fix(prime-agent): explain why a thread cannot switch to Prime Agent Default

[rynfar]

Problem

Choosing Prime Agent Default for a thread that had already run on a named model failed the turn with Model must use a provider/model selector. — an internal-sounding message for an ordinary action in the model picker.

Pylon uses "default" as the sentinel for "let Prime choose". At session start it is handled (model === "default" simply omits the model), but on a later turn applyTurnSelection compares it against context.session.model, sees a difference, and passes it to setModel, where splitModelSelector rejects it for having no /.

Found while adopting 0.7.3 (#48); it is not caused by that release.

How it was fixed

The turn cannot succeed either way. "default" means deferring to Prime's own configured model, and the daemon connection exposes no way to hand model choice back to Prime inside a running session: AgentConnectionModelCatalog is { models, configuredProviders } with no default to re-select, and setModel requires an explicit provider and id. Silently continuing on the previous model would leave the thread running one model behind a Prime Agent Default label, which is exactly the kind of stale label we try not to ship.

So the switch is rejected where the other unsupported turn selections already are, with a message that names the constraint and the way forward. This changes an incomprehensible failure into a comprehensible one rather than turning a working path into a failing one.

Unaffected: naming a different model in an existing thread, and threads that stay on Prime Agent Default (which make no setModel call at all — covered by a second test so the guard cannot over-reject).

A client-side improvement would be to stop offering Prime Agent Default in the picker once a thread has run on a named model. That spans web and mobile and is a separate concern, so it is not in this PR.

Verification

• Reproduced first: the new test failed with expected [ 'default' ] to not include 'default' against the unfixed adapter, confirming setModel("default") was really being called
• vp test run apps/server/src/provider/prime/ — 245 passed, 1 skipped (up from 243; the two new tests are the delta)
• vp run -F t3 typecheck — 0 errors
• vp lint on both touched source files — clean

No UI changes, so no before/after images.

Model: Claude Opus 5 in Pylon (Claude Code harness)

---

^{Need help on this PR? Tag @codesmith with what you need. Autofix is disabled.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

ℹ️ No successful main baseline artifact is available yet. This run establishes the initial measurement.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	—	12.6 KiB	—	15.1 KiB	✅
Codex	Thread snapshot wire	—	6.3 KiB	—	7.3 KiB	✅
Codex	Live turn WebSocket wire	—	6.3 KiB	—	7.8 KiB	✅
Codex	Live turn WebSocket decoded	—	51.8 KiB	—	66.4 KiB	✅
Codex	Live turn messages	—	16	—	21	✅
Claude	Total thread wire	—	12.6 KiB	—	15.1 KiB	✅
Claude	Thread snapshot wire	—	6.3 KiB	—	7.3 KiB	✅
Claude	Live turn WebSocket wire	—	6.3 KiB	—	7.8 KiB	✅
Claude	Live turn WebSocket decoded	—	52.7 KiB	—	66.4 KiB	✅
Claude	Live turn messages	—	16	—	21	✅

Baseline: unavailable · PR result: 50b4f25 · 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: 99.8 KiB
• Claude decoded thread snapshot: 100.6 KiB

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