agentleFS
badgidsbadgids/comfyui-development-skills/skills/comfyui-performance/SKILL.md

comfyui-performance

Skill4 starsChanged 44 days agoMITView on GitHub

Use for ComfyUI inference performance, benchmark design, kernel behavior, VRAM/RAM use, caching, loading time, throughput, latency, or regressions involving ComfyUI core, comfy-kitchen, or benchmark tooling.

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 3 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Before claiming a performance improvement:

    line 41
  2. Distinctive:

    verify output correctness did not regress.

    line 47
  3. Distinctive:

    do not claim a speedup from code inspection alone.

    line 48

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

ComfyUI Performance

Performance claims require measurements on a defined environment.

Sources

Use target/current:

  • Comfy-Org/ComfyUI;
  • Comfy-Org/comfyui-benchmark;
  • Comfy-Org/comfy-kitchen;
  • Comfy-Org/comfy-model-tools for current model packaging or conversion behavior when relevant;
  • Comfy-Org/comfy-quants for current offline quantization formats and export behavior when relevant;
  • relevant profiling/tests and dependency versions.

Procedure

  1. Record hardware, OS, driver/runtime, Python, PyTorch, ComfyUI commit, model, precision, workflow, and warm/cold state as relevant.
  2. Establish a baseline before changing code.
  3. Change one performance variable at a time where possible.
  4. Measure repeated runs and report variance, not a single flattering result.
  5. Separate prompt/setup/load time from steady-state execution when meaningful.
  6. Track memory use alongside speed if an optimization trades one for the other.
  7. Validate output correctness after optimization.
  8. For model conversion or quantization work, verify that the target ComfyUI loader actually supports the produced artifact before making compatibility claims.

Do not

  • claim a speedup from code inspection alone;
  • compare different workflows/models/hardware as if they were equivalent;
  • hide OOMs, fallback paths, or reduced-quality settings in benchmark conclusions.

Acceptance gate

Before claiming a performance improvement:

  • record the test environment and target commit;
  • record a baseline before the change;
  • repeat the after measurement under equivalent conditions;
  • report variance and relevant memory use;
  • verify output correctness did not regress;
  • do not claim a speedup from code inspection alone.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.