hvPlotAgent cannot express any color options, so hvPlot's colormap defaults are unreachable

[ghostiee-11]

hvPlotAgent builds its response model from the declared params of hvPlotUIView via param_to_pydantic (lumen/ai/agents/hvplot.py:51). hvPlotBaseView declares only kind, x, y, by, groupby and geo (lumen/views/base.py:925-970), so the generated model has no color field at all. The agent cannot pick a colormap even when the user explicitly asks for one.

That matters more than it first looks, because hvPlot already does a lot here on its own:

• _default_cmaps (hvplot/converter.py:840-845) maps linear to kbc_r, diverging to coolwarm, categorical to glasbey_category10 and cyclic to colorwheel
• _process_symmetric (hvplot/converter.py:1302-1332) auto-detects diverging data from the 5th/95th percentiles and then selects the diverging colormap (converter.py:1227-1233)
• the strings linear, diverging, categorical and cyclic are accepted as semantic cmap aliases (hvplot/converter.py:1838-1839)

None of that is reachable from Lumen today.

The one color-adjacent value Lumen does inject is cnorm="log" for large datasets (lumen/ai/agents/hvplot.py:72-74), which sets a log color norm without any colormap alongside it.

Declaring cmap plus a few related options would let the agent use the semantic aliases, which seems like a better fit for an LLM than asking it to pick from the full colormap list.

One thing worth being careful about if anyone picks this up: symmetric, colorbar, cnorm, robust and rescale_discrete_levels are all None-defaulted in hvPlot's converter so it can distinguish "unset" from "explicitly False", and each is forwarded only under if x is not None. Declaring symmetric as param.Boolean(default=False) would silently disable the auto-detection above, so these want default=None with allow_None=True.

[ghostiee-11]

Correcting two things I got wrong when I filed this.

Declaring the params would not have worked. View.__init__ puts undeclared keys into self.kwargs (lumen/views/base.py:133), and that is the only channel that reaches the renderer: hvPlotView.get_plot forwards self.kwargs plus five hardcoded params, and hvPlotUIView._get_args (views/base.py:984) filters declared params against hvPlotExplorer | Geographic, neither of which declares a colour option. Declaring cmap would have evicted it from self.kwargs and then dropped it at the whitelist. Passed as a loose kwarg it already reaches the plot today:

v = hvPlotView(pipeline=pl, kind="scatter", x="a", y="b", cmap="fire")
v.get_plot(df).opts.get().kwargs["cmap"]   # 'fire'

So the gap was only in the response model, not the plumbing.

robust is not usable. No hvplot Controls class declares it, so it raises TypeError at hvplot/ui.py:590 whichever route it takes. symmetric is best left out too: Colormapping.symmetric is param.Boolean(default=False) and raises on None, and the explorer already runs the auto-detection internally.

Separately, while working on this I found hvPlotAgent could not build a response model at all, so it raised on every request rather than merely lacking colour. #1935 covers both.