feat(blender): blender-run one-shot headless wrapper + FLEETTOOLS entry

scripts/blender-run launches each call as a docker run --rm of the Blender
image on fv-ml1 GPU 3, capped at 64g / 48 CPUs. It needs no desktop and does
not affect the GUI container's lifecycle. --job DIR stages a local directory
to /tank/blender/jobs/<name>/, runs Blender with that as the cwd, and copies
results back. It always passes --python-exit-code 1, because Blender otherwise
exits 0 when a --python script raises (measured).

Tested headless: Cycles GPU and CPU, EEVEE via EGL, Workbench, an STL
round-trip, and exit codes (3, 7 and 1 pass through). There is no STEP
importer. Written for draupnir's design work, and indexed in FLEETTOOLS with a
detail file.

docs/fleettools/FLEETTOOLS.md

@@ -76,6 +76,12 @@ live contract for its API. Fetch it rather than trusting a transcription.
   *When:* you need images rendered, or a character LoRA trained.
   *Detail:* `/home/lkraven/development/eshpfi-management/docs/fleettools/arbo.md`
 
+- **Blender** — Blender 5.2 LTS on fv-ml1 GPU 3 (borrowed, on demand). `scripts/blender-run` for
+  headless one-shot renders and conversions; `scripts/blender-mcp` for agent-driven scenes
+  (register per task).
+  *When:* 3D rendering, STL → image, scene building.
+  *Detail:* `/home/lkraven/development/eshpfi-management/docs/fleettools/blender.md`
+
 ## Working on the fleet itself
 
 - **elway** — SSH playbook runner for **CHANGING** things.

docs/fleettools/blender.md

@@ -0,0 +1,56 @@
+# Blender: headless and agent-driven 3D on fv-ml1 GPU 3
+
+**Blender 5.2.2 LTS** (bundled Python 3.13), runs on **fv-ml1 GPU 3** (RTX PRO 6000 Blackwell,
+96 GB). Stack and full notes: `/home/lkraven/development/eshpfi-management/stacks/blender/README.md`.
+
+⚠ **GPU 3 is borrowed.** It is the fleet's reserve card for a full-size vLLM seat. Blender
+runs only while in use, and this access ends if a big seat moves onto the card.
+
+## Two ways in
+
+| You are… | Use | Shape |
+|---|---|---|
+| a script or CLI caller (renders, conversions) | `scripts/blender-run` | one-shot `docker run --rm`: no desktop, gone when Blender exits |
+| an agent building scenes interactively | `scripts/blender-mcp` (MCP, per task) | a GUI Blender plus the mcp-for-blender add-on; `up` / `status` / `down` |
+
+Both scripts are in `/home/lkraven/development/eshpfi-management/scripts/` and run from nh3-dev,
+reaching fv-ml1 as `infra-ops@10.251.50.54` over ssh. **No HTTP API** and no openapi.json.
+
+## blender-run (headless)
+
+```sh
+scripts/blender-run --job /path/to/jobdir -- --python render.py -- out.png
+scripts/blender-run -- --python-expr 'import bpy; print(bpy.app.version_string)'
+```
+
+- Always adds `-b --factory-startup --python-exit-code 1`. Arguments go after `--`.
+- **Files:** fv-ml1 does not mount `/mnt/smithy`. `--job DIR` copies DIR to
+  `fv-ml1:/tank/blender/jobs/<name>/`, runs with that as the working directory, and copies
+  new or changed files back into DIR. Nothing is deleted on either side. Use relative
+  input paths inside the job.
+- **Engines headless (tested 2026-09-28):** Cycles on GPU (OptiX/CUDA), Cycles on CPU, EEVEE
+  (EGL, no display needed, ~9 s with the shader compile on first use), Workbench. In 5.2 the
+  EEVEE id is `BLENDER_EEVEE` (`BLENDER_EEVEE_NEXT` is gone). For Cycles GPU, set
+  `prefs.compute_device_type = 'OPTIX'` and enable the OPTIX devices, then
+  `scene.cycles.device = 'GPU'`. Factory startup defaults to CPU.
+- **Import:** STL is built in (`bpy.ops.wm.stl_import`). **No STEP importer** is installed or
+  built in.
+- **Budget:** each run is capped at 64 GB RAM and 48 CPUs, with up to the whole 96 GB of VRAM.
+  Keep to about 2 concurrent renders. It is not on irv-ml1, so irv-ml1's working-set budget
+  does not apply.
+
+⚠ **Foot-guns** (all measured):
+- A `--python` script that raises exits **0** unless `--python-exit-code` is set. blender-run
+  sets it.
+- `render.filepath` must be **absolute**. Blender does not resolve a relative output path
+  against the working directory ("cannot save 'out.png'"). Importers and Python file I/O do.
+- "HIPEW initialization failed" on stderr is harmless: it is the AMD backend probing.
+- The first OptiX render in a process includes about 1-2 s of kernel load.
+
+## blender-mcp (agents)
+
+Register it **per task**, never user- or project-wide (Prime, 2026-09-27):
+`claude mcp add blender -- /home/lkraven/development/eshpfi-management/scripts/blender-mcp`.
+Then run `scripts/blender-mcp up`, wait for `status` to say answering, work, and run `down`
+when finished. Safe mode is on (no os/open/network in agent code), so the startup hook has
+already pointed Cycles at OptiX. Save to `/work/…`. Details and traps are in the stack README.

scripts/blender-run

@@ -0,0 +1,66 @@
+#!/usr/bin/env bash
+# blender-run — one-shot HEADLESS Blender on fv-ml1 GPU 3, for scripted/CLI callers (draupnir etc.).
+# The agent-driven, interactive path is scripts/blender-mcp; this is the batch path.
+#
+#   scripts/blender-run [--job DIR] -- <blender args after -b>
+#   scripts/blender-run --job /mnt/smithy/draupnir/j42 -- --python render.py -- --out out.png
+#   scripts/blender-run -- --python-expr 'import bpy; print(bpy.app.version_string)'
+#
+# Each call is its own `docker run --rm` of the stacks/blender image (Blender 5.2.2 LTS, Python
+# 3.13): no desktop, no MCP socket, gone when Blender exits. So it never collides with the on-demand
+# GUI container, and GPU 3 goes back to 0 when the render ends. Always added: `-b --factory-startup
+# --python-exit-code 1`. Without that last flag, Blender exits 0 even when a --python script raises
+# (measured).
+#
+# ⚠ Give render.filepath an ABSOLUTE path (os.path.abspath). Blender does not resolve a relative
+#   output path against the working directory: "cannot save 'out.png'". Python file I/O and
+#   importers do use the cwd.
+#
+# Files: fv-ml1 does NOT mount /mnt/smithy. With --job DIR, DIR is copied to
+# fv-ml1:/tank/blender/jobs/<name>/, Blender runs WITH THAT AS ITS WORKING DIRECTORY, and new or
+# changed files are copied back into DIR afterwards (nothing is deleted on either side). Use
+# relative paths inside the job. Without --job, the only paths Blender sees are under
+# /work (= fv-ml1:/tank/blender).
+#
+# Budget: GPU 3 (96 GB, borrowed from the vLLM reserve: this ends if a full-size seat moves in).
+# The container is capped at 64 GB RAM / 48 CPUs so a runaway render cannot starve the inference
+# seats on the same host.
+set -euo pipefail
+
+HOST=${BLENDER_SSH_HOST:-infra-ops@10.251.50.54}
+ENV_FILE=/opt/docker/compose/blender/.env
+JOB=""
+while [ $# -gt 0 ]; do
+  case "$1" in
+    --job) JOB=${2:?--job needs a directory}; shift 2 ;;
+    --) shift; break ;;
+    -h|--help) sed -n 2,23p "$0"; exit 0 ;;
+    *) echo "blender-run: unknown option $1 (blender args go after --)" >&2; exit 2 ;;
+  esac
+done
+
+WORKDIR=/work
+if [ -n "$JOB" ]; then
+  [ -d "$JOB" ] || { echo "blender-run: --job $JOB is not a directory" >&2; exit 2; }
+  NAME=$(basename "$(realpath "$JOB")")
+  [[ $NAME =~ ^[A-Za-z0-9._-]+$ ]] || { echo "blender-run: job dir name '$NAME' must be [A-Za-z0-9._-]" >&2; exit 2; }
+  ssh -n -o BatchMode=yes "$HOST" "mkdir -p /tank/blender/jobs/$NAME"
+  rsync -a "$JOB"/ "$HOST:/tank/blender/jobs/$NAME/"
+  WORKDIR=/work/jobs/$NAME
+fi
+
+# Arguments travel as one shell-quoted string: ssh flattens argv into a remote command line.
+ARGS=$(printf '%q ' "$@")
+set +e
+ssh -n -o BatchMode=yes "$HOST" "IMG=\$(grep '^IMAGE=' $ENV_FILE | cut -d= -f2) && \
+  exec docker run --rm --name blender-run-\$\$ --runtime nvidia \
+    -e NVIDIA_VISIBLE_DEVICES=3 -e NVIDIA_DRIVER_CAPABILITIES=all \
+    --user 1002:1003 -e HOME=/tmp --memory 64g --cpus 48 \
+    -v /tank/blender:/work -w $WORKDIR --entrypoint /blender/blender \"\$IMG\" \
+    -b --factory-startup --python-exit-code 1 $ARGS"
+RC=$?
+set -e
+if [ -n "$JOB" ]; then
+  rsync -a --update "$HOST:/tank/blender/jobs/$NAME/" "$JOB"/
+fi
+exit $RC

stacks/blender/README.md

@@ -24,7 +24,15 @@ ssh infra-ops@10.251.50.54 'cd /opt/docker/compose/blender && docker compose dow
 `restart: "no"`, so a reboot never brings it back. **When a big seat moves onto GPU 3, Blender
 stays down.**
 
-## Headless rendering (no desktop needed)
+## Headless rendering: `scripts/blender-run` (for scripts and CLI callers)
+
+A one-shot `docker run --rm` of this image with Blender as the entrypoint. It needs no desktop and
+does not collide with the GUI container's up/down. `--job DIR` stages a local dir to
+`/tank/blender/jobs/<name>/` and copies results back. Engines tested headless on 2026-09-28:
+Cycles GPU and CPU, EEVEE (EGL), Workbench. STL import is built in; there is **no STEP importer**.
+Foot-guns and budget are in `docs/fleettools/blender.md`. First consumer: draupnir.
+
+## Headless rendering inside the running GUI container
 
 ```bash
 ssh infra-ops@10.251.50.54 'docker exec -u abc blender blender -b /work/<file>.blend -E CYCLES -o /work/out/frame_#### -a -- --cycles-device OPTIX'