cli: ui:add solves the exact problem agents hand-roll around, and is invisible at the moment of need

[rickylabs]

Summary

netscript ui:add page/island --query scaffolds the exact page + island + query-loader triad that
both wave-four agents assembled by hand. Neither ran it. ui:add data-table would have given
one of them the settled-incidents table it hand-wrote.

Run 1 lists it under "what I should have used and did not":

"netscript ui:add page/island --query scaffolds the page + island + query-loader triad I
assembled by hand, and ui:add data-table would have given the settled list a real table."

The problem

The command exists, is documented, and is invisible at the moment of need. An agent about to write a
data screen has no prompt to consider it — it is not surfaced by the scaffold, not named in the
generated app, and not reachable from the point where the decision is made.

This is the CLI mirror of the docs-sequencing problem: capability present, activation absent.

Proposal

• Name ui:add in the generated app-scoped conventions file (see the app-scoped AGENTS.md issue),
  at the point where "do not hand-write Button/Input/Card equivalents" is stated.
• Consider a hint from adjacent commands: after netscript service generate, or on first run of a
  route-creating command, mention the ui:add triad once.
• Ensure ui:add --help states what the triad is — page, island, query loader — rather than
  listing flags, so an agent that runs --help once learns the shape.

Acceptance

• ui:add is named in generated app conventions where hand-rolling is discouraged
• ui:add --help describes the generated triad and when to use it
• A follow-up agent run building a data screen either runs ui:add or records why it did not — relocated to verify(wave-five): does the shipped agent surface actually change agent behaviour? #1090 (0.0.5). This is an observation of a future agent run and
  cannot be satisfied by the PR that builds the capability. Implementation landed in feat(cli,mcp): scaffold agent surface — app-scoped conventions and an evidence-gated diagnostic surface #1078.

Notes

Wave-four run-1 self-review (https://github.com/rickylabs/vigil/blob/main/self-review.md).

[rickylabs]

Partially addressed by #1078 — deliberately NOT auto-closed.

Implementation is complete and its acceptance boxes are ticked with evidence in that PR: ui:add named in the generated app conventions where hand-rolling is discouraged, and ui:add --help describing the page + island + query-loader triad and when to use it.

Remaining scope is exactly one box:

A follow-up agent run building a data screen either runs ui:add or records why it did not

That criterion is an observation of a future agent run. It cannot be established from inside the PR
that creates the capability — the PR is the thing the follow-up run would be exercising. So #1078
references this issue without a closing keyword rather than claiming it, and this issue should stay
open until a wave-five run exercises the new surface and the observation is recorded here.

Whoever runs that: the check is whether the agent reaches for the capability from the symptom,
without being told to in its brief. If it does not, record why not — a "did not use it" with a reason
is a more useful result for this issue than a retry that nudges the agent into using it.