Add an example
Workflow
- Goal
- a new example in
examples/shelf/with a checked Steel preset and a preview, following the collection's conventions. - Use when
- adding a scene to the collection, or giving an existing example a Steel preset it does not have yet.
- Done when
shelf check EXAMPLE --everyreportsworksfor every filled table, the example renders and looks right, andPRESETS.mdis regenerated.
Any coding agent, or a person, can follow this workflow from agents/workflows/add-an-example/README.md. Tools that discover skill folders also find it assteel-add-an-example.
Inputs. The scene (a Newton example script, or a new one). Read examples/AGENTS.md
for the presets.toml format, the conversion steps and the rules (never write status = "works" by hand; never
loosen an example’s test).
Steps
- Put the example under
examples/shelf/<NN_section>/<example>/asexample.pypluspresets.toml, converted asexamples/AGENTS.mddescribes (settings out of code into presets,self.preset = presets.load(__file__)). Fill the tables of solvers the code already supports with the values it used; leave othersuntested. - For Steel: run-trial at the new
[steel]table, then autotune-settings fromdefaultsif the table is new. Copy the result into[steel]with a comment per setting saying why. - check-example with
--every. - render-and-grade for this example and visual-review.
- Add the
[gallery]table and render the preview:uv run python -m shelf thumbs EXAMPLE --force.
Deliverables
The example folder, its checked presets.toml, the regenerated PRESETS.md, and the gallery for it.
Steps it uses
- Run a trial runs one example's whole episode with Steel at given settings and lists what is wrong, in fix order.
- Autotune settings repeats trials until no fix is left, then keeps only cheaper settings that stay healthy.
- Check an example runs an example's own test headless with a preset and records the result in
presets.toml. - Render and grade renders examples with their Steel presets on one or more GPUs, records stability metrics, and grades each scene ok, suspect, unstable or failed.
- Visual review looks at frames across each episode and records a grade and a note per scene the metrics may have misjudged.
Source: agents/workflows/add-an-example/README.md, the copy agents read.
Steel docs and tutorials are licensed under Apache-2.0. Third-party notices. "Skild", "Skild AI" and the Skild logo are trademarks of Skild AI and are not licensed under Apache-2.0.