1.9.0: Scenarios — rename Catalogs, edit in the dashboard, generate from a project's browser, export and import

[TonsOfFun]

Release: 1.9.0. Scenario catalogs are new in 1.9, so the rename and dashboard management ship with them. No browser-sidecar/ changes, and nothing depends on the live view's file upload (#588, 1.9.1); the Generate tab uses BrowserPanel as #584 has it.

Stacked on #584 (base feat/claude-code-boot-setup), rebased onto its reworked tip without the upload commit, so this diff shows only the Scenarios work. Retarget to main once #584 merges. Green CI also needs #587 (ruby_llm 2.1.1 breaks two eval tests on main).

What changes

• Rename: "Catalogs" is now Scenarios in the sidebar, page and copy, at /scenarios, and /catalogs redirects there. The database, the API and the YAML catalog: key are unchanged, so existing documents still import.

• Editing in the dashboard: create an empty collection, then add, edit and delete products, sets and scenarios in place. A scenario has a prompt, notes as the rubric, expected tools, contains / never contains, tags, enabled and production-only. Every change is checked the same way an import checks a document, and the collection's YAML is rewritten.

• Generate tab:

  • Pick a project. The page shows the project's explorer controls and the sandbox's live browser.
  • The proposals are the same records as the project's Explorations tab, accepted into a scenario set instead of the evaluation.
  • A missing provider key or an unbooted sandbox is flagged up front, and a project's collection can be created in one click.

• Export: the whole collection, or any checked products, sets and scenarios, as a YAML download or copied to the clipboard.

• Import:

  • From a paste, an uploaded file, or a repository's .activeagents/evals.
  • Preview lists by key what will be created, overwritten, removed or kept before anything is written.
  • "Keep what the document leaves out" is on by default in the UI, so importing an exported selection only updates those entries.

• Accessibility (e0359309): the scenario form's prompt and notes boxes (and every Field/Select) tie their label to the control by id and carry an aria-label. The live test found screen readers reading them as an unnamed "textbox".

Live test (integration build of #584 + #586)

Import preview and apply, adding a scenario, the export round trip and the Generate tab UI all passed. The unnamed text boxes were the only finding, fixed above.

Tests (Ruby 4.0.6), after the rebase

• actionagent: 1695 runs, 0 failures, 7 skips.
• frontend: 599 pass.
• browser-sidecar: 128 pass (unchanged).
• rubocop clean; the bundle is rebuilt.

Open questions

1. Claude Code as a generator isn't built: a Claude Code session driving the sandbox browser and submitting proposals, so no provider key is needed. That's proposed as the next branch.
2. Should the API default to merging on import, like the UI does?
3. Should the API be renamed to match "Scenarios", with the old routes kept as aliases?
4. Should the generator offer "accept and run set"?
5. On the local backend, Run set still needs the first-boot "Run it on this machine" confirmation from the project page.

Review note with how to test: activeagents docs/features/scenarios-dashboard.md.

🤖 Generated with Claude Code

https://claude.ai/code/session_01CJAKwnkBaCGr5qjCcmc3kh