Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add ifBars/blender-agent-studio --skill blender-agent-benchmarkgit clone --depth 1 https://github.com/ifBars/blender-agent-studioWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/ifbars/blender-agent-studio/blender-agent-benchmark)<a href="https://agentmods.dev/skills/ifbars/blender-agent-studio/blender-agent-benchmark"><img src="https://agentmods.dev/badge/skills/ifbars/blender-agent-studio/blender-agent-benchmark.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00068 | $0.01747 |
| Opus 5 | $0.00034 | $0.00873 |
| Sonnet 5 | $0.00014 | $0.00349 |
| Haiku 4.5 | $0.00007 | $0.00175 |
Grade A, and why
blender-agent-benchmark scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured today.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 175 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Blender Agent Benchmark
Read the shared execution guidance once per task for autonomous decisions, evidence cadence, and long-task continuity.
Measure changes with the same tasks, model, effort, limits, Blender build, and evaluator. Preserve natural agent behavior.
Protect benchmark integrity
- Create isolated directories for every condition and repetition.
- Do not leave the other condition's code, renders, metrics, or expected fixes where the agent can discover them.
- Keep the user-facing task prompt identical except for explicit skill invocation in the plugin condition.
- Use
codex exec --ignore-user-configfor the no-plugin baseline. - Use the installed plugin in a fresh invocation for the plugin condition.
- Record CLI version, model, effort, Blender build, duration, tool calls, failures, and output hashes.
- Evaluate outputs after generation. Do not leak hidden rubric details to the agent.
Read references/methodology.md before changing fixtures, scoring, or comparison claims. Read references/open-source-benchmark-landscape.md when designing new suites or borrowing evaluation ideas from other Blender benchmarks. Read references/validated-results.md only when reviewing the plugin's recorded validation result, not while generating a benchmark submission.
Run the suites
For Astra, pass --profile astra (model gpt-6-astra, effort medium). The
sol, terra, and luna profiles select their corresponding GPT-5.6 models
at the same effort. Explicit --reasoning overrides effort, not the model.
Use an explicit profile or --model for reproducible comparisons; runs using
configured default are exploratory because model identity is not pinned.
Separate two experiments: old/revised skills on Astra, then fixed revised skills on Astra/Sol/Terra/Luna. Keep fixture, effective effort, Blender build, limits, permissions, and evaluator fixed. The existing non-regression gate requires matching models; cross-model results are descriptive model comparisons, not proof that a skill revision improved. Never replace historical result labels with Astra or attribute a simultaneous model-and-skill change to either alone.
What ships with it
22 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
- agents/openai.yaml 302 B
- references/astra-0.6-validation.md 4.1 KB
- references/astra-workflow.md 7.1 KB
- references/methodology.md 6.0 KB
- references/open-source-benchmark-landscape.md 7.8 KB
- references/validated-results.md 3.8 KB
- scripts/benchmark_mcp.ts 3.9 KB runs code
- scripts/compare_runs.test.ts 2.7 KB runs code
- scripts/compare_runs.ts 29 KB runs code
- scripts/model-options.ts 1.5 KB runs code
- scripts/rescore_run.ts 2.8 KB runs code
- scripts/run_benchmark.test.ts 4.2 KB runs code
- scripts/run_benchmark.ts 20 KB runs code
- scripts/run_vrchat_hair_benchmark.ts 6.3 KB runs code
- scripts/score.test.ts 7.9 KB runs code
- scripts/score.ts 15 KB runs code
- scripts/tasks.test.ts 3.0 KB runs code
- scripts/tasks.ts 54 KB runs code
- scripts/trace.test.ts 1.2 KB runs code
- scripts/trace.ts 2.8 KB runs code
- scripts/verified_score.test.ts 2.0 KB runs code
- scripts/verified_score.ts 3.1 KB runs code
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- today Changed · +16 lines b136b4a63488
- 8d ago First seen · 159 lines · 68 tokens per session scan A bc3909ccb51f
blender-agent-benchmark is a skill published in the GitHub repository ifBars/blender-agent-studio (14 stars, last pushed today), licensed MIT. It adds 68 tokens to every session and 1,747 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
blender-validation
Blender scene, mesh, material, animation, and export readiness validation.
nemo-evaluator-sdk
Evaluates LLMs across 100+ benchmarks from 18+ harnesses (MMLU, HumanEval, GSM8K, safety, VLM) with multi-backend execution. Use when needing scalable evaluation on local Docker, Slurm HPC, or cloud platforms. NVIDIA's enterprise-grade platform with container-first architecture for reproducible benchmarking.
testing-perf
Performance and load testing patterns — k6 load tests, Locust stress tests, pytest execution optimization (xdist parallel, plugins), test type classification, and performance benchmarking. Use when writing load tests, optimizing test execution speed, or setting up pytest infrastructure.
smart-contract-testing
Guide for testing smart contracts using Foundry and Hardhat, covering unit tests, fuzz testing, invariant testing, fork testing, and gas benchmarking.
solidity-playground-guide
Guide to Lyra Web3 Playground — a browser-based Solidity IDE and smart contract playground. Write, compile, deploy, and interact with contracts directly in the browser. Supports Hardhat/Foundry compilation, multiple chains, and live contract testing.
AI Agent Evaluation
Comprehensive evaluation patterns for AI agents including multi-turn conversation testing, LLM-as-judge frameworks, benchmark suites, regression detection, and systematic eval pipelines for measuring agent quality and safety.