Summary

Adds focused regression coverage for residual high-risk seams on main beyond open coverage PRs #298–#301. No production behavior changes.

Risky behavior now covered

• Codex creation preflight runtime gate — missing managed Codex app-server blocks with installCodexRuntime; available app-server reports runtime_ready: ok.
• Provider-catalog load failure still appends runtime checks — catalog outages cannot hide Codex/runtime readiness (including the Codex repair action).
• Concurrent startup bootstrap failure aggregation — independent runtime + default-skills setup failures compose into startup diagnostics; single-leg failure still surfaces without hanging readiness.
• Equal duplicate ask responses on the live socket — identical retries replay serverRequest/resolved instead of raising interaction_response_conflict; conflicting answers still conflict.
• MCP workspace cache reload after throwing stale close() — config-change reload logs the stale-close failure, drops the old entry, and loads the new tool set.

Test files added/updated

• test/creation-preflight.test.ts
• test/server.test.ts
• test/jsonrpc/flow.prompts.test.ts
• test/mcp.cache.test.ts
• apps/desktop/test/research-readiness.test.tsx (CI stabilizer: desktop bridge + persistence reset)
• apps/mobile/bun.lock (canary Bun lockfile refresh)

Why these tests materially reduce regression risk

These paths gate session creation (Codex readiness), first-connection startup diagnostics, interactive turn recovery after double-submit/reconnect races, and MCP tool correctness after config edits. Regressions here block valid Codex sessions, leave clients with ambiguous startup state, stick turns on equal retries, or keep stale MCP tools attached after reload.

Deliberately avoided duplicating open coverage PRs #298–#301 (all green at audit time).

Validation

• Targeted suites green
• Touched files Biome-clean (biome check)
• bun run lint, bun run typecheck, bun run docs:check clean
• Full bun run test: all 672 files passed (Bun 1.4.0-canary)