fix(server): a Cursor command you stop shows as interrupted

[juliusmarminge]

When you stop a Cursor turn while a command is running, the command is saved as completed, as if it had finished. The run and provider turn are correctly interrupted.

Evidence

Seen live on the V2 head with gpt-5.4-nano. The prompt ran for i in $(seq 1 60); do echo tick $i; sleep 1; done and was stopped about 17s into the command. The provider log has a tool-call-started for the shell call, then run.cancel, then run.completed with status: "cancelled". It never has a tool-call-completed. The projection then held:

provider_turn  status=interrupted
command_execution  status=completed  input="for i in $(seq 1 60); ..."  (no output, no exit code)

Pi and Codex report a stopped command as interrupted, and Claude reports it as failed.

Fix

finalizeTurn closes any tool that never got a tool-call-completed. It used to close them as completed. Now, when the turn ends interrupted, failed, or cancelled, those tools take that status. Tools on a turn that completes normally are unchanged.

Web keeps stopped rows visible once #13790 lands. Without that PR, web hides them, the same as for Codex.

Verification

• turn_interrupt_mid_tool/cursor_output.ts now asserts commandItem.status === "interrupted". The recorded transcript already stops mid-command without a tool-call-completed. vp test run src/orchestration-v2/testkit/OrchestratorReplayFixtures.integration.test.ts -t cursor:
  • Before the fix: expected 'completed' to equal 'interrupted' (1 failed, 9 passed).
  • After: 10 passed.
• vp test run src/orchestration-v2/Adapters/CursorAdapterV2.test.ts: 11 passed.
• vp exec tsc --noEmit -p . in apps/server: no error TS or warning TS. vp run knip:check: clean.
• vp lint on the two touched files reports one warning. It is the unused layer at the end of CursorAdapterV2.ts, which is already on the base and is not touched here.
• Not run: the full server suite, and a live Cursor stop after the fix.

Model: Claude Opus 5.5 (Claude Code)

🤖 Generated with Claude Code

---

[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	—	4.9 KiB	—	6.8 KiB	✅
Codex	Thread snapshot wire	—	3.7 KiB	—	4.9 KiB	✅
Codex	Live turn WebSocket wire	—	1.1 KiB	—	2.0 KiB	✅
Codex	Live turn WebSocket decoded	—	20.4 KiB	—	29.3 KiB	✅
Codex	Live turn messages	—	1	—	8	✅
Claude	Total thread wire	—	4.9 KiB	—	6.8 KiB	✅
Claude	Thread snapshot wire	—	3.7 KiB	—	4.9 KiB	✅
Claude	Live turn WebSocket wire	—	1.2 KiB	—	2.0 KiB	✅
Claude	Live turn WebSocket decoded	—	20.8 KiB	—	29.3 KiB	✅
Claude	Live turn messages	—	2	—	8	✅

Baseline: unavailable · PR result: ebb0000 · 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: 106.1 KiB
• Claude decoded thread snapshot: 106.4 KiB

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

[macroscopeapp]

Approvability

Verdict: Approved at ebb0000

Macroscope's review found this PR approvable — This is a small, self-contained server bug fix that correctly marks unfinished Cursor commands with the enclosing turn’s terminal status. Existing completed-tool behavior is preserved, and the interrupted-mid-tool regression is explicitly tested.

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