Wizard finale hardcodes a `logs` query that fails on AI-only installs

[doppenhe]

Problem

The wizard always ends with:

hyp query sql 'select count(*) from logs'

That query is not valid for every completed setup. A machine that selected only Claude Code or Codex registers ai_gateway_messages; logs belongs to the separately enabled OTEL plugin. The first command the product recommends can therefore target a dataset the install does not have.

The onboarding feedback session reported at 05:04-05:13 that the printed starter query contained log and did not work.

Root cause

writeWalkthroughRunSummary prints the query unconditionally:

src/core/cli/walkthrough.js:638

writeWalkthroughRunSummary receives only configPath and finaleSummary. FinaleSummary records daemon, attach, skill and agent copy, restart, and backfill results, but not selected sources or registered datasets. The fix must pass source or dataset state into the summary, or resolve it from the completed config, before choosing the query.

Suggested fix

Derive the starter query from the datasets registered by the completed setup. For an AI-only install:

hyp query sql 'select count(*) from ai_gateway_messages'

An OTEL install can choose among logs, traces, or metrics. If more than one useful dataset is active, print one known-good first query and point to hyp query status for the rest.

Regression test

Run the summary path with Claude or Codex as the only selected client and assert that the printed command targets ai_gateway_messages. The test should also cover an OTEL selection so the fix does not simply replace one hardcoded dataset with another.

---

Filed from the recorded onboarding feedback session, 05:04-05:13. Root-cause refs against master @ 420cb2a.