fix(supervise): settle a root that never spawned as no-children-spawned

[drewstone]

Problem

A root that ran to completion under budget, selected nothing, and never spawned a child settled no-winner / all-children-down with downCount: 0. That reads as a fleet failure. Fifteen sandbox-placed directors settled that way this week (#1245) while the actual fault was that the root did its research in bash and never recursed — and every reader, including me, went looking at the fleet first. Today's sandbox-mcp-probe-20260916d, which called stop deliberately after spawning nothing, got the same label.

types.ts already acknowledged the ambiguity when driver-failed was carved out; this carves out the other half.

Change

• New reason no-children-spawned on the no-winner lifecycle arm, chosen in the residual bucket when fleetYield.spawned === 0.
• all-children-down now asserts what its name says: at least one child was spawned and none delivered.
• Two tests that pinned the old value on a zero-spawn tree flip; one was about token telemetry and never cared which reason it got.

Release

0.236.0 — the exported reason union widens. A consumer that switched on the reason and treated the old value as "the fleet lost everything" adds the new arm; a consumer that only counted no-winner is unaffected.

Proof, local

tsc --noEmit clean; biome clean; vitest run on supervisor.test.ts, tests/kernel/supervise.test.ts, tests/mcp/delegate.test.ts, tests/kernel/supervise-full-profile-bridge.test.ts, tests/kernel/retained-environment-release.test.ts: 204/204. Build, api:surface, docs freshness pass. Full suite running.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, add credits to your account and enable them for code reviews in your settings.