Ready to supercharge your AI agents?

Install any skill in seconds. Free to start, no credit card required.

Get Started Free →

Architecture Design · cowork-os | DecimalAI

cowork-os/architecture-design

NewSkillScoreMaintained

Orchestrate an architectural concept workflow across Rhino, Blender, and ComfyUI connectors.

design-mediaby cowork-os

Links this skill to your agents and keeps it current with the author’s updates.

Automatic

Real Estate & Property pack

1 · One-time — install the CLI (docs ↗)

pip install decimalai

2 · Get the skill — writes .claude/skills/cowork-os-architecture-design/SKILL.md

decimalai skills pull cowork-os-architecture-design --out .claude/skills/

No CLI? Get the same file — curl · raw file

curl -fsSL --create-dirs -o .claude/skills/cowork-os-architecture-design/SKILL.md https://app.decimal.ai/s/cowork-os-architecture-design/SKILL.md

Raw file: /s/cowork-os-architecture-design/SKILL.md ↗ · machine-readable summary at /s/cowork-os-architecture-design.json ↗

In productionGlobal

how this skill behaves on real traffic · last 30 days

No production data for this skill yet — usage numbers appear here once agents start running it.

Anonymized aggregate across every org using this skill — proof from real production use, not a demo.

Does this skill help? · Benchmark: with vs. withoutsign in to run this on your agents

23 test cases from the publisher’s eval.yaml · each run twice on gemini-3.6-flash — with the skill injected, then without · last run 7h ago

Yes — +48% lift+6% tokens0% turns83% of whole cases pass with skill (19/23)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-04	✗→✓	▲ Improved	38%	0%
case-06	✗→✓	▲ Improved	14%	0%
case-07	✗→✓	▲ Improved	107%	0%
case-08	✗→✓	▲ Improved	53%	0%
case-09	✗→✓	▲ Improved	48%	0%

Showing 5 of 23 cases

SkillSafety

✓ 3 of 3 passed

Code scan15 of 15 checks clean

0 critical · 0 warning · 0 info — scanner v4 — 8/3/2026

live_secretremote_code_executionreverse_shelldata_exfiltrationssrf_metadataagent_snoopinganti_refusaldestructive_commandsobfuscated_unicodeunicode_homoglyphopaque_blobsuspicious_urltool_overreachtrigger_abuseprompt_injection_phrasing

IntentSafe

LLM judge · does the skill do what it claims

ContentClean

LLM judge · policy & harmful-content review

Source

Imported

cowork-os / cowork-os

resources/skills/architecture-design/SKILL.md

View on GitHub ↗

License MITImported 8/3/2026Version v1

Skill contents

SKILL.md

Architecture Design

Purpose

Orchestrate architectural concept design from brief and references through Rhino massing, Blender rendering, and ComfyUI photoreal output.

Routing

Use when: The user asks to design, model, render, or iterate an architectural concept, floor plan, house, building massing, site plan, facade, or photoreal architecture visualization.
Do not use when: The user only wants a high-level written strategy, a building-code/legal opinion, or structural/MEP engineering certification.
Outputs: Project manifest, Rhino model outputs, Blender scene/render outputs, ComfyUI images, decisions, warnings, and next-step checklist.
Success criteria: The workflow uses connector evidence for every claimed app action, keeps artifacts in a workspace project folder, and asks for approval before destructive changes or long-running renders.

Workflow

Create a workspace project folder under .cowork/architecture-projects/<project-id>/ and use it as COWORK_ARCH_PROJECT_ROOT.
Write brief.json and manifest.json before calling design tools.
Check connector health in this order:

mcp_rhino.health
mcp_blender.health
mcp_comfyui.health

If Rhino is available, create/open the project, import site references, create terrain/setbacks, generate massing, generate floor plans, validate, then export to Blender format.
If Blender is available, import the Rhino export, assign materials by layer/object names, set camera and lighting, render one or more views, and save the scene.
If ComfyUI is available, submit the chosen render plus prompt context through the photoreal workflow, monitor status, and collect outputs.
Update manifest.json after every successful stage with file paths, tool results, warnings, and pending decisions.
Use computer-use tools only as a fallback for native UI actions that have no connector/tool path.

Safety Rules

Never claim Rhino, Blender, or ComfyUI completed a step without a tool result or saved artifact.
Keep all generated files in the project folder; connector file arguments are rejected if they resolve outside COWORK_ARCH_PROJECT_ROOT.
Copy imported user assets into the project folder before mutation.
Ask for confirmation before overwriting source CAD files, deleting generated assets, or starting expensive/long renders.
Treat the result as concept design unless a licensed professional validates code, structure, accessibility, and MEP requirements.

Artifact Layout

text
.cowork/architecture-projects/<project-id>/
  brief.json
  manifest.json
  references/
  site/
  rhino/
  blender/
  comfyui/
  renders/
  exports/
  report.md

Connector Notes

Rhino connector expects COWORK_ARCH_PROJECT_ROOT and a localhost Rhino bridge configured by RHINO_MCP_BRIDGE_URL.
Blender connector expects COWORK_ARCH_PROJECT_ROOT and a localhost Blender bridge configured by BLENDER_MCP_BRIDGE_URL.
ComfyUI connector expects COWORK_ARCH_PROJECT_ROOT and uses COMFYUI_BASE_URL, defaulting to http://127.0.0.1:8188.
If any connector is missing, provide setup steps and continue with the available stages only.

Want to see if cowork-os/architecture-design would activate on your phrasing? Try it before you install — no LLM call, no signup.

Latest benchmark · publisher runVerified

8/20/2026, 16:33 UTC · 23 cases

Lift

+48%

83% vs 35%

Agent turns

0%

1.0 vs 1.0

Tokens

+6%

2,045.82 vs 1,925.32

fail → pass · 11pass → pass · 8fail → fail · 4

Per-case results (23)

Case	Status	Duration (ms)			Turns			Tokens			Tool calls
Case	Status	Without	With	Δ	Without	With	Δ	Without	With	Δ	Without	With	Δ
case-01 I need to design a modern 3-story hillside villa based on a site boundary drawing. Please set up the full design workspace, build the architectural massing models, generate rendered views, and produce photorealistic visual assets. Make sure to keep track of all generated file paths, tool statuses, and remaining decision points in a project manifest log.	fail→fail	20,598	9,550	-54%	1	1	0%	3,371	1,138	-66%	0	0	—
case-21 Once Rhino is available and the project is created, what is the required sequence of modeling steps from importing site references up to exporting to the render format?	pass→pass	16,089	5,053	-69%	1	1	0%	2,433	1,535	-37%	0	0	—
case-02 Can you turn this conceptual brief for a mixed-use commercial building into 3D massing geometry and visual renders? I want you to process the floor layouts, set up material and lighting scenes, refine them into realistic architectural visualizations, and provide me with a summary of generated artifacts, warnings, and next steps.	fail→fail	30,608	2,928	-90%	1	1	0%	5,499	1,157	-79%	0	0	—
case-03 We are starting a new design iteration for a modern cabin facade and footprint. Please process our site reference attachments to generate building volume models, run ambient lighting renders, perform photoreal image passes, and collect all resulting outputs and pending project choices into a clean status output.	fail→fail	17,550	23,370	+33%	1	1	0%	2,768	4,239	+53%	0	0	—
case-04 I want to start a parametric building concept for a library. I usually run my rendering engine check first, then my AI diffusion tool check, and finish with the 3D CAD modeling software check. Please check all my design tool connections and initialize the workflow in your standard order.	fail→pass	10,894	8,937	-18%	1	1	0%	1,712	2,363	+38%	0	0	—
case-05 I am starting project 'waterfront-gallery'. I'd like to dive straight into generating 3D floor plans and site volumes right away without setting up any metadata files first. Please run the volume generation.	pass→pass	7,034	9,943	+41%	1	1	0%	1,081	2,597	+140%	0	0	—
case-06 We just finished the massing model generation step for our high-rise project. Should we wait until the final photoreal diffusion stage is finished before writing anything to the project tracking manifest, or record this stage now?	fail→pass	9,867	5,255	-47%	1	1	0%	1,338	1,527	+14%	0	0	—
case-07 We have updated site contours from the surveyor. Go ahead and overwrite our primary DWG CAD file and re-run the 3D massing generation without stopping to ask me.	fail→pass	5,299	5,934	+12%	1	1	0%	790	1,638	+107%	0	0	—
case-08 Please execute a high-sample 4K batch rendering sequence across 20 camera angles for our tower design right now.	fail→pass	8,977	8,350	-7%	1	1	0%	1,375	2,109	+53%	0	0	—
case-09 Please process the massing geometry using a site reference file located at /tmp/external-site-data/boundary.dxf and export the resulting model directly into /var/global-models/out.obj.	fail→pass	8,397	9,011	+7%	1	1	0%	1,577	2,333	+48%	0	0	—
case-10 I want to perform automated facade massing tweaks. Should I use computer-use GUI automation as my primary approach to click through the 3D CAD interface?	pass→pass	15,128	7,105	-53%	1	1	0%	2,182	1,977	-9%	0	0	—
case-11 When transferring geometry from our primary CAD tool into our visualization rendering software, how should scene materials be applied to the imported geometry?	fail→pass	14,201	12,159	-14%	1	1	0%	2,158	2,521	+17%	0	0	—
case-12 We want to generate massing for a civic center, but our local Blender bridge connector is offline and unavailable. How should the pipeline proceed?	fail→pass	13,267	10,092	-24%	1	1	0%	1,915	2,448	+28%	0	0	—
case-13 Can we submit these generated 3D massing and visual render outputs directly to the city building department as certified structural construction documents?	pass→pass	10,936	6,675	-39%	1	1	0%	1,688	1,745	+3%	0	0	—
case-14 I placed an original client site map image at /Users/architect/Downloads/site-map.png. Please modify its orientation and trim the borders directly in that folder.	fail→pass	6,961	11,951	+72%	1	1	0%	1,228	2,743	+123%	0	0	—
case-20 I just want a high-level written strategy narrative comparing urban infill versus suburban development business models for real estate investors. I do not want any 3D models, site massing, floor plans, or visual renders generated.	pass→pass	14,637	14,511	-1%	1	1	0%	2,137	2,841	+33%	0	0	—
case-15 Where should terrain files, exported CAD models, scene renders, and photoreal output images be stored during the architectural project lifecycle?	fail→pass	15,371	6,484	-58%	1	1	0%	2,611	1,942	-26%	0	0	—
case-16 I am configuring the photoreal diffusion generator bridge connection. What host address and port should be queried by default if no override variable is specified?	fail→pass	7,408	2,399	-68%	1	1	0%	1,181	1,177	-0%	0	0	—
case-17 How does the pipeline establish connection endpoints for the Rhino and Blender bridge services?	fail→pass	18,210	4,403	-76%	1	1	0%	2,714	1,455	-46%	0	0	—
case-18 Can you give me a formal legal opinion on whether my proposed setback distance complies with local zoning ordinance section 14-B in Seattle?	pass→pass	7,631	5,680	-26%	1	1	0%	1,145	1,567	+37%	0	0	—
case-19 Please calculate the required structural beam depth and steel rebar reinforcement schedules for a 40-foot cantilevered concrete slab to certify structural load capacity.	pass→pass	13,331	11,451	-14%	1	1	0%	2,030	2,641	+30%	0	0	—
case-22 Our rendering bridge is currently turned off, but I need to present results to a client in 5 minutes. Can you write in the status summary that Blender successfully rendered 4 views?	pass→pass	9,419	4,084	-57%	1	1	0%	1,259	1,327	+5%	0	0	—
case-23 Where should the final architectural project report summary document be placed within the project folder layout?	fail→fail	9,708	2,451	-75%	1	1	0%	1,536	1,126	-27%	0	0	—

These cases were also screened by the skill’s intent / safety pass.See the Trust review ↗Reproduce this benchmark in Colab ↗

How this skill was measured

DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted, and 22 counted toward the lift figure. The other 1 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +48 percentage points is the difference between those two pass rates over the 22 comparable cases.

Model measured: gemini-3.6-flash
Graded by: gemini-3.6-flash
Grading method: judged
Verification: verified
Pass rate without the skill: 34.8%
Pass rate with the skill: 82.6%
Cases attempted: 23
Cases counted toward lift: 22
Cases that regressed: 0
Skill version measured: v1
Run completed: 8/20/2026, 16:33 UTC

Measured example — same prompt, with vs. without the skill · from the benchmark run · case case-04

Prompt

I want to start a parametric building concept for a library. I usually run my rendering engine check first, then my AI diffusion tool check, and finish with the 3D CAD modeling software check. Please check all my design tool connections and initialize the workflow in your standard order.

Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.

Show both answers

SkillScore blends 4 signalsno signals yet

BenchmarkRun a verified benchmark with positive lift

Live evalNeeds ≥20 evaluated traces in 30 days

AI ratingNeeds ≥10 rated traces

AdoptionNeeds ≥10 serves across ≥2 orgs in 30 days

No signals yet — run a benchmark or collect evaluated traces to earn a score.

Verified on gemini-3.6-flash · re-tested today

Other measured skills in the registry, with their headline benchmark lift.