Hermes Overnight Worker: Blender asset skills that run while you sleep, with the owner's yes at both ends
by PaperChaseWebb
An unattended Blender worker for game props. The owner approves tonight's queue, the agent builds each prop headless inside triangle, minute and critique budgets, writes only to a staging folder, a separate critic session reviews every preview, and nothing moves into the real project until the owner says yes in the morning and a read-only check comes back clean. Real scripts, tested on Blender 5.2.3, with a seeded fixture of fifteen planted faults.
- Price
- $29
- Difficulty
- Intermediate
- Setup
- 1 hr 5 min
- Format
- .md file
hermes-overnight-worker-blender-asset-skills.md · 75 KB · instant download after checkout
Want this built for you instead? We can build it for you.
What you end up with
A night of props built and reviewed without anyone watching, the owner's files left untouched, and nothing promoted, published or bought without a yes.
What you get
- 21 sections in one .md file
- The worker folder: worker.json, queue.json, originals, staging, approvals, manifest and promote files
- Three budgets per job (triangles, minutes, critique rounds) and a night budget
- Three kinds of yes: queue before the run, outside asset, and promote after the run
- Four skills in the Agent Skills format: overnight-queue, night-builder, asset-critic, morning-report
- build_prop.py: one headless Blender job that refuses outside staging and enforces the triangle cap
- night.py: the runner with the queue gate, timeouts, critic-is-not-builder refusal and a promote step that never overwrites
- queue_preflight.py: six evening checks before anything is built
- queue_check.py: fifteen read-only checks on the finished night
- A seeded fixture with 15 planted faults, the 15 fixes and the expected output
- A real-Blender demo night (night_demo.sh) with its expected output
- A rules block, ten prompts and setup notes for Grok Bot, Claude Code, Codex and Hermes
- A one-evening install plan, go-live checklist, morning and weekly review, and failure table
- A look at seven Blender agent tools and skill packs, with sources
Who it is for
Solo developers and small teams who already script props in Blender, or want to, and want an agent to work through a list overnight without the risk of it touching the real project, publishing anything or spending money.
Works with
Solid chips: the agent is named in this listing's own tools list. Dashed chip: Plain Markdown, so it also pastes into other agents such as Grok Bot, Hermes, OpenClaw or Muse. We list those as compatible by format only, not as tested.
Tools needed
Cost to run: The pack is the only purchase. Blender is free. Your harness and model usage are billed by those vendors.
Time to result: Set up and tested in one evening; the first watched night runs the same night
How it works
Test the check
Save queue_check.py and fixture/make_fixture.py from the file, build the fixture and run the check on both folders.
Expected: 15 findings on seeded, then 0 on fixed. No Blender needed.
Run the demo night
Save build_prop.py, night.py and fixture/night_demo.sh, then run the demo with the full path to your Blender binary.
Expected: The output matches Test B in section 11: refused without a yes, two crates built, one over budget, promote blocked, then copied, then skipped.
Set up the worker
Make the worker folder, write worker.json with the owner, allowed sources and a night budget, and copy in only the originals tonight's jobs need.
Expected: queue_preflight.py runs on your folder.
Install the skills
Put the four skills and the rules block where your harness reads them, and remove network, store, payment and MCP bridge tools from the night sessions.
Expected: The night session can run night.py and nothing that reaches outside the folder.
Watch the first night
Queue three to five props, get the queue yes, schedule the night and the critic, then read the morning report and promote one prop with the owner's yes.
Expected: queue_check.py prints 0 before promote, and the promoted file is the only change in the project.
Sample prompt from the pack
Show tonight's queue as a numbered card: id, name, brief, source, tris_max, minutes_max, rounds_max, and NEEDS YES for any outside asset. End with the total minutes against the night budget and: reply 'yes queue <run_id>' to approve. Do not write the approval row yourself.
Prompt 2 of 10. Runs after the rules block in section 12.
You are the critic, not the builder. For each done job, open preview.png and result.json and judge them against the brief only. Write critique.md with keep, revise or drop and one line of reasons, then record it with night.py critique. Do not edit any file in staging.
Prompt 5 of 10.
Free preview of the file
# Hermes Overnight Worker: Blender asset skills that run while you sleep, with the owner's yes at both ends Version 1.0, 2026-10-28. By PaperChaseWebb. Works with Grok Bot, Claude Code, Codex and Hermes. Plain Markdown, four Python 3 scripts and one shell script, tested on Blender 5.2.3 LTS. This pack is the worker itself: the folder, the queue, the budgets, the build script, the runner, the check and the morning promote step. You write a short list of props in the evening, the owner approves it, the agent builds them headless in Blender overnight into a staging folder, a separate critic session reviews each one, and in the morning nothing moves into the real project until the owner says yes and the check comes back clean. It does not repeat how to design skills or lock a visual style. Skill design (reference lock, plan with timings, build and critique as three skills) is in the Agent Skills Workshop (pp-agent-skills-workshop). Style locking and assets as code are in Consistent Asset Sets (pp-consistent-asset-sets). This pack is the part those leave to you: the files and rules that make an unattended night safe to leave running. ## Contents 1. What the worker does 2. Who it is for, and when not to use it 3. The night, step by step 4. The worker folder 5. queue.json and the three budgets 6. Approval rules: queue yes, outside asset yes, promote yes 7. The four skills 8. build_prop.py: one headless job 9. night.py: the runner 10. queue_check.py: the read-only check 11. Test
Worth knowing
- It builds scripted, low poly props from parameters. It does not generate meshes from text or images, and it does not replace an artist's judgment on hero assets.
- It never publishes, uploads, buys, installs add-ons or deletes. Those stay daytime tasks with the owner present.
- It runs Blender headless with no MCP bridge. If you also use a live Blender bridge in the daytime, keep it out of the night sessions.
- The skills and prompts guide the agent; night.py and the checks are what enforce the rules. Keep the night session's tools limited as described.
- On 2026-10-28, on Blender 5.2.3 LTS, queue_check.py reported the fifteen expected findings on the seeded fixture, then 0 on the fixed one, queue_preflight.py reported its expected 5 then 0, and night_demo.sh built two crates, stopped one over budget, and refused, blocked and promoted exactly as listed in section 11. Run both tests on your own machine first.
Questions
How is this different from a Blender MCP server or the Hermes Blender plugin?
Those connect an agent to a live Blender session so it can act in the scene while you watch. This pack is for when nobody is watching: a queue the owner approved, budgets the runner enforces, writes to staging only, a separate critic, and a promote step that needs a yes and a clean check. It runs Blender headless and needs no bridge.
Does it need Hermes?
No. The skills use the Agent Skills format, which Hermes, Claude Code and Codex read, and the pack has setup notes for each and for Grok Bot. The scripts are plain Python and run under any of them, or by hand.
What stops the agent from changing my real files?
Originals are hashed before the run and the morning check fails if one changed. Outputs must be inside staging/<run_id>/. Promote copies only rows with the owner's yes, given after the run ended, and never overwrites a file in the project.
How many prompts does it have?
Ten. The pack is built around tested files (four scripts, a seeded fixture and a demo night with expected output) rather than a long prompt list.
Details
Version 1.0 · Updated Oct 28, 2026 · Markdown version of this page
License: Use on your own projects and on installs you do for clients. Do not resell or republish the pack itself. Download terms
Related
The overnight queue rules and the evening preflight check (free sample)
From the Hermes Overnight Worker: the six rules for an unattended agent night (queue yes before the run, three budgets, staging only, a separate critic, nothing published or bought, promote only with a yes), plus a read-only Python check that tests tonight's queue against them, with a worked example and its expected output.
Agent Skills Workshop: write, test and run SKILL.md skills for Hermes, Claude Code and Codex
A bot template and teaching guide for building your own agent skills instead of collecting other people's. A skill card template, a three-skill split for long jobs (reference lock, plan with timings, build and critique), a loop that turns your corrections into skill drafts, an intake check for skills you did not write, four action classes, a SKILLS.md register, an overnight run sheet, eight skill drills, a rules block and 14 prompts.
Consistent Asset Sets: one approved reference, style files and review sheets for AI-made art, props and cards
A small production system for making many AI-made assets that belong together: approve one reference, write STYLE.md from it, request only deltas, build props as code, and review every batch on one sheet with a ten-point check.
Game Dev with AI Agents: method, prompts, skills and check loops
A Markdown playbook for building games with Claude Code, Codex or Claude: a seven-stage method, a staged brief, 16 prompts, five agent skills, self-check loops, engine notes and 30 game concepts.
Make a 3D product ad with Claude Code and headless Blender on an old GPU
A 15-second, four-shot product spot rendered on a 2016 graphics card. Claude Code wrote the whole Blender scene in Python, Blender rendered it with no window open, and ffmpeg put the shots, music and overlay together. The builder posted the render times and what broke.
Cross-Harness Relay: Grok Bot, Claude Code, Codex and Hermes on one job
A bot template and teaching guide for running more than one agent harness on the same work. Six seat cards (front desk, planner, builder, reviewer, tester, night shift), a RELAY.md seat map, a relay packet for every handoff, four loop shapes, routine-or-judgment routing with a fallback when limits run tight, a preflight check, the front desk and night shift split with a proof footer, shared memory rules, eight relay drills, a rules block and 14 prompts.
Ask-first rules for AI agents: hard nos, approval cards and a test set
Paste-in rules that make an agent ask before it sends, spends, posts or deletes, define what a real yes is, and stop retries after a no, plus 12 test scenarios to check it.
Engine Agent Kits: Godot 4, Unity 6 and Unreal 5 rules files, editor bridges and build loops for coding agents
One kit per engine for coding agents: a rules file, a trap table of version mistakes, terminal commands to check, test and build without clicking, ways to give the agent eyes, asset pipelines with a license ledger and a one-milestone work loop.
Unreal from Codex: Unreal MCP setup, a session contract and a tested asset pipeline for coding agents
Connect Codex, Claude Code, Grok Bot or Hermes to the Unreal MCP server in Unreal Engine 5.8 and keep every session in bounds: a session file with a checkpoint commit and a leave-out list, five risk tiers, owner approvals tied to file hashes, an action log with a verify result for every call, and recovery steps for editor restarts, stale tool lists, compiles and timeouts. Adds a generator to Blender to Unreal asset path with a headless Blender prep script, an import script that refuses changed files, a connection probe and a fifteen-check read-only checker.
Game Assets as Code: one spec file, seeded generators and a fifteen-check checker for agent-made game art and sound
Let Grok Bot, Claude Code, Codex or Hermes make your placeholder and jam-ready game assets as code instead of prompts: pixel sprites from letter grids, seeded symmetric sprite variants, seamless tiles, SVG icons, retro sound effects and low-poly Blender props built headless, all from one spec file with a fixed palette and seeds. Every build writes a manifest with file and spec hashes and a review page; your approvals are tied to those hashes, a promote step copies only approved, unchanged files into the game, and a separate release yes covers anything that leaves your machine.