feat(web): Prism settings in the Providers layout, lanes only for Worker

[lukemaj]

This closes #26. The Prism settings page showed a 5x3 grid of lanes, and most of those lanes were never used. The page now uses the Providers layout. On the left is a list of roles, each with an icon, a one-line description and a toggle. On the right is a detail pane with one ordered model list for the selected role. Every change saves on its own, so there are no Save buttons.

Changes

• Roles: the page shows Dispatcher, Worker, Reviewer, Retry (key correction) and Escalation (key recovery). Planner is gone from the page. Only Retry and Escalation can be switched off. The other roles show a locked toggle with the reason in a tooltip and in the detail pane: Dispatcher is needed until router Wight mode: keep a thread working on continue #117 decides, Reviewer because independent review is required, Worker because it is required.
• Lists: Only Worker has easy/medium/hard, shown as tabs. Every other role has one list: model plus effort, drag handle or arrows to reorder, first entry primary, the rest fallbacks.
• Settings schema: a role other than Worker now stores models, and correction/recovery add enabled (default true). Saved per-lane lists still load: the medium list becomes the single list, and the next save writes models.
• Snapshot compatibility with router feat(server): continue idle threads in Wight mode #127: /api/prism/snapshot still sends lanes.easy/medium/hard for every role. A role other than Worker repeats its single list in all three lanes and also sends it as models. Only correction and recovery carry enabled.
• spawn_thread: accepts retry and escalation, and still accepts correction and recovery. A role other than Worker uses its single list and ignores lane.
• Agents panel: each Prism job row is now a heading with no status dot (Job <id> · <status> <route>), so it no longer looks like a second agent.

Proof

• Tests: vp test run passed for packages/contracts/src/prism.test.ts (6), for the server tests roles.test.ts and snapshotRoute.test.ts (21), and for the web tests PrismSettings.logic/state.test.ts and AgentThreadTree.logic.test.ts (21).
• Typecheck: tsc --noEmit passed for contracts, server and web.
• Lint: vp lint is clean on the changed files, and scripts/fork-check.sh reports OK.
• Live run: I ran an isolated dev instance on the worktree's own .t3 state, seeded with old per-lane lists. The page migrated them. Turning Retry off saved correction.enabled: false. GET /api/prism/snapshot returned three lanes for every role, models on every role except Worker, and enabled only on correction (false) and recovery (true).

Screenshots were taken with Playwright's bundled Chromium (headless shell 1243) against that instance. They are cropped to the roles section.

Page (Dispatcher selected, migrated list):

Worker lanes (Hard tab):

Retry switched off:

Not captured: the Agents panel heading. It needs a live Prism job, and the isolated instance has none.

Done with Claude Opus 5.5 in Claude Code.

🤖 Generated with Claude Code

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.5 KiB	13.5 KiB	+23 B (+0.2%)	15.1 KiB	✅
Codex	Thread snapshot wire	7.1 KiB	7.1 KiB	+1 B (+0.0%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.4 KiB	6.5 KiB	+22 B (+0.3%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	56.2 KiB	56.3 KiB	+44 B (+0.1%)	66.4 KiB	✅
Codex	Live turn messages	9	10	+1 (+11.1%)	21	✅
Claude	Total thread wire	13.5 KiB	13.5 KiB	−13 B (−0.1%)	15.1 KiB	✅
Claude	Thread snapshot wire	7.1 KiB	7.1 KiB	+1 B (+0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.5 KiB	6.4 KiB	−14 B (−0.2%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	57.0 KiB	57.0 KiB	0 B (0.0%)	66.4 KiB	✅
Claude	Live turn messages	9	9	0 (0.0%)	21	✅

Baseline: 13b7a4f · PR result: f092bd5 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 113.9 KiB
• Claude decoded thread snapshot: 114.6 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.