>" for a in
("rounds", "judge", "complexity", "measurement", "renders", "cost", "code"))
body = (
- top_bar("3dcode gallery",
+ top_bar("3dcodeverse gallery",
crumbs=f"<a href='/'>gallery</a> <span class='faint'>/</span> "
f"<a href='/?battery={quote(entry.battery)}'>{esc(entry.battery)}</a> "
f"<span class='faint'>/</span> <b>{esc(entry.slug)}</b>",
@@ -357,17 +357,17 @@ def render_detail(entry: RunEntry, urls: UrlMaker, ws: Workspace, rec: RunRecord
+ "</main>"
+ footer(f"{entry.path} · record.json re-read on every request")
)
- return page_shell(f"{entry.slug} — 3dcode gallery", body, extra_css=DETAIL_CSS,
+ return page_shell(f"{entry.slug} — 3dcodeverse gallery", body, extra_css=DETAIL_CSS,
scripts=NEIGHBOUR_JS)
def render_broken_detail(entry: RunEntry, urls: UrlMaker) -> str:
links = " · ".join(f"<a href='{esc(urls.link(entry, link))}'>{esc(link.label)}</a>" for link in entry.links)
- body = (top_bar("3dcode gallery", entry.battery,
+ body = (top_bar("3dcodeverse gallery", entry.battery,
crumbs=f"<a href='/'>gallery</a> <span class='faint'>/</span> <b>{esc(entry.slug)}</b>")
+ "<main class='wrap'>"
+ _panel(entry.slug, f"<div class='err'>{esc(entry.state)}: {esc(entry.error)}</div>"
f"<p style='margin-top:var(--s-3)'>{esc(entry.prompt)}</p>"
f"<div class='linkrow'>{links}</div>")
+ "</main>" + footer(entry.path))
- return page_shell(f"{entry.slug} — 3dcode gallery", body, extra_css=DETAIL_CSS)
+ return page_shell(f"{entry.slug} — 3dcodeverse gallery", body, extra_css=DETAIL_CSS)
diff --git a/harness/codeverse3d/addons/gallery/page.py b/harness/codeverse3d/addons/gallery/page.py
index a047f9cf..5495f9fd 100644
--- a/harness/codeverse3d/addons/gallery/page.py
+++ b/harness/codeverse3d/addons/gallery/page.py
@@ -120,7 +120,7 @@ def _section(label: str, path: str, entries: list[RunEntry], hidden: set[str], u
)
-def render_index(index: GalleryIndex, urls: UrlMaker, *, title: str = "3dcode gallery",
+def render_index(index: GalleryIndex, urls: UrlMaker, *, title: str = "3dcodeverse gallery",
flt: dict[str, str] | None = None, sort: str = "score", view: str = "cards",
note: str = "", extra_html: str = "") -> str:
flt = {k: (flt or {}).get(k, "") for k in FILTER_KEYS}
@@ -138,7 +138,7 @@ def render_index(index: GalleryIndex, urls: UrlMaker, *, title: str = "3dcode ga
right = (f"<button class='btn' id='view-btn' type='button' aria-pressed='{'true' if view == 'table' else 'false'}'>"
f"{'▤ table' if view == 'table' else '▦ cards'}</button>")
body = (
- top_bar("3dcode gallery", f"{len(all_entries)} runs · {len(index.sections)} roots", right=right)
+ top_bar("3dcodeverse gallery", f"{len(all_entries)} runs · {len(index.sections)} roots", right=right)
+ "<main class='wrap'>"
+ (f"<p class='small muted'>{esc(note)}</p>" if note else "")
+ _summary_strip(selected, len(all_entries))
@@ -155,12 +155,12 @@ def render_index(index: GalleryIndex, urls: UrlMaker, *, title: str = "3dcode ga
# ===================================================================== static_site
-def render_static(index: GalleryIndex, *, title: str = "3dcode gallery", embed: bool = False,
+def render_static(index: GalleryIndex, *, title: str = "3dcodeverse gallery", embed: bool = False,
thumb_px: int = THUMB_PX, sort: str = "score", view: str = "cards",
extra_html: str = "") -> str:
note = ("images are inlined; the links open the original run directories on this machine"
if embed else "images and links point at the run directories with file:// — "
- "use `3dcode gallery serve` for a page that works anywhere")
+ "use `3dcodeverse gallery serve` for a page that works anywhere")
return render_index(index, StaticUrls(embed=embed, thumb_px=thumb_px),
title=title, sort=sort, view=view, note=note, extra_html=extra_html)
@@ -168,7 +168,7 @@ def render_static(index: GalleryIndex, *, title: str = "3dcode gallery", embed:
def build_static(roots: list[Path] | list[str], out_html: Path | str, *, title: str | None = None,
embed: bool = False, thumb_px: int = THUMB_PX) -> tuple[Path, int, GalleryIndex]:
index = build_index(roots)
- label = title or ("3dcode gallery — " + ", ".join(s.label for s in index.sections[:4])
+ label = title or ("3dcodeverse gallery — " + ", ".join(s.label for s in index.sections[:4])
+ ("…" if len(index.sections) > 4 else ""))
html = render_static(index, title=label, embed=embed, thumb_px=thumb_px)
return write_text_atomic(Path(out_html), html), len(index.entries()), index
diff --git a/harness/codeverse3d/addons/gallery/server.py b/harness/codeverse3d/addons/gallery/server.py
index dcf1be9e..62f2e2e9 100644
--- a/harness/codeverse3d/addons/gallery/server.py
+++ b/harness/codeverse3d/addons/gallery/server.py
@@ -1,4 +1,4 @@
-"""The local gallery server: ``3dcode gallery serve``."""
+"""The local gallery server: ``3dcodeverse gallery serve``."""
from __future__ import annotations
@@ -52,16 +52,16 @@ class Response:
def _error_page(status: int, title: str, detail: str) -> Response:
- body = (top_bar("3dcode gallery", "", crumbs="<a href='/'>gallery</a>")
+ body = (top_bar("3dcodeverse gallery", "", crumbs="<a href='/'>gallery</a>")
+ f"<main class='wrap'><div class='panel'><h2>{status} — {esc(title)}</h2>"
- f"<p class='muted'>{esc(detail)}</p></div></main>" + footer("3dcode gallery"))
+ f"<p class='muted'>{esc(detail)}</p></div></main>" + footer("3dcodeverse gallery"))
return Response.html(page_shell(f"{status} {title}", body), status=status)
class GalleryApp:
def __init__(self, roots: list[Path] | list[str], *, reload: bool = False,
- title: str = "3dcode gallery", runtime_js: Path | str | None = None) -> None:
+ title: str = "3dcodeverse gallery", runtime_js: Path | str | None = None) -> None:
self.roots = [Path(r).resolve() for r in roots]
self.reload = reload
self.title = title
@@ -290,7 +290,7 @@ def make_server(app: GalleryApp, host: str = DEFAULT_HOST, port: int = DEFAULT_P
def serve(roots: list[Path] | list[str], *, host: str | None = None,
port: int = DEFAULT_PORT, reload: bool = False, open_browser: bool = False,
- title: str = "3dcode gallery", on_start: Any = None) -> None:
+ title: str = "3dcodeverse gallery", on_start: Any = None) -> None:
bind = host or DEFAULT_HOST
app = GalleryApp(roots, reload=reload, title=title)
httpd = make_server(app, bind, port)
diff --git a/harness/codeverse3d/addons/gallery/viewer.py b/harness/codeverse3d/addons/gallery/viewer.py
index c524f9b4..6b90bde9 100644
--- a/harness/codeverse3d/addons/gallery/viewer.py
+++ b/harness/codeverse3d/addons/gallery/viewer.py
@@ -122,7 +122,7 @@ def render_viewer(entry: RunEntry, urls: UrlMaker, rel: str) -> str:
"three/addons/": "/vendor/three/examples/jsm/"}})
src = urls.file(entry, rel)
body = (
- top_bar("3dcode gallery", entry.slug,
+ top_bar("3dcodeverse gallery", entry.slug,
crumbs=f"<a href='/'>gallery</a> <span class='faint'>/</span> "
f"<a href='{esc(urls.detail(entry))}'>{esc(entry.slug)}</a> "
f"<span class='faint'>/</span> <b>{esc(rel)}</b>",
diff --git a/harness/codeverse3d/addons/select.py b/harness/codeverse3d/addons/select.py
index 0a201ac7..897db1bc 100644
--- a/harness/codeverse3d/addons/select.py
+++ b/harness/codeverse3d/addons/select.py
@@ -1,4 +1,4 @@
-"""Which round of a finished run to hand over — and the hand-over itself (``3dcode pick``)."""
+"""Which round of a finished run to hand over — and the hand-over itself (``3dcodeverse pick``)."""
from __future__ import annotations
diff --git a/harness/codeverse3d/cli/__init__.py b/harness/codeverse3d/cli/__init__.py
index 71a17f23..38184625 100644
--- a/harness/codeverse3d/cli/__init__.py
+++ b/harness/codeverse3d/cli/__init__.py
@@ -1 +1 @@
-"""``3dcode`` command line."""
+"""``3dcodeverse`` command line."""
diff --git a/harness/codeverse3d/cli/_common.py b/harness/codeverse3d/cli/_common.py
index a1494332..79719446 100644
--- a/harness/codeverse3d/cli/_common.py
+++ b/harness/codeverse3d/cli/_common.py
@@ -119,7 +119,7 @@ def print_observation(obs: Any, *, as_json: bool) -> None:
def doctor_table(rows: list[tuple[str, str, str]]) -> Table:
- t = Table(title="3dcode doctor")
+ t = Table(title="3dcodeverse doctor")
t.add_column("check", style="bold")
t.add_column("status")
t.add_column("detail")
diff --git a/harness/codeverse3d/cli/_judge.py b/harness/codeverse3d/cli/_judge.py
index f4ce34ea..0c6e27dc 100644
--- a/harness/codeverse3d/cli/_judge.py
+++ b/harness/codeverse3d/cli/_judge.py
@@ -1,4 +1,4 @@
-"""``3dcode judge`` implementation: re-judge a stored round with the inputs the in-run judge
+"""``3dcodeverse judge`` implementation: re-judge a stored round with the inputs the in-run judge
saw, built by ``judges.base.round_input`` from the run directory."""
from __future__ import annotations
diff --git a/harness/codeverse3d/cli/cost_cmd.py b/harness/codeverse3d/cli/cost_cmd.py
index 45c144aa..bd5b8432 100644
--- a/harness/codeverse3d/cli/cost_cmd.py
+++ b/harness/codeverse3d/cli/cost_cmd.py
@@ -1,13 +1,13 @@
-"""``3dcode cost`` — what the runs cost and where the money went, from each run's
+"""``3dcodeverse cost`` — what the runs cost and where the money went, from each run's
``telemetry/cost.jsonl`` ledger.
- 3dcode cost <slug> # one run, from its own ledger
- 3dcode cost --runs-dir eval/bench/out/x # a whole battery, aggregated
- 3dcode cost runs/ eval/bench/out/* # any mix of run dirs / trees
- 3dcode cost show runs/ --md report.md # the same, explicit + full markdown
- 3dcode cost runs/ --recheck # re-price with today's table (drift vs what was billed)
- 3dcode cost prices [--stale] [--unverified]
- 3dcode cost estimate gemini:… --in 12000 --out 800
+ 3dcodeverse cost <slug> # one run, from its own ledger
+ 3dcodeverse cost --runs-dir eval/bench/out/x # a whole battery, aggregated
+ 3dcodeverse cost runs/ eval/bench/out/* # any mix of run dirs / trees
+ 3dcodeverse cost show runs/ --md report.md # the same, explicit + full markdown
+ 3dcodeverse cost runs/ --recheck # re-price with today's table (drift vs what was billed)
+ 3dcodeverse cost prices [--stale] [--unverified]
+ 3dcodeverse cost estimate gemini:… --in 12000 --out 800
"""
from __future__ import annotations
diff --git a/harness/codeverse3d/cli/flywheel_cmd.py b/harness/codeverse3d/cli/flywheel_cmd.py
index 95b4217c..cbea6aa2 100644
--- a/harness/codeverse3d/cli/flywheel_cmd.py
+++ b/harness/codeverse3d/cli/flywheel_cmd.py
@@ -1,4 +1,4 @@
-"""``3dcode flywheel export | pairs | caption | index``."""
+"""``3dcodeverse flywheel export | pairs | caption | index``."""
from __future__ import annotations
@@ -118,7 +118,7 @@ def caption_cmd(
continue
# per run: captioning rewrites record.json (--out writes a side-car instead); a locked
# run is skipped so one live run does not abort an --all batch
- lock = nullcontext() if out else exclusive(ws.root, what=f"3dcode flywheel caption {ws.root.name}")
+ lock = nullcontext() if out else exclusive(ws.root, what=f"3dcodeverse flywheel caption {ws.root.name}")
try:
with lock, run_ledger(ws.root, run=ws.root.name, create=False):
caps = caption_sample(ws, rec, model, out_dir=out, slug=slug)
diff --git a/harness/codeverse3d/cli/inspect_cmd.py b/harness/codeverse3d/cli/inspect_cmd.py
index 3a03999f..5569ea38 100644
--- a/harness/codeverse3d/cli/inspect_cmd.py
+++ b/harness/codeverse3d/cli/inspect_cmd.py
@@ -1,4 +1,4 @@
-"""``3dcode render`` / ``3dcode judge`` — act on one existing run: re-render its working
+"""``3dcodeverse render`` / ``3dcodeverse judge`` — act on one existing run: re-render its working
tree, or re-judge a round into ``artifacts/judge/rNN_cli.json``."""
from __future__ import annotations
@@ -47,7 +47,7 @@ def render(
idx = max(ws.rendered_rounds(), default=0)
out_dir = out or ws.renders_dir(idx) / ("cli" if mode == "shaded" else f"cli_{mode}")
# a graphics render rebuilds the run: one writer per run dir
- with C.mutating(ws, what=f"3dcode render {ws.root.name}", action="render"):
+ with C.mutating(ws, what=f"3dcodeverse render {ws.root.name}", action="render"):
if spec.track is Track.GRAPHICS:
_render_graphics(ws, spec, out_dir)
return
@@ -126,7 +126,7 @@ def judge(
raise C.CliError(str(e), code=2) from e
n_images = J.count_prompt_images(inp, rubric_name, judge_obj)
# writes into the run: one writer per run dir; judged under the run's variant
- with C.mutating(ws, what=f"3dcode judge {ws.root.name}", action="judge"), C.replay_variant(rec):
+ with C.mutating(ws, what=f"3dcodeverse judge {ws.root.name}", action="judge"), C.replay_variant(rec):
try:
with join_post_run(ws):
verdict = judge_obj.judge(inp)
diff --git a/harness/codeverse3d/cli/layout_cmd.py b/harness/codeverse3d/cli/layout_cmd.py
index 5be815e3..3e5b1394 100644
--- a/harness/codeverse3d/cli/layout_cmd.py
+++ b/harness/codeverse3d/cli/layout_cmd.py
@@ -1,5 +1,5 @@
-"""``3dcode show`` — the single-run view: STATUS (works mid-run, without record.json;
-``3dcode status`` is ``show --section status``), DELIVERABLE, QUALITY EVIDENCE and
+"""``3dcodeverse show`` — the single-run view: STATUS (works mid-run, without record.json;
+``3dcodeverse status`` is ``show --section status``), DELIVERABLE, QUALITY EVIDENCE and
COST & SETTINGS."""
from __future__ import annotations
@@ -57,7 +57,7 @@ def print_status(ws: Workspace, record: RunRecord | None, n_events: int) -> None
"generator": spec.backends.generator, "judge": spec.backends.judge, "workspace": ws.root}))
state: dict[str, Any] = {}
if (held := holder_of(ws.root)) is not None:
- state["RUNNING NOW"] = f"pid {held.get('pid', '?')} ({held.get('what') or '3dcode'})"
+ state["RUNNING NOW"] = f"pid {held.get('pid', '?')} ({held.get('what') or '3dcodeverse'})"
if ws.state_path.is_file():
raw = read_json_or_none(ws.state_path)
state["run_state"] = "(unreadable)" if raw is None else ", ".join(
@@ -137,7 +137,7 @@ def print_deliverable(ws: Workspace, record: RunRecord) -> None:
if d is not None:
console.print(_deliverable_table(d))
else:
- console.print(kv_table("artifacts (no deliverable/ yet — `3dcode pick` packages a round)",
+ console.print(kv_table("artifacts (no deliverable/ yet — `3dcodeverse pick` packages a round)",
_unpackaged_artifact_rows(ws)))
@@ -263,7 +263,7 @@ def show(
record: RunRecord | None = load_record(ws)
except RecordError as e:
if section not in ("all", "status"):
- raise C.CliError(f"{e} (`3dcode status {slug}` shows where the run stands)") from e
+ raise C.CliError(f"{e} (`3dcodeverse status {slug}` shows where the run stands)") from e
record, missing = None, str(e)
if section in ("all", "status"):
print_status(ws, record, events)
@@ -280,5 +280,5 @@ def show(
def status(slug: str, runs_dir: RunsDirOpt = None, events: EventsOpt = 8) -> None:
- """Where a run stands, mid-run too: `3dcode show <slug> --section status`."""
+ """Where a run stands, mid-run too: `3dcodeverse show <slug> --section status`."""
show(slug, runs_dir, section="status", events=events)
diff --git a/harness/codeverse3d/cli/main.py b/harness/codeverse3d/cli/main.py
index a186d541..11cfaf21 100644
--- a/harness/codeverse3d/cli/main.py
+++ b/harness/codeverse3d/cli/main.py
@@ -1,4 +1,4 @@
-"""``3dcode`` — the command line: the typer app, every registration, and ``make`` /
+"""``3dcodeverse`` — the command line: the typer app, every registration, and ``make`` /
``resume`` / ``pick`` / ``mcp``. Commands import the harness lazily so ``--help`` stays fast."""
from __future__ import annotations
@@ -120,7 +120,7 @@ def resolve_roots(roots: list[Path] | None) -> list[Path]:
return [fallback]
raise C.CliError(f"no run roots found (looked for runs/ and eval/bench/out/*/runs under {Path.cwd()}, "
f"and {C.EVAL_ROOT / 'bench' / 'out'}/*/runs); "
- f"pass one explicitly: `3dcode gallery serve path/to/runs`")
+ f"pass one explicitly: `3dcodeverse gallery serve path/to/runs`")
return found
@@ -133,7 +133,7 @@ def serve_cmd(
open_browser: Annotated[bool, typer.Option("--open/--no-open", help="open the page in a browser")] = True,
reload: Annotated[bool, typer.Option("--reload", help="re-scan the roots on every page load "
"(cheap: records only, images stay lazy)")] = False,
- title: Annotated[str, typer.Option("--title")] = "3dcode gallery",
+ title: Annotated[str, typer.Option("--title")] = "3dcodeverse gallery",
) -> None:
"""Serve the gallery (and the run directories) on localhost."""
from codeverse3d.addons.gallery.server import GalleryError, serve
@@ -174,11 +174,11 @@ def build_cmd(
warn(f"{broken} run(s) have no usable record.json (shown as broken cards)")
ok(f"gallery of {n} runs → {path} ({path.stat().st_size // 1024} KB)")
if not embed:
- console.print("[dim]file:// links; `--embed` inlines the images, `3dcode gallery serve` makes them clickable[/dim]")
+ console.print("[dim]file:// links; `--embed` inlines the images, `3dcodeverse gallery serve` makes them clickable[/dim]")
app = typer.Typer(
- name="3dcode",
+ name="3dcodeverse",
help="3dcodeverse: LLMs write raw 3D code; the harness builds, judges, refines, records.",
pretty_exceptions_enable=False,
)
@@ -197,7 +197,7 @@ app.add_typer(
app.command("doctor", help="Environment checks.")(doctor)
app.command(
"tools",
- help="List spatial tools or run one: `3dcode tools list` | `3dcode tools <name> --json '{...}' --workspace ws`.",
+ help="List spatial tools or run one: `3dcodeverse tools list` | `3dcodeverse tools <name> --json '{...}' --workspace ws`.",
)(tools)
@app.callback(invoke_without_command=True)
@@ -210,7 +210,7 @@ def _root(
with contextlib.suppress(Exception):
faulthandler.register(signal.SIGUSR1, file=sys.__stderr__, all_threads=True)
if version:
- console.print(f"3dcode {__version__}")
+ console.print(f"3dcodeverse {__version__}")
raise typer.Exit()
if ctx.invoked_subcommand is None:
console.print(ctx.get_help())
@@ -284,7 +284,7 @@ def make(
typer.Option(
"--no-pick",
help="stop after the rounds: no round is picked, no deliverable/ is written "
- "(`3dcode pick <slug>` does it later)",
+ "(`3dcodeverse pick <slug>` does it later)",
),
] = False,
seed: Annotated[
@@ -355,7 +355,7 @@ def make(
except ValueError as e:
raise C.CliError(f"invalid run spec: {e}") from e
# the run mutex: taken before the workspace is created (or --force wipes one), held through the run
- with C.mutating(C.runs_root(runs_dir) / run_slug, what=f"3dcode make {run_slug}", action="create"):
+ with C.mutating(C.runs_root(runs_dir) / run_slug, what=f"3dcodeverse make {run_slug}", action="create"):
ws = C.create_workspace(C.runs_root(runs_dir) / run_slug, force=force)
ws.write_json(ws.spec_path, spec)
ws.commit("spec")
@@ -377,7 +377,7 @@ def make(
)
)
if no_run:
- ok(f"spec written: {ws.spec_path} (not run; `3dcode resume {run_slug}` to start)")
+ ok(f"spec written: {ws.spec_path} (not run; `3dcodeverse resume {run_slug}` to start)")
return
_run_track(spec, ws, resume=False, candidates=candidates, pick=not no_pick)
@@ -400,7 +400,7 @@ def _run_track(spec: Spec, ws, *, resume: bool, candidates: int | None = None, f
record = get_track(spec.track, **options).run(spec, ws, resume=resume, force=force)
except KeyboardInterrupt:
raise C.CliError(
- f"interrupted; resume with `3dcode resume {ws.root.name}`", code=130
+ f"interrupted; resume with `3dcodeverse resume {ws.root.name}`", code=130
) from None
except Exception as e: # the track failed outside its own error handling
from codeverse3d.orchestrator import StateCorrupt
@@ -430,7 +430,7 @@ def _pick_and_package(ws, *, by: str, round_index: int | None = None,
idx = round_index if round_index is not None else select.pick(ws.root, by=by, pairwise_model=judge_model)
if idx is None:
- warn("no round was judged, so none was picked: name one with `3dcode pick <slug> --round N`")
+ warn("no round was judged, so none was picked: name one with `3dcodeverse pick <slug> --round N`")
return None
try:
select.package(ws.root, idx, method="round" if round_index is not None else by)
@@ -513,7 +513,7 @@ def resume(
C.check_minutes(max_minutes)
ws = C.open_workspace(slug, runs_dir)
# the run mutex, before the spec is rewritten
- with C.mutating(ws, what=f"3dcode resume {ws.root.name}"):
+ with C.mutating(ws, what=f"3dcodeverse resume {ws.root.name}"):
spec = C.load_spec(ws)
if candidates and candidates > 1 and not TRACK_INFO[spec.track].best_of_n:
raise _no_best_of_n(spec.track)
@@ -528,7 +528,7 @@ def resume(
if not force and (why := _finished_reason(ws, raised)):
raise C.CliError(
f"run {ws.root.name} already finished ({why}); nothing to resume. "
- f"`3dcode status {ws.root.name}` to look at it, or --force to re-enter it "
+ f"`3dcodeverse status {ws.root.name}` to look at it, or --force to re-enter it "
f"(that re-plans, re-scores and overwrites the final state)."
)
if raised:
@@ -560,7 +560,7 @@ def pick(
raise C.CliError("--judge is the pairwise judge: it goes with --by pairwise (and no --round)", code=2)
ws = C.open_workspace(slug, runs_dir)
# writes deliverable/ + selection.json: one writer per run dir
- with (C.mutating(ws, what=f"3dcode pick {ws.root.name}", action="pick"),
+ with (C.mutating(ws, what=f"3dcodeverse pick {ws.root.name}", action="pick"),
run_ledger(ws.root, run=ws.root.name, create=False)):
try:
idx = _pick_and_package(ws, by=by, round_index=round_index, judge_model=judge)
diff --git a/harness/codeverse3d/cli/skills_cmd.py b/harness/codeverse3d/cli/skills_cmd.py
index 6d88df43..008f8c31 100644
--- a/harness/codeverse3d/cli/skills_cmd.py
+++ b/harness/codeverse3d/cli/skills_cmd.py
@@ -1,4 +1,4 @@
-"""``3dcode skills`` — list and validate the skill library, and report how often agents read it."""
+"""``3dcodeverse skills`` — list and validate the skill library, and report how often agents read it."""
from __future__ import annotations
@@ -36,7 +36,7 @@ def list_skills(
roots = _roots()
found = [(root, s) for root in roots for s in iter_skills(root.path)]
overlay_rows = {root.variant: _overlay_routes(root.variant) for root in roots if root.variant}
- t = Table(title="3dcode skills")
+ t = Table(title="3dcodeverse skills")
for col in ("name", "variant", "evidence", "verified", "lines", "~tokens", "refs", "target", "dir",
*(["routes"] if routes else [])):
t.add_column(col)
diff --git a/harness/codeverse3d/config.py b/harness/codeverse3d/config.py
index e839dce8..0a50fff6 100644
--- a/harness/codeverse3d/config.py
+++ b/harness/codeverse3d/config.py
@@ -58,7 +58,7 @@ Flag = Annotated[bool, WrapValidator(_or_default)]
Count = Annotated[int, Field(ge=0), WrapValidator(_or_default)]
-#: the supported Blender (5.2 LTS); ``3dcode doctor`` fails anything older.
+#: the supported Blender (5.2 LTS); ``3dcodeverse doctor`` fails anything older.
BLENDER_MIN: tuple[int, int] = (5, 2)
@@ -177,7 +177,7 @@ class Judge(BaseModel):
description="the in-run VLM judge's thinking level (C3D_JUDGE__THINKING; models/gemini.THINKING_BUDGET). "
"'low' = 1024 tokens is enough for gemini-3.1-pro; gemini-4-argon thinks only from 'medium' "
"(4096) — measured 2026-10-14 on 19 scene/object rounds: argon@medium within-sd .012 vs .029 at low, "
- "same cost, same mean (D105). `3dcode judge --thinking` replays with another level.")
+ "same cost, same mean (D105). `3dcodeverse judge --thinking` replays with another level.")
class Placement(BaseModel):
diff --git a/harness/codeverse3d/doctor.py b/harness/codeverse3d/doctor.py
index 58273c3c..66cd9fd1 100644
--- a/harness/codeverse3d/doctor.py
+++ b/harness/codeverse3d/doctor.py
@@ -1,4 +1,4 @@
-"""Environment checks behind ``3dcode doctor`` — python deps, Blender, node/three/
+"""Environment checks behind ``3dcodeverse doctor`` — python deps, Blender, node/three/
puppeteer, the GPU probe, keys, the machine-wide in-flight slots, the vendor CLIs, MCP.
"""
diff --git a/harness/codeverse3d/judges/base.py b/harness/codeverse3d/judges/base.py
index cb04ccac..17aa6973 100644
--- a/harness/codeverse3d/judges/base.py
+++ b/harness/codeverse3d/judges/base.py
@@ -51,7 +51,7 @@ class JudgeInput(BaseModel):
)
glb_path: str | None = Field(
default=None,
- description="the round's canonical GLB (object tracks fill it; `3dcode judge` fills it from the stored "
+ description="the round's canonical GLB (object tracks fill it; `3dcodeverse judge` fills it from the stored "
"build) — the D48 slice channel cuts it on gate-ERROR rounds; None disables the channel",
)
diff --git a/harness/codeverse3d/models/pricing.py b/harness/codeverse3d/models/pricing.py
index 7709cbe8..cbe60d19 100644
--- a/harness/codeverse3d/models/pricing.py
+++ b/harness/codeverse3d/models/pricing.py
@@ -136,7 +136,7 @@ def lookup_price(provider: str, model: str) -> Price | None:
@dataclass(frozen=True)
class PriceRow:
- """A resolved price plus its audit trail — what ``3dcode cost prices`` prints."""
+ """A resolved price plus its audit trail — what ``3dcodeverse cost prices`` prints."""
provider: str
model: str
diff --git a/harness/codeverse3d/models/retry.py b/harness/codeverse3d/models/retry.py
index 1b94cae2..100222e6 100644
--- a/harness/codeverse3d/models/retry.py
+++ b/harness/codeverse3d/models/retry.py
@@ -149,7 +149,7 @@ class Slots:
os.close(fd)
def busy(self) -> int:
- """How many slots are held machine-wide (``3dcode doctor``); never takes one."""
+ """How many slots are held machine-wide (``3dcodeverse doctor``); never takes one."""
n = 0
for i in range(self.n):
try:
diff --git a/harness/codeverse3d/orchestrator.py b/harness/codeverse3d/orchestrator.py
index 793e4e08..ab35a872 100644
--- a/harness/codeverse3d/orchestrator.py
+++ b/harness/codeverse3d/orchestrator.py
@@ -138,7 +138,7 @@ class StageRunner:
except (OSError, ValueError) as e: # ValidationError is a ValueError
if self.frozen:
raise RuntimeError(f"stage {name}: its recorded result {path} is unreadable ({e}) and rounds exist, "
- "so it is never re-run; `3dcode resume --force` regenerates the stages") from e
+ "so it is never re-run; `3dcodeverse resume --force` regenerates the stages") from e
self.events.emit("stage.cache_invalid", stage=name, inputs_hash=h, path=str(path),
error=f"{type(e).__name__}: {e}")
log.warning("stage %s: cached result at %s is unusable (%s); re-running", name, path, e)
diff --git a/harness/codeverse3d/proc.py b/harness/codeverse3d/proc.py
index a5ddb99b..284effed 100644
--- a/harness/codeverse3d/proc.py
+++ b/harness/codeverse3d/proc.py
@@ -485,9 +485,9 @@ def _held_message(run_root: Path | str, held: dict, *, action: str) -> str:
name = Path(run_root).name
return (
f"refusing to {action} run {name}: it is being run right now by pid {held.get('pid', '?')} "
- f"(started {started}: {held.get('what') or '3dcode'}). Two processes on one run "
- f"corrupt each other's state. Look at it with `3dcode status {name}`, or stop that "
- f"ONE process with `kill {held.get('pid', '?')}` — never `pkill -f 3dcode`, which kills every "
+ f"(started {started}: {held.get('what') or '3dcodeverse'}). Two processes on one run "
+ f"corrupt each other's state. Look at it with `3dcodeverse status {name}`, or stop that "
+ f"ONE process with `kill {held.get('pid', '?')}` — never `pkill -f 3dcodeverse`, which kills every "
f"other run on this machine too."
)
diff --git a/harness/codeverse3d/prompts/system/harness_contract.md b/harness/codeverse3d/prompts/system/harness_contract.md
index edcab061..72c4fae2 100644
--- a/harness/codeverse3d/prompts/system/harness_contract.md
+++ b/harness/codeverse3d/prompts/system/harness_contract.md
@@ -1,6 +1,6 @@
# 3dcodeverse harness contract (read once, obey always)
-You are a **3D code author** working inside the `3dcodeverse` harness (CLI `3dcode`).
+You are a **3D code author** working inside the `3dcodeverse` harness (CLI `3dcodeverse`).
You write RAW executable code in one language (bpy / CadQuery / three.js / URDF+bpy /
multi-file three.js+GLSL). The harness — not you — builds, exports, measures, renders
and judges. Your code is the deliverable; every GLB / PNG / URDF export is derived from it.
diff --git a/harness/codeverse3d/prompts/variants/argon/README.md b/harness/codeverse3d/prompts/variants/argon/README.md
index 16269238..40372158 100644
--- a/harness/codeverse3d/prompts/variants/argon/README.md
+++ b/harness/codeverse3d/prompts/variants/argon/README.md
@@ -20,7 +20,7 @@ restores the previous snapshot on rollback.
* `variant_hash("argon")` (`codeverse3d/prompts/variants.py`) covers both halves, this README
excluded; the run record names the variant, its version and that hash.
-How a run picks it: `3dcode make --variant argon`, `C3D_VARIANT=argon`, or
+How a run picks it: `3dcodeverse make --variant argon`, `C3D_VARIANT=argon`, or
`C3D_VARIANTS='{"agy:gemini-4-argon": "argon"}'` (an exact generator id, looked up as typed
and then by `<kind>:<served model>` so the effort suffix does not matter). Never a pattern
over the model id (Law 4).
diff --git a/harness/codeverse3d/prompts/variants/argon/variant.toml b/harness/codeverse3d/prompts/variants/argon/variant.toml
index 8fd63fc0..1d7de7cd 100644
--- a/harness/codeverse3d/prompts/variants/argon/variant.toml
+++ b/harness/codeverse3d/prompts/variants/argon/variant.toml
@@ -6,7 +6,7 @@
# so iteration 0 of eval/bench/loop.py measures the base prompts under the variant plumbing;
# the loop writes files beside this one and bumps `version` on accept.
#
-# Selection is explicit, never a regex over the model id: `3dcode make --variant argon`,
+# Selection is explicit, never a regex over the model id: `3dcodeverse make --variant argon`,
# C3D_VARIANT=argon, or C3D_VARIANTS='{"agy:gemini-4-argon": "argon"}' (by the served model, so
# agy:gemini-4-argon-high maps too). `generators` below is informative.
diff --git a/harness/codeverse3d/skills/__init__.py b/harness/codeverse3d/skills/__init__.py
index 9f5e8db6..5a6f0a7a 100644
--- a/harness/codeverse3d/skills/__init__.py
+++ b/harness/codeverse3d/skills/__init__.py
@@ -115,7 +115,7 @@ class ShippedRoot:
def shipped_roots(variants: Iterable[str] | None = None) -> list[ShippedRoot]:
"""The base root, then the skills overlay of every shipped variant that has one.
- ``3dcode skills validate`` / ``list`` walk this so a malformed overlay bundle fails validation
+ ``3dcodeverse skills validate`` / ``list`` walk this so a malformed overlay bundle fails validation
instead of silently not routing. ``variants`` defaults to every variant ``list_variants`` knows.
"""
from codeverse3d.prompts.variants import list_variants
diff --git a/harness/codeverse3d/tracks/lifecycle.py b/harness/codeverse3d/tracks/lifecycle.py
index 9109b4f5..515b434d 100644
--- a/harness/codeverse3d/tracks/lifecycle.py
+++ b/harness/codeverse3d/tracks/lifecycle.py
@@ -330,7 +330,7 @@ class BaseTrack:
f"{len(journal)} recorded round(s) were built from (plan-input fingerprint "
f"{stored} != {fp}: prompt/track/language/constraints/references/planner/seed). "
f"Resuming would record those rounds against the new spec. Fork a new run "
- f"(`3dcode make`), or resume with --force to archive the old rounds under "
+ f"(`3dcodeverse make`), or resume with --force to archive the old rounds under "
f"rounds/pre_force/ and re-plan. Raising budget caps alone never trips this."
)
events.emit("resume.spec_changed", old_fingerprint=stored, new_fingerprint=fp,
diff --git a/harness/docs/ARCHITECTURE.md b/harness/docs/ARCHITECTURE.md
index e3f33bf6..a9b0a492 100644
--- a/harness/docs/ARCHITECTURE.md
+++ b/harness/docs/ARCHITECTURE.md
@@ -1,6 +1,6 @@
# 3dcodeverse harness — architecture
-Package `codeverse3d`, CLI `3dcodeverse` (short alias `3dcode`), repo location `harness/` in the
+Package `codeverse3d`, CLI `3dcodeverse`, repo location `harness/` in the
3dcodeverse checkout. Backend only. Python 3.13 (one fixed version), a small Node runtime
(`runtime_js/`) for everything Three.js / headless Chrome, and moderngl for the graphics track.
How-to is in `docs/RUNBOOK.md`, the skill library in
@@ -29,7 +29,7 @@ of the language — for four **tracks**:
Every run is: **spec → plan → skeleton → [scene stages] → baseline round (optionally best-of-N
candidates) → `max_rounds` refine rounds → finalise → record**. Which round to hand over is
-decided AFTER the run by the `addons/select` reader (`3dcode pick`; `3dcode make` calls it by
+decided AFTER the run by the `addons/select` reader (`3dcodeverse pick`; `3dcodeverse make` calls it by
score): it writes `deliverable/` for that round. The record, the git history of `src/` and every
round's kept build are the flywheel unit.
@@ -80,14 +80,14 @@ codeverse3d/
(GateFinding, GateReport, BuildResult.gates, RenderView, RenderSet, Judgment), run
(+ SkillsUsage/SkillRead), chat, agent (typed AgentJob)
workspace.py run-dir layout + git snapshots
- doctor.py the environment checks behind `3dcode doctor` (deps, Blender, node, GPU probe, keys,
+ doctor.py the environment checks behind `3dcodeverse doctor` (deps, Blender, node, GPU probe, keys,
pool admission, vendor CLIs, MCP, skills)
proc.py stdlib-only subprocess + atomic-JSON primitives (ManagedProcess owns every child's
lifecycle: group kill on ANY exception, bounded pumps, stdin writer; run_subprocess,
write_json_atomic, scrub_secrets), the tolerant readers/writer (read_json_or_none,
iter_jsonl_lines, read_jsonl_lenient, append_jsonl_line), the JSONL event log (EventLog),
the run lock (ONE writer per run dir: fcntl.flock at <runs>/.locks/<slug>.lock naming the
- holder, printed by `3dcode status`), fan_out, sha256_file, version_line. RULE: any
+ holder, printed by `3dcodeverse status`), fan_out, sha256_file, version_line. RULE: any
stdlib-only file / JSON / JSONL helper lives HERE. A leaf: imports nothing from codeverse3d
models/ ChatModel (base.py), parts.py; gemini.py (request/response shapes, dead-key + 429 rotation);
anthropic.py openai.py; retry.py (KeyPool with outcomes ok | 429 | 5xx | error | dead | skip,
@@ -136,9 +136,9 @@ codeverse3d/
MeteredAgent — one row per ChatModel.generate; one session row for a backend that does NOT
meter itself; run_ledger is context-local so parallel batteries work) tally.py (a block's
money = its ledger rows, its lost seconds → StepTime) guard.py (a call's cost estimated
- before it is sent — `3dcode cost estimate`)
+ before it is sent — `3dcodeverse cost estimate`)
judges/ base.py (JudgeInput; round_input + plan_summary, the ONE payload builder the in-run judge,
- `3dcode judge`, calibration and eval all use), rubrics.py + rubrics/*.yaml (defect
+ `3dcodeverse judge`, calibration and eval all use), rubrics.py + rubrics/*.yaml (defect
checklists, the wire schema, caps and scoring), prompt_builder.py (image prep, montages, the
judge messages), vlm_judge.py (+ the reference/likeness judges), pairwise.py. No Judge
Protocol: a judge is duck-typed `.judge(JudgeInput) -> Judgment`
@@ -166,11 +166,11 @@ codeverse3d/
addons/ optional tools that READ finished runs; outside cli/ nothing imports them
(tests/core/test_addons_boundary.py). dataset/ = export.py, pack.py, pairs.py, refine.py,
captions.py, index.py, sample.py, quality.py (tiers + code fingerprints + duplicate groups)
- — `3dcode flywheel …`; costreport/ = audit.py, report.py, caching.py — `3dcode cost`;
+ — `3dcodeverse flywheel …`; costreport/ = audit.py, report.py, caching.py — `3dcodeverse cost`;
calibration.py (judge repeatability over recorded runs); skill_targets.py (measured targets
+ claims per skill bundle); select.py (round_rows / pick / summarise / package: which round
- of a finished run to hand over — `3dcode pick`, and `3dcode make` after the run)
- addons/gallery/ THE local run gallery (`3dcode gallery serve|build`): cards.py, compare.py (side-by-side
+ of a finished run to hand over — `3dcodeverse pick`, and `3dcodeverse make` after the run)
+ addons/gallery/ THE local run gallery (`3dcodeverse gallery serve|build`): cards.py, compare.py (side-by-side
arms), index.py (run roots → typed RunEntry, tolerant of half-written records), model.py,
page.py (cards + table + filters + per-filter summary), detail.py (/run/<battery>/<slug>),
code.py (src browser), viewer.py (GLB orbit viewer on the vendored three.js), urls.py
@@ -190,7 +190,7 @@ codeverse3d/
cli/ main.py (app wiring, make/resume/mcp + the doctor/tools/gallery commands), inspect_cmd.py
(render/judge on one existing run), layout_cmd.py (THE single-run view: `show` = STATUS /
DELIVERABLE / QUALITY EVIDENCE / COST & SETTINGS; `status` = `show --section status`),
- flywheel_cmd.py, cost_cmd.py (`3dcode cost`), skills_cmd.py (`3dcode skills
+ flywheel_cmd.py, cost_cmd.py (`3dcodeverse cost`), skills_cmd.py (`3dcodeverse skills
list|show|validate|report`)
eval/bench/ the battery launcher and reports (`python -m bench.run_bench` / `bench.report`,
compare_backends.py, ab_plan.py, complexity_report.py, prompts/*.yaml); never imported by
@@ -209,7 +209,7 @@ tests/ one directory per package area (the count is `pytest --col
One run is one directory, `runs/<slug>/`: `src/` is the code (git, one commit per round),
`artifacts/` the last round's build plus every round's own `artifacts/rNN/`, and a pick writes
`deliverable/` + `selection.json`; `telemetry/` holds the cost ledger and settings
-(`workspace.py` names the buckets; `3dcode show <slug>` walks a run).
+(`workspace.py` names the buckets; `3dcodeverse show <slug>` walks a run).
## 4. Per-language authoring contracts (raw code; the harness owns export)
@@ -282,7 +282,7 @@ shuffled per sample. The wire schema is **observe-then-score** (summary, streng
top; degraded verdicts are glitches, not scores.
The judge's structure facts come from the gates, rendered from the stored `GateReport` so
-`3dcode judge <slug>` and calibration see exactly what the in-run judge saw: on the static track
+`3dcodeverse judge <slug>` and calibration see exactly what the in-run judge saw: on the static track
the connectivity gate's **contact ledger** (parts / contacts / floating / deepest overlap, the
plan's PLANNED JOINS as contact or OPEN with the gap, the lowest point above the floor); on an
object-track round whose connectivity gate carries an ERROR, the payload grows **cross-section
@@ -291,7 +291,7 @@ GLB on the two vertical centre planes, red hatch only on the gate's ERROR penetr
appended after the montages with a factual rig-text block. A clean round's payload is
byte-identical to the unconditional one (test-pinned).
-`PairwiseJudge` (position-swapped, tie on disagreement; only `3dcode pick --by pairwise` asks
+`PairwiseJudge` (position-swapped, tie on disagreement; only `3dcodeverse pick --by pairwise` asks
it); `ReferenceJudge` for image-conditioned specs. Rubrics (pass threshold 0.70 unless named):
`static_object` (0.72), `articulated` (requires pose views: the `missing_pose_sheet` cap), `scene`
(placement caps floating 0.60 / sunken 0.65 / interpenetration 0.70 / scale 0.75 / free_exempt
@@ -324,7 +324,7 @@ the run's minutes, a failing sibling raised after every sibling has finished and
under a `Settings` switch (`layouts`: `zone_layouts`, on); `*` is not cached — it runs every
session. Every other stage is cached under its key (`stages/<name>.json`): what its result
depends on, named in the declaration. **Once rounds exist** a recorded stage is served whatever
-its key (`stage.frozen`), and the skeleton never runs at all; `3dcode resume --force` is the one
+its key (`stage.frozen`), and the skeleton never runs at all; `3dcodeverse resume --force` is the one
way to regenerate a stage under them: it archives the round journal + `record.json` to
`rounds/pre_force/` and the run starts again at round 0. The plan itself is `run()`'s, before
the graph (`plan_stage_key`).
@@ -373,7 +373,7 @@ then max_rounds refine rounds, EACH built on the round before it — a FIXED cou
stop reasons: max_rounds | budget (the wall clock — BudgetGuard / max_minutes) | agent_quota | no_change |
no_refine_tasks | judge_unavailable | failed
finalise: the workspace ENDS AT THE LAST ROUND; finalize_record → record.json. No best round, no deliverable/
-after the run (`3dcode make`, unless --no-pick): addons/select.pick(by="score") — highest effective score, ties →
+after the run (`3dcodeverse make`, unless --no-pick): addons/select.pick(by="score") — highest effective score, ties →
fewer gate errors → the earlier round — then package(): deliverable/ + selection.json
```
@@ -395,8 +395,8 @@ the shader-traps skill). `scene_probe` also holds the scene draw budget: the ce
## 8. Cost and time
-`docs/RUNBOOK.md` §2 gives expected cost and time per track. `3dcode cost <slug>` prints a
-run's recorded spend, `3dcode cost estimate` a call's expected price.
+`docs/RUNBOOK.md` §2 gives expected cost and time per track. `3dcodeverse cost <slug>` prints a
+run's recorded spend, `3dcodeverse cost estimate` a call's expected price.
## 9. Known limits
@@ -421,13 +421,13 @@ run's recorded spend, `3dcode cost estimate` a call's expected price.
* Anthropic / OpenAI backends are mock-tested only.
* Budget checks run between steps: a round the clock cuts mid-way is rolled back (its spend
stays in `aborted_rounds`) — give scenes `--max-minutes 60`.
-* A sustained Gemini outage can still fail a round (`3dcode resume` re-uses cached stages). A
+* A sustained Gemini outage can still fail a round (`3dcodeverse resume` re-uses cached stages). A
storm-dead agent session gets the round loop's ONE transport retry (fresh key), no single-shot
fallback.
## 10. Flywheel
-`record.json` per run + git history; `3dcode flywheel export` writes sample folders
+`record.json` per run + git history; `3dcodeverse flywheel export` writes sample folders
(`<out>/<track>/<language>/<slug>/{code.<ext>, src/**, robot.urdf, meta.json, captions.json,
renders/}`) of the PICKED round (the last built round when none was judged) with **quality
tiers** from that round's verdict (A passed & 0 gate errors, B passed, C score ≥ 0.6, D else),
@@ -441,13 +441,13 @@ written in response, both code snapshots and whether the score moved
(`addons/dataset/refine.RefineTransition`). That is the supervision the harness produces that a
one-shot corpus cannot. `finetune/llamafactory/` turns those rows into training files; the
harness writes the measurement, not the trainer's format. `flywheel caption` adds {detailed,
-instruction, factory} captions (`--out` for side-car mode); the gallery is `3dcode gallery build
+instruction, factory} captions (`--out` for side-car mode); the gallery is `3dcodeverse gallery build
--embed`.
-`codeverse3d/addons/gallery/` is the **local** answer to the same question: `3dcode gallery
+`codeverse3d/addons/gallery/` is the **local** answer to the same question: `3dcodeverse gallery
serve` indexes `runs/` + every `eval/bench/out/*/runs`, serves the page **and the run
directories** on 127.0.0.1 (sheet, renders, `src/`, `object.glb` in an orbit viewer built on the
vendored three.js, `record.json`), and re-reads a run's record per request so a battery that is
still writing shows up live (RUNBOOK §4). A URL can only name a `(battery, slug)` the scanner
found under a declared root, and `urls.safe_join` refuses anything that escapes that run
-directory. `3dcode gallery build [--embed]` writes the same page as one file.
+directory. `3dcodeverse gallery build [--embed]` writes the same page as one file.
diff --git a/harness/docs/INSTALL.md b/harness/docs/INSTALL.md
index 97a71bef..6bc427ab 100644
--- a/harness/docs/INSTALL.md
+++ b/harness/docs/INSTALL.md
@@ -59,7 +59,7 @@ offline suite and `ruff` are run locally before every push (see `CLAUDE.md`).
|---|---|---|---|
| **python** | **3.13** | `requires-python = ">=3.13"`; ruff `target-version = "py313"`; `PY_FLOOR` in `tests/core/test_portability.py`; `MIN_PY_MINOR` in `setup.sh` | the four places are pinned together by the portability test |
| **node** | **20.6.0+** (developed on 24 LTS) | `runtime_js/package.json` `engines.node`; `codeverse3d.spatial.node.NODE_MIN` fails every node workload with an actionable message | 20.6 is the `--import` module-hook floor |
-| **Blender** | **5.2 LTS** only | `BLENDER_MIN = (5, 2)` in `codeverse3d/config.py`; `3dcode doctor` fails anything older; `Settings.resolve_blender()` | `blender`, `urdf_blender` and `scene_blender` are the users; the prompts and lints state 5.2 API facts |
+| **Blender** | **5.2 LTS** only | `BLENDER_MIN = (5, 2)` in `codeverse3d/config.py`; `3dcodeverse doctor` fails anything older; `Settings.resolve_blender()` | `blender`, `urdf_blender` and `scene_blender` are the users; the prompts and lints state 5.2 API facts |
| **OS** | Linux x86_64 (WSL2 works) | — | macOS should work (nothing is Linux-specific except `resource.setrlimit` guards) but is not tested |
Moving to another Python later is the same four-line change (`requires-python`, ruff
@@ -90,7 +90,7 @@ cd 3dcodeverse
```
Layout: the harness is the `harness/` subdirectory (python dist `3dcodeverse`, import package
-`codeverse3d`, CLIs `3dcodeverse` and `3dcode`). See the repo `README.md` for the other
+`codeverse3d`, CLI `3dcodeverse`). See the repo `README.md` for the other
top-level components.
Use a virtualenv or a conda env. It is only optional when your interpreter already owns its
@@ -111,8 +111,8 @@ pip install -e harness # core only
pip install -e 'harness[all,dev]' # everything (what setup.sh does)
```
-`-e` (editable) is the intended mode: the harness is developed in place and the console scripts
-(`3dcodeverse`, `3dcode`) point back at the checkout.
+`-e` (editable) is the intended mode: the harness is developed in place and the console script
+`3dcodeverse` points back at the checkout.
Core dependencies (always installed): pydantic + pydantic-settings (contracts, settings), typer +
rich (CLI), jinja2 (prompt templates), pyyaml, numpy, trimesh (mesh measurement), python-fcl
@@ -124,8 +124,8 @@ rich (CLI), jinja2 (prompt templates), pyyaml, numpy, trimesh (mesh measurement)
| `urdf` | yourdfpy, scipy | the `articulated_object` track: URDF parse/validate + joint math |
| `graphics` | moderngl | the `graphics` track (`glsl_shader`, `opengl_python`) — headless GL rendering in `spatial/gl_render.py` |
| `mesh` | shapely, networkx, matplotlib | `cross_section` (filled area / hollow ratio from shapely, the image drawn in matplotlib), mesh split / connectivity (networkx), judge cross-section slices on gate-ERROR rounds (without the extra the judge logs a warning and sends the payload without slices) |
-| `mcp` | mcp | `3dcode mcp` / `codeverse3d.spatial.mcp_server`, i.e. spatial tools for the `agy`, `claude-code` and `codex` backends |
-| `flywheel` | pyarrow | `3dcode flywheel export --pack` (parquet shards). A plain `flywheel export` needs nothing: without pyarrow it writes `metadata.jsonl` and skips `metadata.parquet` with a warning |
+| `mcp` | mcp | `3dcodeverse mcp` / `codeverse3d.spatial.mcp_server`, i.e. spatial tools for the `agy`, `claude-code` and `codex` backends |
+| `flywheel` | pyarrow | `3dcodeverse flywheel export --pack` (parquet shards). A plain `flywheel export` needs nothing: without pyarrow it writes `metadata.jsonl` and skips `metadata.parquet` with a warning |
| `all` | all of the above | — |
| `dev` | pytest, pytest-timeout, ruff | the test suite and the linter |
@@ -257,7 +257,7 @@ exec "$BLENDER_HOME/blender" "$@"
frames: Cycles on CUDA with GPU denoising and persistent data when a CUDA device exists, CPU
otherwise; `Settings.render.blender_frame_time_limit_s` (default 20 s) caps a heavy scene's
sampling per frame. **OptiX is never enumerated** (it is unavailable under WSL2 and probing it
-crashes Blender). `3dcode doctor` times one small frame on exactly this path.
+crashes Blender). `3dcodeverse doctor` times one small frame on exactly this path.
### How the harness finds it
@@ -439,16 +439,16 @@ real data:
```bash
cd harness
-3dcode make "a slow rotating neon hex grid" \
+3dcodeverse make "a slow rotating neon hex grid" \
--track graphics --language glsl_shader \
--generator single-shot:gemini:gemini-3.8-flash \
--rounds 0 --max-minutes 20 \
--runs-dir /tmp/c3d_smoke
-3dcode status <slug> --runs-dir /tmp/c3d_smoke
+3dcodeverse status <slug> --runs-dir /tmp/c3d_smoke
```
If node and Blender are installed, the equivalent object-track smoke is
-`3dcode make "a simple wooden stool" --track static_object --language blender
+`3dcodeverse make "a simple wooden stool" --track static_object --language blender
--generator single-shot:gemini:gemini-3.8-flash --rounds 0` (~$0.1, a few minutes).
Artifacts land in `<runs-dir>/<slug>/`:
@@ -465,12 +465,12 @@ artifacts/judge/rNN.json the VLM verdict and the computed score
trajectories/ per-call transcripts
```
-`3dcode status <slug>` prints the rounds table, cost and last events; the full operating guide is
+`3dcodeverse status <slug>` prints the rounds table, cost and last events; the full operating guide is
`docs/RUNBOOK.md`.
> **`capacity storm N/10 … waiting Ns`** means the Gemini endpoint is busy, not that your
> install is broken: the harness backs off (up to 10 waits, ~10 min in total) and rotates keys,
-> which is why the smoke above gets `--max-minutes 20`. `3dcode resume <slug> --runs-dir …`
+> which is why the smoke above gets `--max-minutes 20`. `3dcodeverse resume <slug> --runs-dir …`
> continues from the last completed stage. To prove the key path alone, `3dcodeverse doctor
> --live` is the one-call version.
@@ -479,7 +479,7 @@ trajectories/ per-call transcripts
## 11. Uninstall / cleanup
```bash
-pip uninstall 3dcodeverse # removes the editable install + 3dcodeverse/3dcode
+pip uninstall 3dcodeverse # removes the editable install + the 3dcodeverse command
rm -rf harness/runtime_js/node_modules # re-creatable with `npm ci`
rm -rf ~/.cache/puppeteer # downloaded Chrome builds
rm -rf ~/.cache/codeverse3d # harness cache — see below
diff --git a/harness/docs/RUNBOOK.md b/harness/docs/RUNBOOK.md
index 3f3edf92..ea276d33 100644
--- a/harness/docs/RUNBOOK.md
+++ b/harness/docs/RUNBOOK.md
@@ -1,15 +1,15 @@
# Runbook — operating the harness
Paths are relative to the harness root (`harness/` in the checkout) unless absolute. The CLI is
-`3dcodeverse` with short alias `3dcode` (used below). Installing: `docs/INSTALL.md`.
+`3dcodeverse`. Installing: `docs/INSTALL.md`.
## 1. Install and check
```bash
-pip install -e harness # once, from the checkout root; entry points 3dcodeverse and 3dcode
+pip install -e harness # once, from the checkout root; entry point 3dcodeverse
cd harness/runtime_js && npm install # [email protected], puppeteer (chrome cached)
-3dcode doctor # python deps, blender, node/three/puppeteer, chrome WebGL, keys, CLIs, git, mcp
-3dcode doctor --live # + one tiny Gemini call ("pong", ~$0.00001)
+3dcodeverse doctor # python deps, blender, node/three/puppeteer, chrome WebGL, keys, CLIs, git, mcp
+3dcodeverse doctor --live # + one tiny Gemini call ("pong", ~$0.00001)
python -m pytest tests -q # offline suite; live tests are opt-in (blender/node/GL run when the binaries exist)
python -m pytest tests -q -m "not live and not blender and not node" # pure-python subset
python -m pytest tests -q -m live # OPT-IN: real API calls
@@ -23,20 +23,20 @@ and description is its `Settings` field in `codeverse3d/config.py`.
```bash
# static object — Blender (default language, multi-file src/model.py + src/parts/*.py)
-3dcode make "a mid-century wooden dining chair" --track static_object --language blender
-3dcode make "a brass desk lamp" --track static_object --language cadquery
-3dcode make "a classic park bench" --track static_object --language threejs --generator agy:gemini-4-argon
+3dcodeverse make "a mid-century wooden dining chair" --track static_object --language blender
+3dcodeverse make "a brass desk lamp" --track static_object --language cadquery
+3dcodeverse make "a classic park bench" --track static_object --language threejs --generator agy:gemini-4-argon
# articulated object — bpy links + URDF
-3dcode make "a bedside cabinet with one hinged door and one drawer" --track articulated_object --language urdf_blender
+3dcodeverse make "a bedside cabinet with one hinged door and one drawer" --track articulated_object --language urdf_blender
# scene — multi-file three.js + GLSL (+ optional bpy GLB assets chosen by the planner), or bpy on Blender
-3dcode make "a small japanese garden at dusk with a koi pond" --track scene --language scene_threejs --rounds 2 --max-minutes 60
-3dcode make "a small japanese garden at dusk with a koi pond" --track scene --language scene_blender --rounds 2 --max-minutes 90
+3dcodeverse make "a small japanese garden at dusk with a koi pond" --track scene --language scene_threejs --rounds 2 --max-minutes 60
+3dcodeverse make "a small japanese garden at dusk with a koi pond" --track scene --language scene_blender --rounds 2 --max-minutes 90
# graphics — animated shader / raw OpenGL program
-3dcode make "neon cyberpunk rain on a window with bokeh city lights" --track graphics --language glsl_shader
-3dcode make "instanced pastel cubes with bloom" --track graphics --language opengl_python
+3dcodeverse make "neon cyberpunk rain on a window with bokeh city lights" --track graphics --language glsl_shader
+3dcodeverse make "instanced pastel cubes with bloom" --track graphics --language opengl_python
```
-Useful flags (`3dcode make --help`):
+Useful flags (`3dcodeverse make --help`):
* `--generator`, `--planner`, `--judge` — backend ids (§3).
* `--image <png>` (repeatable) — reference images → `ReferenceJudge` + silhouette gate + IoU
@@ -46,7 +46,7 @@ Useful flags (`3dcode make --help`):
* `--candidates N` — best-of-N baseline: N parallel candidates in `<ws>/_cand/`, quick-judged,
the highest quick score kept (fewer gate errors on a tie); multiplies baseline cost ≈ N;
object and graphics tracks only (a scene writes its baseline in its stages).
-* `--no-pick` — package nothing (no `deliverable/`, no `selection.json`; `3dcode pick` later).
+* `--no-pick` — package nothing (no `deliverable/`, no `selection.json`; `3dcodeverse pick` later).
* `--max-minutes`, `--dim height=0.45`, `--must`, `--must-not`, `--style`, `--tag`, `--seed` (a
label: it re-keys the plan, nothing else is seeded by it), `--slug`, `--runs-dir`, `--force`,
`--no-run` (workspace + spec.json only).
@@ -58,7 +58,7 @@ Useful flags (`3dcode make --help`):
Expected cost/time with gemini-3.8-flash: object tracks ≈ $0.7–0.9 and 12–36 min for baseline +
1 refine; best-of-2 single-shot ≈ $0.25 / 8 min; graphics single-shot ≈ $0.05 / 2 min per judged
round; scenes ≈ $2.3 and 30 min before the first judged round, then ≈ $0.36 / ~7 min per refine
-(give scenes `--max-minutes 60`). Actual spend: `3dcode cost <slug>`.
+(give scenes `--max-minutes 60`). Actual spend: `3dcodeverse cost <slug>`.
## 3. Backends
@@ -95,22 +95,22 @@ How `agents/backends.py` runs agy (do not undo):
hides only noise (renders/judge output, `trajectories/`); build/census/measurement JSON stay
readable.
* MCP server argv is `[sys.executable, -m, codeverse3d.spatial.mcp_server, --workspace, ws]`;
- server name `3dcode` → tools appear as `mcp__3dcode__<name>` (claude). `3dcode mcp
+ server name `3dcode` → tools appear as `mcp__3dcode__<name>` (claude). `3dcodeverse mcp
--workspace runs/<slug>` execs the same server; `python -m codeverse3d.spatial.mcp_server
--workspace ws --list` prints the tools.
## 4. Where outputs land
Everything lands in `runs/<slug>/` — code in `src/` (git, one commit per round), each round's
-build in `artifacts/rNN/`, and after a pick `deliverable/` + `selection.json`. `3dcode status <slug>` prints the stop reason, baseline → picked score,
+build in `artifacts/rNN/`, and after a pick `deliverable/` + `selection.json`. `3dcodeverse status <slug>` prints the stop reason, baseline → picked score,
the rounds table, cost and the last events.
### Looking at results locally
```bash
-3dcode gallery serve # ./runs + every eval/bench/out/*/runs → http://127.0.0.1:8765/
-3dcode gallery serve ../eval/bench/out/static_v2_flash/runs --port 9000 --reload --no-open
-3dcode gallery build --out gallery.html [--embed] # one self-contained file (--embed inlines the sheets)
+3dcodeverse gallery serve # ./runs + every eval/bench/out/*/runs → http://127.0.0.1:8765/
+3dcodeverse gallery serve ../eval/bench/out/static_v2_flash/runs --port 9000 --reload --no-open
+3dcodeverse gallery build --out gallery.html [--embed] # one self-contained file (--embed inlines the sheets)
```
`serve` indexes the run roots and serves the run **directories** too, so every link works:
@@ -128,7 +128,7 @@ bookmarked. `--reload` re-scans on every page load (a run with no `record.json`
## 5. Resume, re-render, re-judge, export
```bash
-3dcode resume <slug> [--candidates N] [--no-pick]
+3dcodeverse resume <slug> [--candidates N] [--no-pick]
# continues from run_state + stages/*.json and from the LAST round. A
# max_rounds run resumes only with a raised --rounds, a budget run with any
# raised cap; an agent_quota, judge_unavailable or failed run resumes as is.
@@ -137,42 +137,42 @@ bookmarked. `--reload` re-scans on every page load (a run with no `record.json`
# recorded stage never runs again; `resume --force` regenerates the stages:
# it archives the round journal + record.json to rounds/pre_force/ and the
# run starts again at round 0
-3dcode pick <slug> [--by score|pairwise] [--round N] [--judge MODEL]
+3dcodeverse pick <slug> [--by score|pairwise] [--round N] [--judge MODEL]
# hand over a round: deliverable/ + selection.json. score = highest
# effective score, ties → fewer gate errors → the earlier round; pairwise =
# the pairwise judge between the top two when within 0.03; --round N = that
# round. `make`/`resume` run `pick --by score` for you
-3dcode render <slug> [--mode shaded|wire|normals|clay|silhouette] [--out dir]
-3dcode judge <slug> [--round N] [--rubric static_object] [--model gemini:gemini-3.1-pro-preview] [--n 3] [--thinking off|low|medium|high]
+3dcodeverse render <slug> [--mode shaded|wire|normals|clay|silhouette] [--out dir]
+3dcodeverse judge <slug> [--round N] [--rubric static_object] [--model gemini:gemini-3.1-pro-preview] [--n 3] [--thinking off|low|medium|high]
# re-judges a round's recorded renders → artifacts/judge/rNN_cli.json
-3dcode tools list [--cards] · 3dcode tools measure --workspace runs/<slug> · 3dcode tools gl_frames --workspace … --json '{"times":[0,1,2.5]}'
+3dcodeverse tools list [--cards] · 3dcodeverse tools measure --workspace runs/<slug> · 3dcodeverse tools gl_frames --workspace … --json '{"times":[0,1,2.5]}'
# panel says ok | FAIL (the verdict) | error (the tool could not run); exit 1 on either
-3dcode flywheel export runs/ dataset/ [--min-score 0.7] [--only-passed] [--pack] [--include-unbuilt]
+3dcodeverse flywheel export runs/ dataset/ [--min-score 0.7] [--only-passed] [--pack] [--include-unbuilt]
[--captions-dir caps/] [--drop-duplicates]
-3dcode flywheel pairs runs/ pairs.jsonl [--min-delta 0.05]
-3dcode flywheel refine runs/ refine.jsonl [--with-code]
+3dcodeverse flywheel pairs runs/ pairs.jsonl [--min-delta 0.05]
+3dcodeverse flywheel refine runs/ refine.jsonl [--with-code]
# one row per round the harness asked to change (addons/dataset/refine.RefineTransition);
# training formats live in finetune/llamafactory/, not here
-3dcode flywheel caption <slug> [--model …] [--out caps/] # --out = side-car mode, run untouched
+3dcodeverse flywheel caption <slug> [--model …] [--out caps/] # --out = side-car mode, run untouched
python -m codeverse3d.addons.calibration runs/<slug> [runs/<slug2> …] --model gemini:gemini-3.1-pro-preview --n 3 --out out/
# re-judges recorded rounds; writes calibration_<model>.md/.json (never touches runs/)
```
-A run that crashed outside its own handling leaves `record.json` with `status=failed`; `3dcode
+A run that crashed outside its own handling leaves `record.json` with `status=failed`; `3dcodeverse
resume` retries from the last completed stage/round (cached plan/skeleton/scene stages are
reused — this also recovers from a provider outage). A refine session that dies on the
VENDOR's usage limit stops the run at `agent_quota`: every round so far is kept and picked from,
-and `3dcode resume` continues it once the limit resets. Ctrl-C and `kill <pid>` (SIGTERM) are
+and `3dcodeverse resume` continues it once the limit resets. Ctrl-C and `kill <pid>` (SIGTERM) are
safe: `make` / `resume` kill every agent / build process group the run has open, record the
stop, and end within `proc.STOP_GRACE_S` (30 s). Kill by PID, never by pattern: a
-`pkill -f "3dcode make"` takes every other run on the box with it.
+`pkill -f "3dcodeverse make"` takes every other run on the box with it.
Two things to keep apart on one machine: run the offline test suite and a battery at different
times (both hold Blender, a browser pool and most cores; a red browser test under load is not
evidence until it reproduces on a quiet box), and give two concurrent batteries separate
`C3D_CACHE_DIR`s and half the `--parallel` each. A worktree needs its own
`runtime_js/node_modules` (copy, do not symlink: `serve.cjs` refuses a real path outside its
-root) and `PYTHONPATH=<worktree>/harness` before `3dcode make`, because the console script
+root) and `PYTHONPATH=<worktree>/harness` before `3dcodeverse make`, because the console script
imports the editable install, not the cwd.
## 6. Benchmarks
@@ -239,7 +239,7 @@ storm's rows (`--redo-status error,infra_failed,budget`) rather than reading the
* **New rubric**: `judges/rubrics/<name>.yaml` with `pass_threshold`, `criteria[{id, weight,
floor, title, description, anchors, kind}]`, `caps[{id, cap, when:
gate|acceptance|console|missing_views, gate, severity, kinds}]`, `defects[{id, text, penalty,
- cap}]`. Tracks pick rubrics in `tracks/*.py`; `3dcode judge` maps track → rubric in
+ cap}]`. Tracks pick rubrics in `tracks/*.py`; `3dcodeverse judge` maps track → rubric in
`cli/_judge.py::rubric_for`.
* **New backend**: ChatModel → `models/<provider>.py` + registry + prices; CodingAgent →
`agents/backends.py` using `cli_common` + registry + `materialize.py`.
diff --git a/harness/docs/SKILLS.md b/harness/docs/SKILLS.md
index a60e6493..6b4ffe07 100644
--- a/harness/docs/SKILLS.md
+++ b/harness/docs/SKILLS.md
@@ -119,8 +119,8 @@ when the control comes back opened (git reads untracked files too), the session'
is `None`, not a number.
```
-3dcode skills report eval/bench/out/<battery> # deep-read rate per skill per backend (--json)
-3dcode skills validate [--strict] # bundles, overlays, routes, claims
+3dcodeverse skills report eval/bench/out/<battery> # deep-read rate per skill per backend (--json)
+3dcodeverse skills validate [--strict] # bundles, overlays, routes, claims
```
Target: ≥ 60 % deep-read, counted only over sessions the control says were measurable. A skill
@@ -150,14 +150,14 @@ found and opened through each discovery root.
A row may also pin a sentence in a bundle file (`file = "references/x.md"`), a JS constant
(`js = "runtime_js/lib/backdrop.mjs:GROUND_SPAN_M"`), a sequence (`sep`, `last_sep`) and the
line it must stand on (`context = "WARN below"`). The prompt corpus pins its numbers the
- same way in `codeverse3d/prompts/_claims.toml` (`check_prompt_claims`, run by `3dcode skills
+ same way in `codeverse3d/prompts/_claims.toml` (`check_prompt_claims`, run by `3dcodeverse skills
validate`).
4. Its rows in `ROUTES` (`codeverse3d/skills/registry.py`), in the same commit.
5. A `Target` row in `codeverse3d/addons/skill_targets.py` and the matching `target_*`
frontmatter keys — the ONE deterministic quantity the bundle claims to move, its direction,
and its baseline; if no deterministic instrument can see the claim, say so with
`measurable=False` and a `caveat`.
-6. `3dcode skills validate --strict` and `pytest tests/skills`.
+6. `3dcodeverse skills validate --strict` and `pytest tests/skills`.
Body rules: ≤ 350 lines and ≤ 2,500 tokens; no code fence over 12 lines; never restate a frame,
unit or naming rule (`conventions.py` owns those); every corpus percentage carries its battery,
diff --git a/harness/pyproject.toml b/harness/pyproject.toml
index 5148a464..940cc014 100644
--- a/harness/pyproject.toml
+++ b/harness/pyproject.toml
@@ -42,28 +42,20 @@ cad = ["cadquery>=2.4"] # language `cadquery`
urdf = ["yourdfpy>=0.0.56", "scipy>=1.11"] # track `articulated_object` (urdf_blender)
graphics = ["moderngl>=5.8"] # track `graphics` (glsl_shader / opengl_python)
mesh = ["shapely>=2.0", "networkx>=3.2", "matplotlib>=3.8"] # cross_section area, mesh split, judge slices (D48)
-mcp = ["mcp>=1.0"] # `3dcode mcp` spatial-tool server for the agentic CLIs
-flywheel = ["pyarrow>=15"] # `3dcode flywheel export --pack` (parquet)
+mcp = ["mcp>=1.0"] # `3dcodeverse mcp` spatial-tool server for the agentic CLIs
+flywheel = ["pyarrow>=15"] # `3dcodeverse flywheel export --pack` (parquet)
all = ["3dcodeverse[cad,urdf,graphics,mesh,mcp,flywheel]"]
dev = ["pytest>=8", "pytest-timeout>=2.3", "pytest-xdist>=3.6", "ruff>=0.5", "pyarrow>=15"]
[project.scripts]
-3dcode = "codeverse3d.cli.main:app"
3dcodeverse = "codeverse3d.cli.main:app"
[tool.setuptools.packages.find]
include = ["codeverse3d*"]
[tool.setuptools.package-data]
-# Every non-python file a runtime reads at call time. The scene starter tree was once
-# missing and `spatial/js/*` matched nothing at all (no such directory), so a wheel
-# install shipped no starter files: skeleton.write_example() rglob'd an absent
-# directory and wrote ZERO files while reporting success. tests/core/test_portability.py
-# fails on a glob that matches nothing (languages/**/*.md died with the last README).
-# `skills/**/*` carries the SKILL.md bundles + their references/: an installed wheel that
-# ships none of them routes an empty set, and the failure is silent (the run still passes,
-# the skill simply never appears). tests/skills/test_packaging.py builds a wheel and
-# asserts the files are in it.
+# Runtime data, including starter trees and skill bundles. tests/core/test_portability.py
+# checks that every data file matches a glob and every glob matches an existing file.
codeverse3d = ["prompts/**/*", "judges/rubrics/*.yaml", "languages/**/wrappers/*",
"languages/**/starter/**/*", "skills/**/*"]
@@ -85,9 +77,8 @@ timeout = 600
# that happens to have API keys configured fires real requests, and during a provider
# outage it hangs instead of failing — which is exactly what happened on 2026-08-24.
# Everyone's default run is the offline suite (run locally before every push; there is no CI).
-# -n auto: the offline suite is ~1 900 mostly-independent tests and this box has 24
-# cores — serial it is ~3 min, parallel ~15 s. -p no:randomly is NOT set: xdist
-# distributes by file, and every test that touches shared state already uses tmp_path.
+# xdist distributes test items with work stealing. Tests isolate mutable state with tmp_path;
+# nested pytest invocations must use -n0 to avoid launching another worker pool.
addopts = ["-m", "not live", "-n", "auto", "--dist", "worksteal"]
markers = [
"live: needs network + API keys (deselect with -m 'not live')",
diff --git a/harness/runtime_js/README.md b/harness/runtime_js/README.md
index 5d4f928d..751b8346 100644
--- a/harness/runtime_js/README.md
+++ b/harness/runtime_js/README.md
@@ -21,7 +21,7 @@ npm dependencies themselves bottom out at node 18, and `node:util.parseArgs`
older *async* loader hooks (`lib/resolve_three_async.mjs`, a separate loader
thread) instead of the in-thread `module.registerHooks`; the resolutions are
identical. The python side refuses to spawn an older node with an actionable
-message (`codeverse3d.spatial.node.NODE_MIN`, mirrored by `3dcode doctor`'s `node`
+message (`codeverse3d.spatial.node.NODE_MIN`, mirrored by `3dcodeverse doctor`'s `node`
row), and `tests/core/test_portability.py` pins the two declarations together.
The floor is exercised, not assumed: the 139 `node`-marked tests all pass on node
20.19.5 (`C3D_BINARIES__NODE=/path/to/node20 pytest tests -m "node and not live and not blender"`).
diff --git a/harness/tests/agents/test_antigravity.py b/harness/tests/agents/test_antigravity.py
index 4eba6ade..826be50e 100644
--- a/harness/tests/agents/test_antigravity.py
+++ b/harness/tests/agents/test_antigravity.py
@@ -132,11 +132,3 @@ def test_without_keys_agy_is_unavailable_and_never_starts(tmp_ws: Workspace, fak
assert not ok and "API keys" in why
res = a.run(AgentJob(workspace=str(tmp_ws.root), prompt="hello", label="nokey", timeout_s=30))
assert not res.ok and res.exit_reason == "error" and not (tmp_ws.src / "hello.txt").exists()
-
-
-def _effort_job(tmp_ws: Workspace, **kw) -> AgentJob:
- return AgentJob(workspace=str(tmp_ws.root), prompt="hello", timeout_s=30, spatial_tools=False, **kw)
-
-
-
-
diff --git a/harness/tests/agents/test_materialize.py b/harness/tests/agents/test_materialize.py
index 61852889..8256d49b 100644
--- a/harness/tests/agents/test_materialize.py
+++ b/harness/tests/agents/test_materialize.py
@@ -1,4 +1,4 @@
-"""materialize_workspace: bodies, ignore file, codex overrides."""
+"""Codex MCP environment overrides preserve the selected harness."""
from __future__ import annotations
@@ -7,31 +7,7 @@ import subprocess
import sys
import tomllib
-from codeverse3d.agents.cli_common import default_mcp_command
-from codeverse3d.agents.materialize import (
- codex_mcp_overrides,
- materialize_workspace,
-)
-from codeverse3d.workspace import Workspace
-
-CONTRACT = "## blender contract\nWrite pure bpy into src/model.py."
-
-
-def _mat(ws: Workspace, kind: str = "agy"):
- return materialize_workspace(ws, agent_kind=kind, contract_md=CONTRACT,
- mcp_command=default_mcp_command(ws))
-
-
-
-
-
-
-
-
-
-
-
-
+from codeverse3d.agents.materialize import codex_mcp_overrides
def test_codex_mcp_preserves_explicit_engine_selection(tmp_path, monkeypatch):
diff --git a/harness/tests/conftest.py b/harness/tests/conftest.py
index 7000813f..9192869f 100644
--- a/harness/tests/conftest.py
+++ b/harness/tests/conftest.py
@@ -81,15 +81,3 @@ def assert_pid_gone(pid: int, *, group: bool = False, timeout_s: float = 10.0) -
return
time.sleep(0.05)
raise AssertionError(f"{'process group' if group else 'pid'} {pid} survived the kill")
-
-
-#: how many tests this session collected. tests/install/test_hermetic.py needs a
-#: ceiling for the counts docs/INSTALL.md quotes, and used to get it by spawning a
-#: nested `pytest tests --collect-only` — 8-10 s of the suite's wall clock to learn a
-#: number this process already knows. Under xdist every worker collects the whole
-#: suite before running its share, so this is the full count in each of them.
-COLLECTED: dict[str, int] = {}
-
-
-def pytest_collection_modifyitems(session, config, items): # noqa: ARG001
- COLLECTED["n"] = len(items)
diff --git a/harness/tests/core/test_doctor.py b/harness/tests/core/test_doctor.py
index 34579ac9..508529d6 100644
--- a/harness/tests/core/test_doctor.py
+++ b/harness/tests/core/test_doctor.py
@@ -1,4 +1,4 @@
-"""``3dcode doctor``: the GPU probe hands the browser back, and a broken install is a row, never a crash."""
+"""``3dcodeverse doctor``: the GPU probe hands the browser back, and a broken install is a row, never a crash."""
from __future__ import annotations
diff --git a/harness/tests/core/test_proc.py b/harness/tests/core/test_proc.py
index a07729d6..b7108982 100644
--- a/harness/tests/core/test_proc.py
+++ b/harness/tests/core/test_proc.py
@@ -2,7 +2,6 @@
from __future__ import annotations
-import os
import sys
import time
from pathlib import Path
@@ -31,17 +30,6 @@ def test_a_detached_descendant_holding_the_pipes_cannot_extend_the_timeout(tmp_p
assert elapsed < 8.0, f"run_subprocess returned only after {elapsed:.1f}s"
-# --------------------------------------------------------------------------- ManagedProcess lifecycle
-#: spawns a grandchild, writes its OWN pid (== the new session's pgid) to argv[1] and
-#: hangs — so the probe below is a real GROUP probe, not just "the leader died".
-_PID_THEN_SLEEP = (
- "import os, pathlib, subprocess, sys, time\n"
- "subprocess.Popen(['sleep', '300'])\n"
- "pathlib.Path(sys.argv[1]).write_text(str(os.getpid()))\n"
- "time.sleep(300)\n"
-)
-
-
@pytest.mark.timeout(120)
def test_huge_output_is_bounded_head_and_tail(tmp_path: Path):
"""Large output is bounded while preserving its head, tail, and truncation marker; one bad byte
@@ -63,12 +51,3 @@ def test_huge_output_is_bounded_head_and_tail(tmp_path: Path):
assert "bytes dropped" in r.stdout # marker sits between them
cut = r.stdout.index("bytes dropped")
assert "x" * 1000 in r.stdout[:cut] and "x" * 1000 in r.stdout[cut:]
-
-
-def _alive(pid: int) -> bool:
- try:
- os.kill(pid, 0)
- except ProcessLookupError:
- return False
- with open(f"/proc/{pid}/stat") as f: # a zombie is dead
- return f.read().split()[2] != "Z"
diff --git a/harness/tests/core/test_runlock.py b/harness/tests/core/test_runlock.py
index abfddf54..dd536cd2 100644
--- a/harness/tests/core/test_runlock.py
+++ b/harness/tests/core/test_runlock.py
@@ -58,7 +58,7 @@ def test_a_second_process_is_refused_and_a_killed_holder_never_wedges_the_run(tm
msg = str(got.value)
assert str(child.pid) in msg # names the holder, so a human can kill THAT pid
assert "pkill" in msg, "and must warn against the pattern kill that took out 13 runs"
- with pytest.raises(CliError) as ei, mutating(run_root, what="3dcode make slug"):
+ with pytest.raises(CliError) as ei, mutating(run_root, what="3dcodeverse make slug"):
pytest.fail("the mutation boundary must not be entered")
assert ei.value.exit_code == 2
assert (run_root / "spec.json").exists(), "the live holder's workspace must survive"
diff --git a/harness/tests/cost/test_cli.py b/harness/tests/cost/test_cli.py
index 16ccf945..c0d98fef 100644
--- a/harness/tests/cost/test_cli.py
+++ b/harness/tests/cost/test_cli.py
@@ -1,4 +1,4 @@
-"""``3dcode cost``."""
+"""``3dcodeverse cost``."""
from __future__ import annotations
diff --git a/harness/tests/flywheel_cli/test_cli.py b/harness/tests/flywheel_cli/test_cli.py
index 428c2c67..fcae67f3 100644
--- a/harness/tests/flywheel_cli/test_cli.py
+++ b/harness/tests/flywheel_cli/test_cli.py
@@ -17,7 +17,7 @@ runner = CliRunner()
@pytest.fixture
def made_run(tmp_path: Path):
- """`3dcode make --no-run` into a fresh runs dir → ``(runs_dir, run_dir)``."""
+ """`3dcodeverse make --no-run` into a fresh runs dir → ``(runs_dir, run_dir)``."""
def make(*extra: str) -> tuple[Path, Path]:
runs = tmp_path / "runs"
r = runner.invoke(app, ["make", "a clay pot", "--runs-dir", str(runs), "--no-run", *extra])
@@ -241,7 +241,7 @@ def test_resume_refuses_a_finished_run_and_never_re_enters_it(made_run, stub_tra
def test_a_replay_judges_the_picked_round_on_the_runs_rubric_and_books_into_the_record(tmp_path: Path, monkeypatch):
- """`3dcode judge` replays the run it reads. r0 0.55, r1 unjudged (a judge outage), r2 0.90, of a
+ """`3dcodeverse judge` replays the run it reads. r0 0.55, r1 unjudged (a judge outage), r2 0.90, of a
static `--image` run:
* no --round after `pick --round 0` is r0, the packaged round — not the best-scored r2 (N55a);
* the unjudged r1 replays on the run's reference, not the track's static_object (N49);
diff --git a/harness/tests/flywheel_cli/test_run_layout_cli.py b/harness/tests/flywheel_cli/test_run_layout_cli.py
index 0f39c0f4..7a4876b8 100644
--- a/harness/tests/flywheel_cli/test_run_layout_cli.py
+++ b/harness/tests/flywheel_cli/test_run_layout_cli.py
@@ -1,4 +1,4 @@
-"""deliverable/ (addons/select.package) + telemetry/ (record.package_run), `3dcode show`, and
+"""deliverable/ (addons/select.package) + telemetry/ (record.package_run), `3dcodeverse show`, and
export / gallery on BOTH layouts (old runs must keep working)."""
from __future__ import annotations
diff --git a/harness/tests/flywheel_cli/test_select.py b/harness/tests/flywheel_cli/test_select.py
index 5e9ba2e9..bca272d3 100644
--- a/harness/tests/flywheel_cli/test_select.py
+++ b/harness/tests/flywheel_cli/test_select.py
@@ -1,5 +1,5 @@
"""``codeverse3d.addons.select``: which round of a finished run is handed over, and how
-(``round_rows`` / ``pick`` / ``summarise`` / ``package``, ``3dcode pick``, ``make --no-pick``)."""
+(``round_rows`` / ``pick`` / ``summarise`` / ``package``, ``3dcodeverse pick``, ``make --no-pick``)."""
from __future__ import annotations
@@ -90,7 +90,7 @@ def test_an_unjudged_run_shows_one_fallback_round_in_every_reader(tmp_path, monk
asked: list[int] = []
monkeypatch.setattr(J, "load_round", lambda ws, rec, idx: asked.append(idx))
runner.invoke(app, ["judge", str(ws.root)])
- assert asked == [0] # `3dcode judge`'s default round
+ assert asked == [0] # `3dcodeverse judge`'s default round
class _Pairwise:
diff --git a/harness/tests/gallery/test_cli.py b/harness/tests/gallery/test_cli.py
index ae726b6a..698a2994 100644
--- a/harness/tests/gallery/test_cli.py
+++ b/harness/tests/gallery/test_cli.py
@@ -1,4 +1,4 @@
-"""``3dcode gallery build|serve``."""
+"""``3dcodeverse gallery build|serve``."""
from __future__ import annotations
diff --git a/harness/tests/gallery/test_server.py b/harness/tests/gallery/test_server.py
index 3a653aac..a894003c 100644
--- a/harness/tests/gallery/test_server.py
+++ b/harness/tests/gallery/test_server.py
@@ -25,7 +25,7 @@ def app(gallery_tree: dict[str, Path]) -> GalleryApp:
def test_routes(app: GalleryApp):
r = app.route("/")
assert r.status == 200 and r.content_type.startswith("text/html")
- assert b"3dcode gallery" in r.body
+ assert b"3dcodeverse gallery" in r.body
assert app.route("/nope").status == 404
# the detail of a broken / pending run does not crash
diff --git a/harness/tests/graphics/test_track.py b/harness/tests/graphics/test_track.py
index dadaa0eb..befa9905 100644
--- a/harness/tests/graphics/test_track.py
+++ b/harness/tests/graphics/test_track.py
@@ -44,7 +44,7 @@ def _writer(job, ws):
def test_a_replayed_round_reads_the_payload_the_in_run_judge_read(tmp_path):
- """``3dcode judge`` / calibration rebuild the payload the in-run judge read (``judges.base.round_input``)."""
+ """``3dcodeverse judge`` / calibration rebuild the payload the in-run judge read (``judges.base.round_input``)."""
from codeverse3d.cli._judge import build_judge_input
from codeverse3d.record.record import load_record
diff --git a/harness/tests/install/test_docs.py b/harness/tests/install/test_docs.py
index aa5467da..ae7f9381 100644
--- a/harness/tests/install/test_docs.py
+++ b/harness/tests/install/test_docs.py
@@ -30,7 +30,7 @@ ALLOWED_BINARIES = {
"git", "python", "python3", "pip", "pip3", "pytest", "conda", "uv", "node", "npm", "npx", "nvm",
"sudo", "apt-get", "apt", "brew", "dnf", "pacman",
# this project
- "3dcode", "3dcodeverse", "setup.sh", "blender", "gemini", "claude", "codex", "agy",
+ "3dcodeverse", "setup.sh", "blender", "gemini", "claude", "codex", "agy",
}
_ENV_ASSIGN = re.compile(r"^[A-Za-z_][A-Za-z0-9_]*=")
@@ -98,6 +98,20 @@ def into_absent_eval(path: Path) -> bool:
# --------------------------------------------------------------------------- tests
+def test_the_installed_cli_has_one_name_and_reports_it_in_its_version() -> None:
+ from typer.testing import CliRunner
+
+ from codeverse3d import __version__
+ from codeverse3d.cli.main import app
+
+ with (HARNESS / "pyproject.toml").open("rb") as fh:
+ scripts = tomllib.load(fh)["project"]["scripts"]
+ assert scripts == {"3dcodeverse": "codeverse3d.cli.main:app"}
+ result = CliRunner().invoke(app, ["--version"])
+ assert result.exit_code == 0, result.output
+ assert result.output.strip() == f"3dcodeverse {__version__}"
+
+
@pytest.mark.parametrize("doc", existing_docs(), ids=lambda p: str(p.relative_to(REPO)))
def test_owned_docs_have_valid_commands_links_and_paths(doc: Path) -> None:
bad: list[str] = []
diff --git a/harness/tests/install/test_hermetic.py b/harness/tests/install/test_hermetic.py
deleted file mode 100644
index 08961448..00000000
--- a/harness/tests/install/test_hermetic.py
+++ /dev/null
@@ -1,27 +0,0 @@
-"""Hermetic-suite guard: credential-sensitive offline tests pass without API keys."""
-
-from __future__ import annotations
-
-import os
-import subprocess
-import sys
-from pathlib import Path
-
-HARNESS = Path(__file__).resolve().parents[2]
-KEY_ENVS = ("GEMINI_API_KEYS", "GEMINI_API_KEY", "GOOGLE_API_KEY", "GOOGLE_GENAI_API_KEY")
-
-#: the one that PORT-2 caught; inside the offline (pure python) subset
-CREDENTIAL_FREE = ("tests/models/test_health.py::test_probe_model_treats_a_503_as_final",)
-
-
-def test_passes_with_no_api_keys_in_the_environment(tmp_path: Path) -> None:
- """Credential-sensitive offline tests pass in one key-free subprocess."""
- env = {k: v for k, v in os.environ.items() if k not in KEY_ENVS}
- env["HOME"] = str(tmp_path) # no ~/.config/astra3d/gemini_keys.env either
- env["PYTHONPATH"] = str(HARNESS)
-
- # -n0: the suite's addopts turn xdist ON, and a nested run must not fork 24 more workers
- r = subprocess.run([sys.executable, "-m", "pytest", *CREDENTIAL_FREE, "-q", "-n0", "-p", "no:cacheprovider"],
- cwd=HARNESS, env=env, capture_output=True, text=True, timeout=300)
-
- assert r.returncode == 0, f"needs credentials:\n{r.stdout[-3000:]}\n{r.stderr[-2000:]}"
diff --git a/harness/tests/judges/test_calibration.py b/harness/tests/judges/test_calibration.py
index 3c00d949..9b206301 100644
--- a/harness/tests/judges/test_calibration.py
+++ b/harness/tests/judges/test_calibration.py
@@ -67,9 +67,9 @@ def test_calibrate_offline(tmp_path):
def test_calibration_judges_what_the_in_run_judge_saw(tmp_path, monkeypatch):
- """It rebuilt a thinner JudgeInput than `3dcode judge` (no previous verdict, track
+ """It rebuilt a thinner JudgeInput than `3dcodeverse judge` (no previous verdict, track
context, stored clay views or GLB for the D48 slices), re-rendered clay views a round
- had stored, and always used VlmJudge where `3dcode judge` picks ReferenceJudge."""
+ had stored, and always used VlmJudge where `3dcodeverse judge` picks ReferenceJudge."""
import codeverse3d.addons.calibration as cal
import codeverse3d.judges.vlm_judge as vj
from codeverse3d.contracts.spec import ReferenceImage
diff --git a/harness/tests/models/test_health.py b/harness/tests/models/test_health.py
index b4b20d16..ed6c21d5 100644
--- a/harness/tests/models/test_health.py
+++ b/harness/tests/models/test_health.py
@@ -16,27 +16,34 @@ def test_probe_never_raises_when_every_call_fails(monkeypatch):
assert h.reasons and "provider on fire" in h.reasons[0]
-def test_probe_model_treats_a_503_as_final(monkeypatch):
+def test_probe_model_treats_a_503_as_final(monkeypatch, tmp_path):
""""No retries" includes the storm branch, which max_attempts alone does not bound."""
import codeverse3d.models.gemini as gm
+ from codeverse3d import config
+ from codeverse3d.contracts.chat import ChatMessage, ChatRequest, ChatResponse
from codeverse3d.models.health import _bare_model
+ # Exercise this offline without credentials or a user's config; no nested pytest needed.
+ for name in ("C3D_GEMINI_API_KEYS", "GEMINI_API_KEYS", "GEMINI_API_KEY", "GOOGLE_API_KEY", "GOOGLE_GENAI_API_KEY"):
+ monkeypatch.delenv(name, raising=False)
+ monkeypatch.setattr(config, "_USER_CONFIG", tmp_path / "config.yaml")
+ monkeypatch.setattr(config, "_LEGACY_KEYS_FILE", tmp_path / "gemini_keys.env")
+ monkeypatch.chdir(tmp_path)
+ config.get_settings.cache_clear()
+ assert not config.get_settings().gemini_api_keys
+
captured: dict = {}
def fake_rotate(pool, call, **kw):
captured.update(kw)
- return "unused"
+ return ChatResponse(text="pong")
monkeypatch.setattr(gm, "rotate_with_retries", fake_rotate)
# a non-live test must pass with no credentials (PORT-2)
monkeypatch.setattr(gm, "_default_keys", lambda: ["fake-key-for-tests"])
m = _bare_model("gemini:gemini-3.8-flash", 30.0)
assert m.storm_attempts == 0 and m.max_attempts == 1
- import contextlib
-
- from codeverse3d.contracts.chat import ChatMessage, ChatRequest
-
- with contextlib.suppress(Exception): # the fake returns a non-response; only the kwargs matter
- m.generate(ChatRequest(messages=[ChatMessage.user("pong")], max_output_tokens=8))
+ response = m.generate(ChatRequest(messages=[ChatMessage.user("pong")], max_output_tokens=8))
+ assert response.text == "pong"
assert captured.get("storm_attempts") == 0, captured.keys()
assert captured.get("max_attempts") == 1
diff --git a/harness/tests/orchestrator_tracks/test_fixed_rounds.py b/harness/tests/orchestrator_tracks/test_fixed_rounds.py
index dde431b7..4035f7a4 100644
--- a/harness/tests/orchestrator_tracks/test_fixed_rounds.py
+++ b/harness/tests/orchestrator_tracks/test_fixed_rounds.py
@@ -4,8 +4,6 @@ each built on the round before it, and only the clock or a hard failure ends it
from __future__ import annotations
from codeverse3d.contracts.common import Language
-from codeverse3d.contracts.run import RunStatus
-from codeverse3d.proc import EventLog
from codeverse3d.tracks.static_object import StaticObjectTrack
from codeverse3d.workspace import Workspace
from tests.orchestrator_tracks.conftest import make_spec
@@ -25,20 +23,6 @@ def _track(chair_plan, settings, scores, *, agent=None) -> StaticObjectTrack:
runtime=FakeRuntime(Language.CADQUERY))
-def test_a_run_is_the_baseline_plus_max_rounds_whatever_the_judge_says(tmp_path, chair_plan, settings):
- """A run that passes at r01 (once a "pass" stop) still runs every round; the regressing twin
- is ``test_every_refine_starts_from_the_previous_rounds_commit`` and the D84 loop test."""
- scores, max_rounds = (0.55, 0.85, 0.95), 3
- ws = Workspace(tmp_path / "runs" / "fixed")
- rec = _track(chair_plan, settings, scores).run(make_spec(language=Language.CADQUERY, max_rounds=max_rounds), ws)
- assert len(rec.rounds) == max_rounds + 1
- assert [r.kind for r in rec.rounds] == ["baseline"] + ["refine"] * max_rounds
- assert rec.status is RunStatus.MAX_ROUNDS and rec.extra["stop_reason"] == "max_rounds"
- assert all(r.judgment is not None for r in rec.rounds), "the judge still scores every round"
- stops = [e for e in EventLog(ws.events_path).read() if e["event"] == "stop"]
- assert [e["reason"] for e in stops] == ["max_rounds"]
-
-
def test_every_refine_starts_from_the_previous_rounds_commit(tmp_path, chair_plan, settings):
seen: dict[int, str] = {}
diff --git a/harness/tests/orchestrator_tracks/test_refine_groups_and_judge_views.py b/harness/tests/orchestrator_tracks/test_judge_inputs.py
similarity index 51%
rename from harness/tests/orchestrator_tracks/test_refine_groups_and_judge_views.py
rename to harness/tests/orchestrator_tracks/test_judge_inputs.py
index 9dbb2b1d..fb74a00c 100644
--- a/harness/tests/orchestrator_tracks/test_refine_groups_and_judge_views.py
+++ b/harness/tests/orchestrator_tracks/test_judge_inputs.py
@@ -1,36 +1,15 @@
-"""Refine grouping and judge-view flags."""
+"""The track's judge inputs: selected views, artifact paths and plan coordinates."""
from __future__ import annotations
from codeverse3d.contracts.artifacts import RenderSet, RenderView
-from codeverse3d.orchestrator import RefineTask, plan_refine_groups
+from codeverse3d.contracts.common import Language
+from codeverse3d.contracts.plan import BBox, PartPlan, StaticPlan
from codeverse3d.tracks.scene import ScenePipeline
from codeverse3d.workspace import Workspace
-from tests.orchestrator_tracks.fakes import (
- FakeServices,
-)
+from tests.orchestrator_tracks.fakes import FakeServices
-# --------------------------------------------------------------------- plan_refine_groups
-def _task(target: str, files: list[str] | None = None, priority: int = 2) -> RefineTask:
- return RefineTask(target=target, kind="geometry", instruction=f"fix {target}", priority=priority, files=files or [])
-
-
-def test_plan_refine_groups_fans_out_only_when_allowed_and_disjoint():
- tasks = [_task("Seat", ["src/parts/seat.js"]), _task("Leg", ["src/parts/leg.js"])]
- groups, parallel = plan_refine_groups(tasks, allow_fanout=True, parallel_min_tasks=2)
- assert parallel and len(groups) == 2
-
- groups, parallel = plan_refine_groups(tasks, allow_fanout=False, parallel_min_tasks=2)
- assert not parallel and len(groups) == 1
- assert groups[0].files == ["src/parts/leg.js", "src/parts/seat.js"] and len(groups[0].tasks) == 2
-
- # unknown file ownership collapses into one group even when fan-out is allowed
- groups, parallel = plan_refine_groups([_task("Seat", ["src/parts/seat.js"]), _task("overall")], allow_fanout=True)
- assert not parallel and len(groups) == 1
-
-
-# --------------------------------------------------------------------- judge-view flags (F30 consumption)
def _rs(flags: list[bool | None]) -> RenderSet:
views = [RenderView(name=f"v{i}", path=f"/tmp/v{i}.png", judge=f) for i, f in enumerate(flags)]
return RenderSet(views=views, renderer="fake")
@@ -58,3 +37,16 @@ def test_judge_view_flags_and_path_reconstruction(tmp_path):
fixed = resolve_paths(ws, rel)
assert fixed.out_dir == str(ws.root / "artifacts/renders/r00")
assert fixed.views[0].path == str(ws.root / "artifacts/renders/r00/a.png")
+
+
+def test_plan_digest_is_labelled_w_h_d_in_the_measurement_frame():
+ from codeverse3d.judges.base import plan_summary
+
+ plan = StaticPlan(object_name="Cabinet", summary="A cabinet.",
+ overall_bbox=BBox(center=(0, 0, 0.4), extents=(0.48, 0.52, 0.80)), # Z-up plan: W×D×H
+ parts=[PartPlan(name="Body", role="body", description="box",
+ bbox=BBox(center=(0, 0, 0.4), extents=(0.48, 0.52, 0.80)))])
+ zup = plan_summary(plan, Language.BLENDER)
+ assert "Overall 0.48×0.80×0.52 m (W×H×D)" in zup # H and D swapped into the GLB/measurement order
+ yup = plan_summary(plan, Language.THREEJS)
+ assert "Overall 0.48×0.52×0.80 m (W×H×D)" in yup # identity for Y-up plans
diff --git a/harness/tests/orchestrator_tracks/test_lifecycle_regressions.py b/harness/tests/orchestrator_tracks/test_lifecycle_regressions.py
index 93479c7e..c9b92253 100644
--- a/harness/tests/orchestrator_tracks/test_lifecycle_regressions.py
+++ b/harness/tests/orchestrator_tracks/test_lifecycle_regressions.py
@@ -1,4 +1,4 @@
-"""Lifecycle regressions: aborted rounds, degraded verdicts, the skeleton guard, finalise rebuilds."""
+"""Lifecycle recovery: aborted rounds, degraded verdicts, resume and finalise rebuilds."""
from __future__ import annotations
@@ -6,14 +6,11 @@ import json
import pytest
-from codeverse3d.contracts.artifacts import BuildResult, GateReport, Judgment
-from codeverse3d.contracts.common import Language, Track
-from codeverse3d.contracts.plan import BBox, PartPlan, ScenePlan, StaticPlan
+from codeverse3d.contracts.artifacts import BuildResult, Judgment
+from codeverse3d.contracts.common import Language
from codeverse3d.contracts.run import RoundRecord, RunStatus
-from codeverse3d.orchestrator import RoundPolicy, RunState
+from codeverse3d.orchestrator import RoundPolicy
from codeverse3d.proc import EventLog
-from codeverse3d.tracks.planner import plan_example
-from codeverse3d.tracks.scene import SceneTrack
from codeverse3d.tracks.static_object import StaticObjectTrack
from codeverse3d.workspace import Workspace
from tests.orchestrator_tracks.conftest import fake_clock, make_spec
@@ -187,43 +184,6 @@ def test_skeleton_never_reruns_over_existing_rounds(tmp_path, chair_plan, settin
assert "stage.skipped" in kinds
-# --------------------------------------------------------------------- finding: scene track rebuild path
-def test_scene_refine_emits_rebuild_task_when_build_failed(tmp_path, settings):
- plan = ScenePlan.model_validate(plan_example(Track.SCENE))
- spec = make_spec(Track.SCENE, Language.SCENE_THREEJS)
- ws = Workspace(tmp_path / "runs" / "r")
- ws.create()
- track = SceneTrack(services=FakeServices(), settings=settings, runtime=FakeRuntime(Language.SCENE_THREEJS))
- ctx = track.build_context(spec, ws, EventLog(ws.events_path), RunState())
- ctx.plan = plan
- (ws.src / "zones").mkdir(parents=True, exist_ok=True)
- (ws.src / "zones" / "quay.js").write_text("// broken\n")
- last = RoundRecord(index=0, kind="baseline",
- build=BuildResult(ok=False, language="scene_threejs", error_type="ProbeError",
- error_message="Cannot read properties of undefined (reading 'update')",
- error_file="src/zones/quay.js"),
- gates=[GateReport(gate="lint:scene_threejs", passed=True)])
- tasks, instructions = track.refine_tasks(ctx, last)
- assert len(tasks) == 1 and tasks[0].kind == "rebuild"
- assert tasks[0].files_hint == ["src/zones/quay.js"]
- assert "ProbeError" in tasks[0].prompt and "Cannot read properties" in tasks[0].prompt
- assert instructions and "rebuild" in instructions[0]
-
-
-# --------------------------------------------------------------------- finding: plan digest frame order
-def test_plan_digest_is_labelled_w_h_d_in_the_measurement_frame():
- from codeverse3d.judges.base import plan_summary
-
- plan = StaticPlan(object_name="Cabinet", summary="A cabinet.",
- overall_bbox=BBox(center=(0, 0, 0.4), extents=(0.48, 0.52, 0.80)), # Z-up plan: W×D×H
- parts=[PartPlan(name="Body", role="body", description="box",
- bbox=BBox(center=(0, 0, 0.4), extents=(0.48, 0.52, 0.80)))])
- zup = plan_summary(plan, Language.BLENDER)
- assert "Overall 0.48×0.80×0.52 m (W×H×D)" in zup # H and D swapped into the GLB/measurement order
- yup = plan_summary(plan, Language.THREEJS)
- assert "Overall 0.48×0.52×0.80 m (W×H×D)" in yup # identity for Y-up plans
-
-
def test_a_successful_re_finalise_clears_the_stale_rebuild_failed_flag(tmp_path, chair_plan, settings, monkeypatch):
"""A failed finalise rebuild is flagged; a later resume that rebuilds successfully clears the flag."""
spec = make_spec(max_rounds=3)
diff --git a/harness/tests/orchestrator_tracks/test_likeness_routing.py b/harness/tests/orchestrator_tracks/test_likeness_routing.py
index 7f05c077..dee6e006 100644
--- a/harness/tests/orchestrator_tracks/test_likeness_routing.py
+++ b/harness/tests/orchestrator_tracks/test_likeness_routing.py
@@ -41,7 +41,7 @@ def test_make_judge_routes_photos_to_the_likeness_judge_and_a_replay_picks_the_s
assert BaseTrack.make_judge(me, _ctx(Track.STATIC_OBJECT, ["/x.png"])) == "R"
assert BaseTrack.make_judge(me, _ctx(Track.GRAPHICS, [])) == "P"
assert calls[0] == ("likeness", "shader")
- # a replay picks the judge the live run did: `3dcode judge` / calibration and the live run share
+ # a replay picks the judge the live run did: `3dcodeverse judge` / calibration and the live run share
# `judge_for` — a graphics or scene run with photos was re-judged by ReferenceJudge (silhouette +
# diff pass) where the run used likeness
from codeverse3d.cli._judge import make_judge
diff --git a/harness/tests/orchestrator_tracks/test_orchestrator.py b/harness/tests/orchestrator_tracks/test_orchestrator.py
index 6945f390..19e41fc3 100644
--- a/harness/tests/orchestrator_tracks/test_orchestrator.py
+++ b/harness/tests/orchestrator_tracks/test_orchestrator.py
@@ -20,6 +20,7 @@ from codeverse3d.orchestrator import (
StageRunner,
build_refine_instructions,
plan_parallel_groups,
+ plan_refine_groups,
)
from codeverse3d.proc import EventLog, fan_out
@@ -126,3 +127,21 @@ def test_two_env_errors_of_one_gate_are_two_refine_tasks(chair_plan):
tasks = build_refine_instructions(None, [GateReport(gate="scene_placement", passed=False, findings=env)], [], chair_plan)
lines = [t.instruction for t in tasks]
assert len(tasks) == 2 and any("fog" in x for x in lines) and any("background" in x for x in lines), lines
+
+
+def _task(target: str, files: list[str] | None = None, priority: int = 2) -> RefineTask:
+ return RefineTask(target=target, kind="geometry", instruction=f"fix {target}", priority=priority, files=files or [])
+
+
+def test_plan_refine_groups_fans_out_only_when_allowed_and_disjoint():
+ tasks = [_task("Seat", ["src/parts/seat.js"]), _task("Leg", ["src/parts/leg.js"])]
+ groups, parallel = plan_refine_groups(tasks, allow_fanout=True, parallel_min_tasks=2)
+ assert parallel and len(groups) == 2
+
+ groups, parallel = plan_refine_groups(tasks, allow_fanout=False, parallel_min_tasks=2)
+ assert not parallel and len(groups) == 1
+ assert groups[0].files == ["src/parts/leg.js", "src/parts/seat.js"] and len(groups[0].tasks) == 2
+
+ # unknown file ownership collapses into one group even when fan-out is allowed
+ groups, parallel = plan_refine_groups([_task("Seat", ["src/parts/seat.js"]), _task("overall")], allow_fanout=True)
+ assert not parallel and len(groups) == 1
diff --git a/harness/tests/orchestrator_tracks/test_refine_evidence.py b/harness/tests/orchestrator_tracks/test_refine_evidence.py
index cf3de7b4..93ece677 100644
--- a/harness/tests/orchestrator_tracks/test_refine_evidence.py
+++ b/harness/tests/orchestrator_tracks/test_refine_evidence.py
@@ -1,10 +1,19 @@
-"""D112 (iteration 5, r1_07): the refine prompt carries the placement rows itself; the session had burned 15 minutes paging scene_probe.json."""
+"""Scene refinement carries gate evidence and rebuilds failed source files."""
from __future__ import annotations
+from codeverse3d.contracts.artifacts import BuildResult, GateReport
+from codeverse3d.contracts.common import Language, Track
+from codeverse3d.contracts.plan import ScenePlan
from codeverse3d.contracts.run import RoundRecord
-from codeverse3d.orchestrator import TaskGroup, build_refine_instructions
+from codeverse3d.orchestrator import RunState, TaskGroup, build_refine_instructions
+from codeverse3d.proc import EventLog
from codeverse3d.spatial.scene_placement import placement_findings
+from codeverse3d.tracks.planner import plan_example
+from codeverse3d.tracks.scene import SceneTrack
+from codeverse3d.workspace import Workspace
+from tests.orchestrator_tracks.conftest import make_spec
+from tests.orchestrator_tracks.fakes import FakeRuntime, FakeServices
from tests.orchestrator_tracks.test_env_stage_honest import LANGUAGES, _ctx
from tests.scene_gates.test_placement import _row, _table
@@ -21,3 +30,25 @@ def test_the_refine_prompt_states_each_failed_row_its_gap_and_tolerance(tmp_path
assert f"{name} is floating {gap} m above Ground (touches nothing; indoor tolerance 0.02 m" in prompt, name
assert "`artifacts/scene_probe.json` is the probe driver's raw dump" in prompt
assert "`check_placement` returns the placement rows" in prompt
+
+
+def test_scene_refine_emits_rebuild_task_when_build_failed(tmp_path, settings):
+ plan = ScenePlan.model_validate(plan_example(Track.SCENE))
+ spec = make_spec(Track.SCENE, Language.SCENE_THREEJS)
+ ws = Workspace(tmp_path / "runs" / "r")
+ ws.create()
+ track = SceneTrack(services=FakeServices(), settings=settings, runtime=FakeRuntime(Language.SCENE_THREEJS))
+ ctx = track.build_context(spec, ws, EventLog(ws.events_path), RunState())
+ ctx.plan = plan
+ (ws.src / "zones").mkdir(parents=True, exist_ok=True)
+ (ws.src / "zones" / "quay.js").write_text("// broken\n")
+ last = RoundRecord(index=0, kind="baseline",
+ build=BuildResult(ok=False, language="scene_threejs", error_type="ProbeError",
+ error_message="Cannot read properties of undefined (reading 'update')",
+ error_file="src/zones/quay.js"),
+ gates=[GateReport(gate="lint:scene_threejs", passed=True)])
+ tasks, instructions = track.refine_tasks(ctx, last)
+ assert len(tasks) == 1 and tasks[0].kind == "rebuild"
+ assert tasks[0].files_hint == ["src/zones/quay.js"]
+ assert "ProbeError" in tasks[0].prompt and "Cannot read properties" in tasks[0].prompt
+ assert instructions and "rebuild" in instructions[0]
diff --git a/harness/tests/orchestrator_tracks/test_tracks.py b/harness/tests/orchestrator_tracks/test_tracks.py
index c8752ba6..831f33e5 100644
--- a/harness/tests/orchestrator_tracks/test_tracks.py
+++ b/harness/tests/orchestrator_tracks/test_tracks.py
@@ -47,6 +47,10 @@ def test_static_track_end_to_end_agent_path(tmp_path, chair_plan, settings):
settings=settings, runtime=FakeRuntime(Language.THREEJS))
rec = track.run(make_spec(max_rounds=3), ws)
assert rec.status is RunStatus.MAX_ROUNDS and [r.score for r in rec.rounds] == [0.55, 0.7, 0.85, 0.85]
+ assert [r.kind for r in rec.rounds] == ["baseline", "refine", "refine", "refine"]
+ assert rec.extra["stop_reason"] == "max_rounds"
+ stops = [e for e in EventLog(ws.events_path).read() if e["event"] == "stop"]
+ assert [e["reason"] for e in stops] == ["max_rounds"]
assert rec.rounds[1].instructions[0].startswith("[gate/gate:contract] Seat") # gate findings lead the refine
# Law 6: one commit per round, and the workspace ends at the last one
assert len({r.commit for r in rec.rounds}) == 4 and ws.head() == rec.rounds[3].commit
diff --git a/harness/tests/prompts/manifest.json b/harness/tests/prompts/manifest.json
index 38a68e7f..ce43cd9a 100644
--- a/harness/tests/prompts/manifest.json
+++ b/harness/tests/prompts/manifest.json
@@ -11,7 +11,7 @@
"chat:planner:planner/messages": {"bytes": 1328, "sha256": "313c87553cf38df423dc9245a8af4cca5a41a844b489e13a5dc08e8c44eabba5"},
"chat:planner:planner/sampling": {"bytes": 10001, "sha256": "0ff37a3ae43fdf878dd0911a9954d57e49365ad788c84818ea01d350310ab401"},
"chat:planner:planner/system": {"bytes": 20030, "sha256": "45dccfcf72b4c6e46b7ceeaae8af0ebddf6f6e6b5970cda5fe900aa47682bf06"},
- "context:runs/run/files": {"bytes": 45730, "sha256": "e317d817f6eeae64383e6e85433580ffaea0c7e50ab14033e5a3e4cbf0890be8"},
+ "context:runs/run/files": {"bytes": 45740, "sha256": "c19f00e9a22215fc6d155e4640aba039343e5d2fe9981a2fdc7953e9cad8b80a"},
"job:baseline@r00/prompt": {"bytes": 19680, "sha256": "60580670e903b8d76bdad9f76ffb88dcdd42b923a30348858cd5724dd4eb2e2d"},
"job:baseline@r00/scope": {"bytes": 383, "sha256": "7187fd726a9f499668ae3df0399ed4a3e08b7cb7a9de5ba753b8775448ea876e"},
"job:baseline@r00/system": {"bytes": 3271, "sha256": "29f5a9b2e7b9a879be58963ce95ae6fbec7460dae6785d542b214d6b850e21f4"},
@@ -72,9 +72,9 @@
"chat:planner:planner/messages": {"bytes": 348, "sha256": "f50fc3cdf3a426fc9c9b830b6e68d2cfe9fd9b7125550362b018f9c9b454b25c"},
"chat:planner:planner/sampling": {"bytes": 3421, "sha256": "918170272cea656e8c542c54b777c57d1a4567a8cc36a0c1cd825cb81a01b9bb"},
"chat:planner:planner/system": {"bytes": 11919, "sha256": "194d0e2a724c4799816033228af15cac568dd51c14ed85bff9c94ae5ba281866"},
- "context:runs/run/_cand/c0/files": {"bytes": 32350, "sha256": "1e653c3d53c920e53ec7142071b1a187518c5f3a6a679321f38e29ad71a82d5d"},
- "context:runs/run/_cand/c1/files": {"bytes": 32350, "sha256": "1e653c3d53c920e53ec7142071b1a187518c5f3a6a679321f38e29ad71a82d5d"},
- "context:runs/run/files": {"bytes": 32350, "sha256": "1e653c3d53c920e53ec7142071b1a187518c5f3a6a679321f38e29ad71a82d5d"},
+ "context:runs/run/_cand/c0/files": {"bytes": 32360, "sha256": "7ece7fd3e4274f3c5749772aaf0774de530f1f0a2d50b48aaf998dc6576a067c"},
+ "context:runs/run/_cand/c1/files": {"bytes": 32360, "sha256": "7ece7fd3e4274f3c5749772aaf0774de530f1f0a2d50b48aaf998dc6576a067c"},
+ "context:runs/run/files": {"bytes": 32360, "sha256": "7ece7fd3e4274f3c5749772aaf0774de530f1f0a2d50b48aaf998dc6576a067c"},
"job:baseline_c0@r00/prompt": {"bytes": 11347, "sha256": "5aca38cfae577e7e5e5974c2b6cd64e55ec35a1dfab33eaf62eff70652854df5"},
"job:baseline_c0@r00/scope": {"bytes": 383, "sha256": "51d22961200ef987a7c652af48ff2897b910803baa672d3081f53c3fb819a685"},
"job:baseline_c0@r00/system": {"bytes": 3173, "sha256": "c3879a563e2a512b82358d0c3927dcbfe9347725e592035a9b64c5d1b34e6654"},
@@ -107,7 +107,7 @@
"chat:planner:planner/messages": {"bytes": 348, "sha256": "f50fc3cdf3a426fc9c9b830b6e68d2cfe9fd9b7125550362b018f9c9b454b25c"},
"chat:planner:planner/sampling": {"bytes": 3421, "sha256": "918170272cea656e8c542c54b777c57d1a4567a8cc36a0c1cd825cb81a01b9bb"},
"chat:planner:planner/system": {"bytes": 11919, "sha256": "194d0e2a724c4799816033228af15cac568dd51c14ed85bff9c94ae5ba281866"},
- "context:runs/run/files": {"bytes": 32350, "sha256": "1e653c3d53c920e53ec7142071b1a187518c5f3a6a679321f38e29ad71a82d5d"},
+ "context:runs/run/files": {"bytes": 32360, "sha256": "7ece7fd3e4274f3c5749772aaf0774de530f1f0a2d50b48aaf998dc6576a067c"},
"job:baseline@r00/prompt": {"bytes": 11347, "sha256": "5aca38cfae577e7e5e5974c2b6cd64e55ec35a1dfab33eaf62eff70652854df5"},
"job:baseline@r00/scope": {"bytes": 379, "sha256": "831e9c99bdd04789b89bfddb89c97939a813323c9173d742adabbb5f1f8a7dc1"},
"job:baseline@r00/system": {"bytes": 3173, "sha256": "c3879a563e2a512b82358d0c3927dcbfe9347725e592035a9b64c5d1b34e6654"},
@@ -143,7 +143,7 @@
"chat:planner:planner/messages": {"bytes": 348, "sha256": "f50fc3cdf3a426fc9c9b830b6e68d2cfe9fd9b7125550362b018f9c9b454b25c"},
"chat:planner:planner/sampling": {"bytes": 3421, "sha256": "918170272cea656e8c542c54b777c57d1a4567a8cc36a0c1cd825cb81a01b9bb"},
"chat:planner:planner/system": {"bytes": 11919, "sha256": "194d0e2a724c4799816033228af15cac568dd51c14ed85bff9c94ae5ba281866"},
- "context:runs/run/files": {"bytes": 32350, "sha256": "1e653c3d53c920e53ec7142071b1a187518c5f3a6a679321f38e29ad71a82d5d"},
+ "context:runs/run/files": {"bytes": 32360, "sha256": "7ece7fd3e4274f3c5749772aaf0774de530f1f0a2d50b48aaf998dc6576a067c"},
"job:baseline@r00/prompt": {"bytes": 11347, "sha256": "5aca38cfae577e7e5e5974c2b6cd64e55ec35a1dfab33eaf62eff70652854df5"},
"job:baseline@r00/scope": {"bytes": 379, "sha256": "831e9c99bdd04789b89bfddb89c97939a813323c9173d742adabbb5f1f8a7dc1"},
"job:baseline@r00/system": {"bytes": 3173, "sha256": "c3879a563e2a512b82358d0c3927dcbfe9347725e592035a9b64c5d1b34e6654"},
@@ -200,7 +200,7 @@
"chat:planner:planner/messages": {"bytes": 348, "sha256": "f50fc3cdf3a426fc9c9b830b6e68d2cfe9fd9b7125550362b018f9c9b454b25c"},
"chat:planner:planner/sampling": {"bytes": 3421, "sha256": "918170272cea656e8c542c54b777c57d1a4567a8cc36a0c1cd825cb81a01b9bb"},
"chat:planner:planner/system": {"bytes": 14279, "sha256": "b9199fdf3a0bc64a88587f4b8c8efe1443bc20164e1b9e5eada29c7762adff2f"},
- "context:runs/run/files": {"bytes": 37232, "sha256": "3201918ccba99763efa1cc13a9a5bd9f9f2f69121ee61b8484c97666bcd36b6d"},
+ "context:runs/run/files": {"bytes": 37242, "sha256": "d87cbc72fb19833f5ddeb69af936e4389fda7554566c6f8a6b437ba6b788d58c"},
"job:baseline@r00/prompt": {"bytes": 13687, "sha256": "ac1b0fc66ea413b427ddc1e90d80bf967048ec80216aeaef9c4c104c1df5ada3"},
"job:baseline@r00/scope": {"bytes": 358, "sha256": "b0e9a906ab8df3ff56acc08198289aa3958c470e5fd16a4e0225178a2d88e715"},
"job:baseline@r00/system": {"bytes": 2729, "sha256": "d12d4891ce72156aa561bdfa0ed1fb20c5826412024b0be07b423949528e5df6"},
@@ -259,7 +259,7 @@
"chat:planner:zone-layout#2/messages": {"bytes": 485, "sha256": "67e917190ee227c5ed3415ab4b03c072af313d7040704e7de671e63c42e2ab52"},
"chat:planner:zone-layout#2/sampling": {"bytes": 3397, "sha256": "160445a83ede28afc0191547fd42d9341e694ef70aaad380bebf60ead0a65c49"},
"chat:planner:zone-layout#2/system": {"bytes": 3326, "sha256": "494acec302dd23fa2fc6ee1eb6d48e873d6cf0ad98bad413e27578cf062313d4"},
- "context:runs/run/files": {"bytes": 60438, "sha256": "45f8c3c7e2df56537869918b6cf091c1985d102bdd660bb126bbfb0a8c9f7472"},
+ "context:runs/run/files": {"bytes": 60448, "sha256": "5b958ecd92a1fa44ec9dec8e0514b9996a1b43a816029ae352d9b586f9fd8295"},
"job:asset_bollard@r00/prompt": {"bytes": 13922, "sha256": "180d385bb533dedc37b3d0af79ceb3e07cf0a08e9600d7fadf17f47e8804608c"},
"job:asset_bollard@r00/scope": {"bytes": 342, "sha256": "d7d32c17ba05779e24c5336cf50025dfae81e0cf5bfa6be0476b61904b0ad8f8"},
"job:asset_bollard@r00/system": {"bytes": 2685, "sha256": "48a4603263c98cfaa5b3bb557c785530d917681f68f727d010cc51da13656b11"},
@@ -389,7 +389,7 @@
"chat:planner:zone-layout#2/messages": {"bytes": 485, "sha256": "67e917190ee227c5ed3415ab4b03c072af313d7040704e7de671e63c42e2ab52"},
"chat:planner:zone-layout#2/sampling": {"bytes": 3397, "sha256": "160445a83ede28afc0191547fd42d9341e694ef70aaad380bebf60ead0a65c49"},
"chat:planner:zone-layout#2/system": {"bytes": 3326, "sha256": "494acec302dd23fa2fc6ee1eb6d48e873d6cf0ad98bad413e27578cf062313d4"},
- "context:runs/run/files": {"bytes": 55616, "sha256": "e616e7e5b285d8891c6e98a0d45af7d9aad5eab1aae2e5c220c63b508b31c2dd"},
+ "context:runs/run/files": {"bytes": 55626, "sha256": "210ef1318159cf47ac914cf0013d623c34ee57bc7723ca301da1e1a3881b5b86"},
"job:asset_bollard@r00/prompt": {"bytes": 5182, "sha256": "eb2371d5d91daf0664ea4155eed95618b5798277e9b8e5cd118e48181aa499a3"},
"job:asset_bollard@r00/scope": {"bytes": 357, "sha256": "450d659126efa73cdfe3371f18464e6ab40230af78367de6994d4712c9a8aaaf"},
"job:asset_bollard@r00/system": {"bytes": 997, "sha256": "20d72b3bb074ab09ce4b2b6656b8bd0fcffef4fd3d437f2c171cb3efd96d475c"},
@@ -475,7 +475,7 @@
"chat:planner:zone-layout#2/sampling": {"bytes": 3397, "sha256": "160445a83ede28afc0191547fd42d9341e694ef70aaad380bebf60ead0a65c49"},
"chat:planner:zone-layout#2/system": {"bytes": 3361, "sha256": "9acfcedb047a4f1623c483cb99c1f8aa30fd1158e56f88d1c1518f138db1b34f"},
"context:runs/run/_assets/crate/files": {"bytes": 32742, "sha256": "fee09c023ada7ce7895e070195596c0b5103ecc5dabf1cd86dbc5352775c90b2"},
- "context:runs/run/files": {"bytes": 55616, "sha256": "e616e7e5b285d8891c6e98a0d45af7d9aad5eab1aae2e5c220c63b508b31c2dd"},
+ "context:runs/run/files": {"bytes": 55626, "sha256": "210ef1318159cf47ac914cf0013d623c34ee57bc7723ca301da1e1a3881b5b86"},
"job:asset_crate@r00/prompt": {"bytes": 18032, "sha256": "27dfd5f4bedcc036f93c844230347401da2c4c4f8b1d300bfaece147b704c3fc"},
"job:asset_crate@r00/scope": {"bytes": 430, "sha256": "3c4b45fd7cc847cc6a6a5f3b83784cddab99440be445d61f924f56b834cd1476"},
"job:asset_crate@r00/system": {"bytes": 3716, "sha256": "6068eff7b8c36b4edcc5d64ead79bd3de7386e0bbaf4fb141210883fb34f8518"},
@@ -522,7 +522,7 @@
"chat:planner:planner/messages": {"bytes": 204, "sha256": "b91c27cbf7a8071bf320a68a6113bd25896bf0422f92b9087fb155bad18dbdc1"},
"chat:planner:planner/sampling": {"bytes": 8047, "sha256": "ac5eba7173d30fcdfc0682ff102fc65e9f4325a5a4f1fbdaca4b0dcd4547d4b1"},
"chat:planner:planner/system": {"bytes": 26641, "sha256": "97cc784b73ca9dfb4f46cc62c0db7c73d9cb6085db85b7feba1895d98e0fa6f6"},
- "context:runs/run/files": {"bytes": 55616, "sha256": "e616e7e5b285d8891c6e98a0d45af7d9aad5eab1aae2e5c220c63b508b31c2dd"},
+ "context:runs/run/files": {"bytes": 55626, "sha256": "210ef1318159cf47ac914cf0013d623c34ee57bc7723ca301da1e1a3881b5b86"},
"job:env@r00/prompt": {"bytes": 36302, "sha256": "f12e0b146910ce6a10b44229fb8b080fc72943831b04231f0522f6fdfb3892b2"},
"job:env@r00/scope": {"bytes": 334, "sha256": "b94c09720c9e481759ea0408d7ef708c4d5a3e47b1c2b28a8d901a7c77d7c66b"},
"job:env@r00/system": {"bytes": 5625, "sha256": "3e4c806c832c2fb4ab9dd3b1d5063c611b07258ce2fa6d9b3f25df5a467734c7"},
@@ -566,7 +566,7 @@
"chat:planner:zone-layout#2/messages": {"bytes": 485, "sha256": "67e917190ee227c5ed3415ab4b03c072af313d7040704e7de671e63c42e2ab52"},
"chat:planner:zone-layout#2/sampling": {"bytes": 3397, "sha256": "160445a83ede28afc0191547fd42d9341e694ef70aaad380bebf60ead0a65c49"},
"chat:planner:zone-layout#2/system": {"bytes": 3326, "sha256": "494acec302dd23fa2fc6ee1eb6d48e873d6cf0ad98bad413e27578cf062313d4"},
- "context:runs/run/files": {"bytes": 55616, "sha256": "e616e7e5b285d8891c6e98a0d45af7d9aad5eab1aae2e5c220c63b508b31c2dd"},
+ "context:runs/run/files": {"bytes": 55626, "sha256": "210ef1318159cf47ac914cf0013d623c34ee57bc7723ca301da1e1a3881b5b86"},
"job:env@r00/prompt": {"bytes": 36302, "sha256": "f12e0b146910ce6a10b44229fb8b080fc72943831b04231f0522f6fdfb3892b2"},
"job:env@r00/scope": {"bytes": 334, "sha256": "b94c09720c9e481759ea0408d7ef708c4d5a3e47b1c2b28a8d901a7c77d7c66b"},
"job:env@r00/system": {"bytes": 5625, "sha256": "3e4c806c832c2fb4ab9dd3b1d5063c611b07258ce2fa6d9b3f25df5a467734c7"},
@@ -623,7 +623,7 @@
"chat:planner:zone-layout#2/messages": {"bytes": 485, "sha256": "67e917190ee227c5ed3415ab4b03c072af313d7040704e7de671e63c42e2ab52"},
"chat:planner:zone-layout#2/sampling": {"bytes": 3397, "sha256": "160445a83ede28afc0191547fd42d9341e694ef70aaad380bebf60ead0a65c49"},
"chat:planner:zone-layout#2/system": {"bytes": 3326, "sha256": "494acec302dd23fa2fc6ee1eb6d48e873d6cf0ad98bad413e27578cf062313d4"},
- "context:runs/run/files": {"bytes": 55616, "sha256": "e616e7e5b285d8891c6e98a0d45af7d9aad5eab1aae2e5c220c63b508b31c2dd"},
+ "context:runs/run/files": {"bytes": 55626, "sha256": "210ef1318159cf47ac914cf0013d623c34ee57bc7723ca301da1e1a3881b5b86"},
"job:env@r00/prompt": {"bytes": 36302, "sha256": "f12e0b146910ce6a10b44229fb8b080fc72943831b04231f0522f6fdfb3892b2"},
"job:env@r00/scope": {"bytes": 334, "sha256": "b94c09720c9e481759ea0408d7ef708c4d5a3e47b1c2b28a8d901a7c77d7c66b"},
"job:env@r00/system": {"bytes": 5625, "sha256": "3e4c806c832c2fb4ab9dd3b1d5063c611b07258ce2fa6d9b3f25df5a467734c7"},
@@ -724,9 +724,9 @@
"chat:planner:planner/messages": {"bytes": 1354, "sha256": "5b1a589e27562a3668649aed4be4857a34804a2acdc140eb8fb3630d970f71a6"},
"chat:planner:planner/sampling": {"bytes": 6272, "sha256": "04dbfbbd42d8b93a12571a1908beacc7eb471e6f2c0fc077371aa86e2a99fb40"},
"chat:planner:planner/system": {"bytes": 20977, "sha256": "0a3c2ea2bfb39981a166952054d1dcc10dffdf097d493e89b03e0b8586d2ee8e"},
- "context:runs/run/_cand/c0/files": {"bytes": 47082, "sha256": "431132a1be77fd78faf44676066d73c10c7dcdcc024fe50a177a579427482987"},
- "context:runs/run/_cand/c1/files": {"bytes": 47082, "sha256": "431132a1be77fd78faf44676066d73c10c7dcdcc024fe50a177a579427482987"},
- "context:runs/run/files": {"bytes": 47082, "sha256": "431132a1be77fd78faf44676066d73c10c7dcdcc024fe50a177a579427482987"},
+ "context:runs/run/_cand/c0/files": {"bytes": 47092, "sha256": "abca5605a55de083655700701faed3bbfd7ba04281b47344a8394c0042f39dc1"},
+ "context:runs/run/_cand/c1/files": {"bytes": 47092, "sha256": "abca5605a55de083655700701faed3bbfd7ba04281b47344a8394c0042f39dc1"},
+ "context:runs/run/files": {"bytes": 47092, "sha256": "abca5605a55de083655700701faed3bbfd7ba04281b47344a8394c0042f39dc1"},
"job:baseline_c0@r00/prompt": {"bytes": 20387, "sha256": "158c7483180b37e34a73f10bf8cabd1f708d89920c60b791f3815bdf4703ef8d"},
"job:baseline_c0@r00/scope": {"bytes": 434, "sha256": "2654f6339a446c7181633e50fb347ef3b3002dd0dc96cd6f88db21018f2fd4ba"},
"job:baseline_c0@r00/system": {"bytes": 2865, "sha256": "e8bdde0b2251c9f422c9a02c8c457543798280f5215e740312ffb73f1d0d867f"},
@@ -770,7 +770,7 @@
"chat:planner:planner/messages": {"bytes": 1354, "sha256": "8681af8220fc76fb89bbaacf34c658644272af66ace3045ccef202e75e29482f"},
"chat:planner:planner/sampling": {"bytes": 6272, "sha256": "04dbfbbd42d8b93a12571a1908beacc7eb471e6f2c0fc077371aa86e2a99fb40"},
"chat:planner:planner/system": {"bytes": 15857, "sha256": "faf1d558f09254b16c3f0680aa2117737c72dea402b3b7a0be721c46bdd169fe"},
- "context:runs/run/files": {"bytes": 36698, "sha256": "f295bfd1aa9416926881f65ac65e05a98d799205ea689bc9600719c6fcd61ab2"},
+ "context:runs/run/files": {"bytes": 36708, "sha256": "e11219b9bf4cd4c8e5db008fc15ac93da157c9216bed852b0e2f0bef2eef6dc0"},
"job:baseline@r00/prompt": {"bytes": 15196, "sha256": "479f99ceb71d4bddfecbd0b62b61d92d3756aa69c2e67c260d33e93913c4e25f"},
"job:baseline@r00/scope": {"bytes": 354, "sha256": "2b65c930399cda88341114c6d8289228fedde80a1d17568900e1fe5a055915bd"},
"job:baseline@r00/system": {"bytes": 2205, "sha256": "2ca6cc8997115b5940dcef874355b65c82f25ee8000b964fe0ad96dce25ac0e8"},
@@ -811,7 +811,7 @@
"chat:planner:planner/messages": {"bytes": 1354, "sha256": "5b1a589e27562a3668649aed4be4857a34804a2acdc140eb8fb3630d970f71a6"},
"chat:planner:planner/sampling": {"bytes": 6272, "sha256": "04dbfbbd42d8b93a12571a1908beacc7eb471e6f2c0fc077371aa86e2a99fb40"},
"chat:planner:planner/system": {"bytes": 18287, "sha256": "0e78778a5c81b420699c7148ee039722877c818cea9844f793d543fccc8ef97b"},
- "context:runs/run/files": {"bytes": 41638, "sha256": "b8118f228681041fae720e09b470add3d86bb308ed3ababa21551d3a7494f3b6"},
+ "context:runs/run/files": {"bytes": 41648, "sha256": "bae912e0692ef880672ffb83c7810f5d877bb00c4a8f667dde4af796256875ae"},
"job:baseline@r00/prompt": {"bytes": 17698, "sha256": "f39ab241b54cdce8db4eda26ba44d4664b8869ec13e57b8b5e863fb9d982b459"},
"job:baseline@r00/scope": {"bytes": 432, "sha256": "b7970aa82707a446f41764e551d45617767914f38d6861453d2d61991d48ed05"},
"job:baseline@r00/system": {"bytes": 3208, "sha256": "e24e99f321231e5587cae7f4fbbbcc50fbd254768f4f4c860c4dfa8f83804810"},
@@ -850,7 +850,7 @@
"chat:planner:planner/messages": {"bytes": 1354, "sha256": "5b1a589e27562a3668649aed4be4857a34804a2acdc140eb8fb3630d970f71a6"},
"chat:planner:planner/sampling": {"bytes": 6272, "sha256": "04dbfbbd42d8b93a12571a1908beacc7eb471e6f2c0fc077371aa86e2a99fb40"},
"chat:planner:planner/system": {"bytes": 18287, "sha256": "0e78778a5c81b420699c7148ee039722877c818cea9844f793d543fccc8ef97b"},
- "context:runs/run/files": {"bytes": 41638, "sha256": "b8118f228681041fae720e09b470add3d86bb308ed3ababa21551d3a7494f3b6"},
+ "context:runs/run/files": {"bytes": 41648, "sha256": "bae912e0692ef880672ffb83c7810f5d877bb00c4a8f667dde4af796256875ae"},
"job:baseline@r00/prompt": {"bytes": 17667, "sha256": "92e0968c790d93dcd7b2237f826f314fbbc2f0b82977552f8b5875df7eb1c18c"},
"job:baseline@r00/scope": {"bytes": 648, "sha256": "eed11065615f36b3d1b869300d16518bd4ec6e258630c204fdeff14e676cc2d8"},
"job:baseline@r00/system": {"bytes": 3208, "sha256": "e24e99f321231e5587cae7f4fbbbcc50fbd254768f4f4c860c4dfa8f83804810"},
@@ -881,7 +881,7 @@
"chat:planner:planner/messages": {"bytes": 1354, "sha256": "5b1a589e27562a3668649aed4be4857a34804a2acdc140eb8fb3630d970f71a6"},
"chat:planner:planner/sampling": {"bytes": 6272, "sha256": "04dbfbbd42d8b93a12571a1908beacc7eb471e6f2c0fc077371aa86e2a99fb40"},
"chat:planner:planner/system": {"bytes": 18287, "sha256": "0e78778a5c81b420699c7148ee039722877c818cea9844f793d543fccc8ef97b"},
- "context:runs/run/files": {"bytes": 41638, "sha256": "b8118f228681041fae720e09b470add3d86bb308ed3ababa21551d3a7494f3b6"},
+ "context:runs/run/files": {"bytes": 41648, "sha256": "bae912e0692ef880672ffb83c7810f5d877bb00c4a8f667dde4af796256875ae"},
"job:baseline@r00/prompt": {"bytes": 17698, "sha256": "f39ab241b54cdce8db4eda26ba44d4664b8869ec13e57b8b5e863fb9d982b459"},
"job:baseline@r00/scope": {"bytes": 432, "sha256": "b7970aa82707a446f41764e551d45617767914f38d6861453d2d61991d48ed05"},
"job:baseline@r00/system": {"bytes": 3208, "sha256": "e24e99f321231e5587cae7f4fbbbcc50fbd254768f4f4c860c4dfa8f83804810"},
diff --git a/harness/tests/prompts/test_manifest.py b/harness/tests/prompts/test_manifest.py
index 438abe03..ed8aa139 100644
--- a/harness/tests/prompts/test_manifest.py
+++ b/harness/tests/prompts/test_manifest.py
@@ -10,20 +10,19 @@ import pytest
from tests.prompts import manifest as M
[email protected](scope="module", params=M.ALL)
-def scenario(request, tmp_path_factory) -> tuple[str, dict]:
- """One run per scenario for both tests below (module scope keeps them together)."""
- name = request.param
- return name, M.digest(M.run_scenario(name, tmp_path_factory.mktemp(name.replace(".", "_"))))
-
[email protected](("section", "hint"), [("payloads", M.PAYLOAD_HINT), ("stage_keys", M.STAGE_HINT)],
- ids=["payloads", "stage_keys"])
-def test_the_run_matches_the_manifest(scenario, section: str, hint: str) -> None:
- name, got = scenario
[email protected]("name", M.ALL)
+def test_the_run_matches_the_manifest(name: str, tmp_path) -> None:
+ """Check both sections in one case so xdist cannot run the same scenario twice."""
stored = M.load()
assert name in stored, f"{name} is not in {M.MANIFEST.name}: bless it"
- lines = M.diff(got[section], stored[name][section])
- assert not lines, f"{name}: {section} changed\n" + "\n".join(lines) + "\n\n" + hint
+ got = M.digest(M.run_scenario(name, tmp_path))
+ changes = []
+ for section, hint in (("payloads", M.PAYLOAD_HINT), ("stage_keys", M.STAGE_HINT)):
+ lines = M.diff(got[section], stored[name][section])
+ if lines:
+ changes.append(f"{name}: {section} changed\n" + "\n".join(lines) + "\n\n" + hint)
+ assert not changes, "\n\n".join(changes)
+
def test_the_manifest_holds_exactly_the_scenarios() -> None:
assert sorted(M.load()) == M.ALL, "bless after adding or removing a scenario"
diff --git a/harness/tests/scene_runtime/lib/_probe.py b/harness/tests/scene_runtime/lib/_probe.py
index fca893ca..1ff88c5e 100644
--- a/harness/tests/scene_runtime/lib/_probe.py
+++ b/harness/tests/scene_runtime/lib/_probe.py
@@ -2,7 +2,6 @@
from __future__ import annotations
import json
-import re
import shutil
import tempfile
from pathlib import Path
@@ -17,35 +16,6 @@ LIB_DIR = HARNESS / "codeverse3d" / "languages" / "scene_threejs" / "starter" /
pytestmark = pytest.mark.node
-# The two hooks patchStandard injects into, as a material three would.
-# Nothing else is in the source, so what comes back is the patch itself:
-# `compile(mat)` runs a material's onBeforeCompile over it.
-SHADER_JS = """
-const fake = () => ({
- vertexShader: 'void main() {\\n#include <begin_vertex>\\n}',
- fragmentShader: 'void main() {\\n#include <color_fragment>\\n}',
- uniforms: {},
-});
-function compile(mat) {
- const shader = fake();
- mat.onBeforeCompile(shader);
- return shader;
-}
-"""
-
-
-def _find(pattern: str, src: str) -> re.Match:
- """``re.search`` that fails naming the pattern and the source it missed."""
- m = re.search(pattern, src)
- assert m, f"{pattern} not in\n{src}"
- return m
-
-
-def _main_body(src: str) -> str:
- """The shader from ``void main`` on: where the patch bodies land."""
- return src[src.index("void main"):]
-
-
def _three_ready() -> None:
if not (node_modules_dir() / "three" / "package.json").is_file():
pytest.skip("runtime_js/node_modules/three missing (npm ci in runtime_js)")
diff --git a/harness/tests/skills/test_targets.py b/harness/tests/skills/test_targets.py
index ffaffd5f..5f0035bc 100644
--- a/harness/tests/skills/test_targets.py
+++ b/harness/tests/skills/test_targets.py
@@ -15,12 +15,6 @@ from codeverse3d.skills.registry import ROUTED_SKILLS
BUNDLES = bundle_dirs()
pytestmark = pytest.mark.skipif(not BUNDLES, reason=f"no bundles in {skills_dir()} yet")
-def _live_kinds() -> set[str]:
- from codeverse3d.skills import registry
-
- return {v for k, v in vars(registry).items()
- if k.isupper() and isinstance(v, str) and "/" in v}
-
# --------------------------------------------------------------------------- the table
def test_target_table_exactly_covers_the_shipped_and_routed_bundles():
"""No orphan rows, unmeasured routes, duplicate rows or bundles maintained by taste."""
diff --git a/harness/tests/skills/test_variants.py b/harness/tests/skills/test_variants.py
index 4973b4fa..4e949476 100644
--- a/harness/tests/skills/test_variants.py
+++ b/harness/tests/skills/test_variants.py
@@ -149,7 +149,7 @@ def test_routes_toml_typing_errors_are_refused(variant, text, problem):
BROKEN_SKILL = "---\nname: c3d-bbox-contract\ndescription: the closing fence is missing\n# body\n"
def _cli(*args: str, width: int = 300):
- """``3dcode skills ...`` through typer's runner, the CLI console wide enough for one table row per
+ """``3dcodeverse skills ...`` through typer's runner, the CLI console wide enough for one table row per
line: an xdist worker runs under ``COLUMNS=80``, which rich pins on the console when it is built, so
the width is set on the console itself (and restored) rather than through the environment."""
from unittest import mock
diff --git a/harness/tests/spatial_tools/test_tools_cli.py b/harness/tests/spatial_tools/test_tools_cli.py
index 49ef6800..837a688a 100644
--- a/harness/tests/spatial_tools/test_tools_cli.py
+++ b/harness/tests/spatial_tools/test_tools_cli.py
@@ -1,4 +1,4 @@
-"""`3dcode tools <name>`: the three-state panel and the exit code behind it."""
+"""`3dcodeverse tools <name>`: the three-state panel and the exit code behind it."""
from __future__ import annotations
@@ -8,7 +8,7 @@ from codeverse3d.spatial.registry import ToolContext
def test_cli_exit_code_follows_the_verdict_not_the_failure(stool_ctx: ToolContext) -> None:
- """`3dcode tools` exits 1 on ``not ok`` — a FAIL verdict included (docs/COST.md §30)."""
+ """`3dcodeverse tools` exits 1 on ``not ok`` — a FAIL verdict included (docs/COST.md §30)."""
from codeverse3d.cli.main import app
runner = CliRunner()