refactor(provider-cursor): follow the Effect service conventions

[juliusmarminge]

Part of the provider-package audit. Every module in provider-cursor should now follow docs/internals/effect-services.md.

Module state becomes service-owned

• The shell-spawn unhandledRejection guard used to be installed globally behind a module-level flag the first time sdk.ts was imported.
  • It now belongs to a CursorSdk service. Its layer adds the listener with Effect.acquireRelease and removes it on release.
  • The "another handler owns the decision" logic and the stderr output are unchanged.
  • The runner and text generation reach Agent and createAgentPlatform through this service, so the guard is armed whenever they are in use.
• The server provides CursorSdk.layer once, next to the runner layer in ProviderOrchestrationAdapterInfrastructure. The guard is now armed when that layer builds instead of on first import, and there is still exactly one guard while the server runs.
• The sandbox-support warm-up used to be a module-level memoized promise. It's now a Ref and Deferred owned by the runner. It still warms at most once, later callers wait for that attempt, and failures are ignored.

Conventions

• CursorSdkCatalogShape and CursorAgentSdkRunnerShape are inlined into their services.
• effect/http modules are imported as subpath namespaces.
• makeCursorAgentSdkRunner is an Effect.fn that yields CursorSdk.
• The snapshot error's detail no longer copies cause.message.
• An unused default adapter layer is removed.

Left for a follow-up: keychainToken.ts still memoizes the macOS keychain read in module state, and its timeout error extends Error. Its main consumer, the Promise-based Cursor usage reader in apps/server, moves into the package with the usage-reader work, so it'll be converted there.

Tests: provider-cursor 102 passed. Server ProviderInstanceRegistry 8 passed. Typecheck is clean for provider-cursor and apps/server.

Model: Claude Opus 5.5 via Claude Code in T3 Code.

🤖 Generated with Claude Code

---

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This refactor moves Cursor process state and its unhandled-rejection guard into a scoped Effect service, changing production layer wiring and listener lifetime. It also adds a file-level globalConsole:off static-analysis suppression, which requires human review under the configured criteria.

^{You can add or adjust custom eligibility rules. Learn more.}

[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	5.0 KiB	5.0 KiB	0 B (0.0%)	6.8 KiB	✅
Codex	Thread snapshot wire	3.8 KiB	3.8 KiB	0 B (0.0%)	4.9 KiB	✅
Codex	Live turn WebSocket wire	1.2 KiB	1.2 KiB	0 B (0.0%)	2.0 KiB	✅
Codex	Live turn WebSocket decoded	20.9 KiB	20.9 KiB	0 B (0.0%)	29.3 KiB	✅
Codex	Live turn messages	2	2	0 (0.0%)	8	✅
Claude	Total thread wire	5.0 KiB	5.0 KiB	−24 B (−0.5%)	6.8 KiB	✅
Claude	Thread snapshot wire	3.8 KiB	3.8 KiB	0 B (0.0%)	4.9 KiB	✅
Claude	Live turn WebSocket wire	1.2 KiB	1.2 KiB	−24 B (−2.0%)	2.0 KiB	✅
Claude	Live turn WebSocket decoded	21.2 KiB	21.2 KiB	−41 B (−0.2%)	29.3 KiB	✅
Claude	Live turn messages	2	1	−1 (−50.0%)	8	✅

Baseline: f853d51 · PR result: 2c41887 · 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: 108.5 KiB
• Claude decoded thread snapshot: 108.8 KiB

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