fix benchmark startup and cleanup checks

[kevinjosethomas]

The new prompt bar changed the text that our startup benchmark waits for, so both main and the Bun PRs time out even after the application is ready. Some runs also lose a sample when a session exits between being listed and stopped.

This recognizes both prompt layouts and still checks that typing works without submitting a prompt. Cleanup accepts an already-exited session only when the stop command reports that exact missing session and a fresh list confirms it is gone. Other stop failures still fail the benchmark.

Tested both prompt layouts and reproduced the cleanup race in Linux sandboxes. With the fixes, the full Node.js-versus-Bun benchmark passed all measurements: three installs and ten startup/runtime trials per side, with no failures. All 44 benchmark tests and repository checks also pass. This must land on main before GitHub's trusted benchmark runs can use it.

ENG-6119

[github-actions]

Prime Agent performance — failed

PR 88e71260 compared with main 5c141697.

Overall: 0 regressed · 0 improved · 2 no clear change · 15 unavailable.

Metric	Main	This PR	Change
Cold startup	—	—	—
Warm startup	—	—	—
Installation	26.79 s	26.02 s	≈ -0.78 s (-2.90%)
Compressed release artifacts	11.16 MB	11.16 MB	≈ -0.00026 MB (-0.00%)
Installed footprint	—	—	—
Idle memory, summed RSS	—	—	—

Python runtime

Metric	Main	This PR	Change
Python kernel startup	—	—	—
Python cell round trip	—	—	—
Empty bash command	—	—	—
Bash git status	—	—	—
Bash 32 KiB output	—	—	—
35 cells / 9 shell calls	—	—	—
Python interrupt to done	—	—	—
Python state snapshot	—	—	—
Python state restore	—	—	—
Python idle RSS	—	—	—
Python RSS after pandas workload	—	—	—

Sandbox cost: ~$0.1944 — no inference calls.
Run, logs, and downloadable raw results

Methodology and samples

Main resolved at 2026-09-12T01:36:31.324376+00:00. Harness 5c141697.
Linux x64, 4 vCPU, 8 GB RAM, 20 GB disk; region us.
Image: node:24-bookworm@sha256:be23f54a88d34e8824c741b19b91064094f92c1c97b194144bfc8b50d67258e2.
Stock tools, skills, daemon, and Python bootstrap enabled; fresh homes and a fixed Git fixture.
Onboarding is dismissed; the editor starts without a selected model or submitted prompt.
Medians shown. Arrows require a 20% timing/memory change plus absolute floors and IQR.
These practical noise floors are not a statistical significance test.
Cold means stopped Prime processes; OS filesystem caches are not flushed.
No model requests or credentials. Installation excludes build/setup time.
Installer tarballs use loopback; npm/Python downloads use the network with fresh caches.
Artifact size counts release tarballs; footprint after first use includes registry packages.
MB is decimal. Summed RSS can double-count shared pages; PSS is recorded when available.
Provisioning, setup, and build durations are recorded separately in the raw results.
Kernel probes use the installed JSONL runtime, outside the TUI/TypeScript host.
Per trial: 50 Python cells, 5 calls per shell case, and one 35-cell mix (9 git status calls).
Cell/shell values are batch means; other runtime timings are single operations.
State fixture: a 10,000-row × 8-column integer DataFrame and a 10,000-integer list.
Restore runs in a fresh kernel, including pandas imports; kernel startup is excluded.
Kernel RSS covers the isolated Python process; loaded RSS follows the pandas workload.
Costs estimate full sandbox lifetimes at configured rates, including setup and build.
Budget target: $1; not a billing cap. Checks are informational.

Metric	Main successful/attempted	PR successful/attempted	Main spread	PR spread
Cold startup	0/7	0/7	—	—
Warm startup	0/7	0/7	—	—
Installation	3/3	3/3	range 1.53 s	range 1.19 s
Compressed release artifacts	1/1	1/1	—	—
Installed footprint	0/0	0/0	—	—
Idle memory, summed RSS	0/0	0/0	—	—
Python kernel startup	0/0	0/0	—	—
Python cell round trip	0/0	0/0	—	—
Empty bash command	0/0	0/0	—	—
Bash git status	0/0	0/0	—	—
Bash 32 KiB output	0/0	0/0	—	—
35 cells / 9 shell calls	0/0	0/0	—	—
Python interrupt to done	0/0	0/0	—	—
Python state snapshot	0/0	0/0	—	—
Python state restore	0/0	0/0	—	—
Python idle RSS	0/0	0/0	—	—
Python RSS after pandas workload	0/0	0/0	—	—

Failures:

• Benchmark exceeded its runtime limit
• main cold trial 0: TimeoutError: Timed out waiting for the expected terminal state
• main cold trial 1: TimeoutError: Timed out waiting for the expected terminal state
• main cold trial 2: TimeoutError: Timed out waiting for the expected terminal state
• main cold trial 3: TimeoutError: Timed out waiting for the expected terminal state
• main cold trial 4: TimeoutError: Timed out waiting for the expected terminal state
• main cold trial 5: TimeoutError: Timed out waiting for the expected terminal state
• main cold trial 6: TimeoutError: Timed out waiting for the expected terminal state
• main warm trial 0: TimeoutError: Timed out waiting for the expected terminal state
• main warm trial 1: TimeoutError: Timed out waiting for the expected terminal state
• main warm trial 2: TimeoutError: Timed out waiting for the expected terminal state
• main warm trial 3: TimeoutError: Timed out waiting for the expected terminal state
• main warm trial 4: TimeoutError: Timed out waiting for the expected terminal state
• main warm trial 5: TimeoutError: Timed out waiting for the expected terminal state
• main warm trial 6: TimeoutError: Timed out waiting for the expected terminal state
• PR cold trial 0: TimeoutError: Timed out waiting for the expected terminal state
• PR cold trial 1: TimeoutError: Timed out waiting for the expected terminal state
• PR cold trial 2: TimeoutError: Timed out waiting for the expected terminal state
• PR cold trial 3: TimeoutError: Timed out waiting for the expected terminal state
• PR cold trial 4: TimeoutError: Timed out waiting for the expected terminal state
• PR cold trial 5: TimeoutError: Timed out waiting for the expected terminal state
• PR cold trial 6: TimeoutError: Timed out waiting for the expected terminal state
• PR warm trial 0: TimeoutError: Timed out waiting for the expected terminal state
• PR warm trial 1: TimeoutError: Timed out waiting for the expected terminal state
• PR warm trial 2: TimeoutError: Timed out waiting for the expected terminal state
• PR warm trial 3: TimeoutError: Timed out waiting for the expected terminal state
• PR warm trial 4: TimeoutError: Timed out waiting for the expected terminal state
• PR warm trial 5: TimeoutError: Timed out waiting for the expected terminal state
• PR warm trial 6: TimeoutError: Timed out waiting for the expected terminal state