feat: add the tilemap-palette-gridpaintingstate skill

[ziyiunity]

Adds tilemap-palette-gridpaintingstate — reads and manages the live Tile Palette painting state (active brush, active target, palette). 2 files.

Why any of this changed

This skill was written for Unity AI Assistant, where a built-in tool ran C# inside the Editor and handed the result back to the model. That tool does not exist in a third-party coding agent, so the skill is re-pointed at the Unity CLI: unity command eval against the project's com.unity.pipeline package.

It is not a rename. The execution contract differs. AI Assistant injected a result object the script reported through; the CLI returns a value on stdout. So every reporting site needed a decision — is this operator-facing (a console log), or does the model have to read it back (a returned value)? Abort paths became throw, so a failure is loud rather than a log line nobody reads.

Prerequisites are not restated here. They are delegated to the unity-cli skill, which owns installing the CLI, confirming a connected Editor, adding the Pipeline package, and telling a genuinely absent Editor apart from one stuck in Safe Mode after a compile error.

The description is byte-identical to the original, and the domain logic is untouched.

The README skill list is deliberately not touched, so these skills stay independently reviewable and mergeable in any order. It gets one entry per skill in a follow-up.

What changed in this skill specifically

• Two RunCommand instructions became the CLI equivalent.
• Added a prerequisites section delegating to unity-cli. This skill needed the clearest statement of why an Editor is mandatory: GridPaintingState and GridPaintPaletteWindow describe the state of an open window, and nothing on disk holds it — so there is no file-reading fallback at all.
• These types are internal to UnityEditor.Tilemaps. Editor-side eval reaches them (measured), so the skill says to use them directly and fall back to reflection only if a compile error actually reports a visibility problem.
• Fixed a malformed link: **API Reference** : [references/references.md] had no target, so it rendered as literal text rather than a link to the file.
• Generate standalone RunCommand scripts only became Generate standalone snippets only — no MenuItem, no AssetPostprocessor, keeping the original intent (don't wrap the work in Editor plumbing) in wording that means something here.

Tested

End-to-end against a Unity 6 project with the Tile Palette window open, asked in natural language which palette and brush were active. No errors, and the internal types resolved without needing reflection.

[ziyiunity]

One thing I should have said when I opened this, prompted by the review on #9.

I've been asking each of you whether the content is still accurate — but the more valuable question is upstream of that. If your read is that this skill shouldn't exist any more, that's the most useful answer you can give me, not the least. Models have moved since these were written, and a skill that no longer beats what a capable model does unprompted is worse than absent: its description sits in context every session whether it fires or not, and when it does fire it can steer toward a more rigid workflow than the model would have chosen.

That's what happened on #9 — the owner's read was that current models do it better without the skill, and we're closing it on that basis. No hard feelings if the same applies here.

I can tell you that a skill compiles, drives the Editor correctly, and produces the right end state — I verified all of that against a live Unity 6 Editor. What I can't tell is whether it makes the outcome better than not having it. That part only you can answer.

[ChuanXin-Unity]

This particular skill may not be needed anymore.

I have tried cloning and running the skill as per the README into Claude Code. Using Opus 5 and Unity CLI, it was able to do the tasks I have set with and without the skill (more steps without the skill, but still successful). I guess the result may be better if Fable was used?

[ziyiunity]

This particular skill may not be needed anymore.

I have tried cloning and running the skill as per the README into Claude Code. Using Opus 5 and Unity CLI, it was able to do the tasks I have set with and without the skill (more steps without the skill, but still successful). I guess the result may be better if Fable was used?

Closing this — your A/B is the answer, and thank you for actually running it rather than just reading the diff.

On your Fable question, I'm not going to hold the batch on it, but if the Fable numbers come back clearly different I'll reopen with them attached. Thanks!

[ziyiunity]

Closing this — your A/B is the answer, and thank you for actually running it rather than reading the diff.

Both arms completing the task means the skill isn't buying correctness, and "more steps without it" isn't a strong enough return to justify shipping and maintaining it. The same thing came back on another skill in this batch, which we also dropped, so this is a pattern rather than a one-off.

On your Fable question — you're right that the gap should widen on a smaller model, and that's the version of the test that would actually justify keeping this. I'm not holding the batch on it, but if those numbers come back clearly different I'll reopen with them attached.

The tilemap skills that did earn their place are in Unity-Technologies/skills#42, which you've already approved.