feat(xreview-scout-codex): run on gpt-6.1-sol at high reasoning effort

[bdchatham]

Summary

The codex scout moves from gpt-5.6-sol to gpt-6.1-sol and sets executor.reasoning_effort: high. Before this, it set no effort and ran at the harness default.

Why this model

gpt-6.1-sol (released 2026-09-29) is the Codex CLI's default model as of v0.159.1, and OpenAI describes it as "near-Astra performance for complex work at a lower cost". The comment that tied the pin to openai/codex-action's old default is replaced.

Verification

• The CI bundle check, run locally with omnigent==0.9.0: ok for all 4 bundles.
• The deployed seigent server runs omnigent 0.16.0.dev0, which has ExecutorSpec.reasoning_effort. CI's pinned 0.9.0 parses the bundle but drops the field, so CI does not assert it.
• Not verified: whether the runner image applies the effort for codex, and whether our OpenAI key can use gpt-6.1-sol. The first review after deploy settles both: its log must show scout reported, not a scout failure note.

Rollout

Ships on the next bundle deploy. Independent of the driver PRs.

🤖 Generated with Claude Code

[cursor]

PR Summary

Medium Risk
Changes the production codex scout’s model and reasoning settings; runtime behavior depends on the runner honoring effort and API access to gpt-6.1-sol, which CI cannot fully verify.

Overview
The xreview-scout-codex bundle now pins gpt-6.1-sol (replacing gpt-5.6-sol) and sets executor.reasoning_effort: high so the codex scout runs above the harness default.

CI is aligned with the server parser: verify-agent-bundles.yml installs omnigent==0.16.0 (was 0.9.0), and both the PR gate and the ECR server in-image bundle check assert reasoning_effort: high for xreview-scout-codex via a new optional expectation column/field (None skips the check for other bundles).

^{Reviewed by Cursor Bugbot for commit fd4cefb. Bugbot is set up for automated code reviews on this repo. Configure here.}

[seidroid]

Moves the codex scout from gpt-5.6-sol to gpt-6.1-sol and sets executor.reasoning_effort: high. Nothing blocks: codex's one finding, that gpt-6.1-sol is not a supported model id, was dropped because OpenAI released that model on 2026-09-29 and Codex CLI 0.159.1 made it the default; codex raised nothing else, and the only open risk is the effort field, which nothing checks before deploy.

Non-blocking

1 finding on the changed lines, as inline comments.

1 nit, not posted on the code

• agents/xreview-scout-codex/config.yaml:31 — The new reason ("the Codex CLI's default model") doesn't fit this bundle well: it runs the codex harness, and line 42 says the image carries no Codex CLI. The removed reason, matching the model ai-review's codex pass resolves to, applied directly to this bundle. If that is still the goal, say so here.

_{seidroid review · decision approve · session 9ebbe45d0f8b408199ec3249d21e48ff · turn resp_claude_b6fb989970bc0366ac5a17513858c035 · item b9fa60044aed5e6d93d9e998958c793f}

Findings: 0 blocking | 1 non-blocking | 1 posted inline

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 9267267. Configure here.}

[bdchatham]

@seidroid review

[seidroid]

Since my last review, the bundle check now runs on omnigent 0.16.0 and asserts reasoning_effort == 'high' for the codex scout. Running the check on 0.16.0 passes as committed and fails once the field is removed, which resolves my open thread, so nothing blocks; codex's only finding, the model-id claim, is dropped again because OpenAI released gpt-6.1-sol on 2026-09-29 and Codex CLI 0.159.1 made it the default.

1 nit, not posted on the code

• agents/xreview-scout-codex/config.yaml:31 — The new reason ("the Codex CLI's default model") doesn't fit this bundle well: it runs the codex harness, and line 42 says the image carries no Codex CLI. The removed reason, matching the model ai-review's codex pass resolves to, applied directly to this bundle. If that is still the goal, say so here.

_{seidroid review · decision approve · session 9ebbe45d0f8b408199ec3249d21e48ff · turn resp_claude_a6883a6d3220e6a05fd6858a54c83843 · item 537bd76159bc5556bc8c12ef7fd35519}

Findings: 0 blocking | 0 non-blocking | 0 posted inline

[bdchatham]

@seidroid review

[seidroid]

The only change since my last review rewords the model comment in the codex scout config. It no longer leans on the Codex CLI default (my earlier nit) and now gives a reason that fits this bundle; the workflow assertion I checked last time is unchanged, so nothing blocks. Codex's only finding, that gpt-6.1-sol is not a valid model id, is dropped for the third time: OpenAI released gpt-6.1-sol on 2026-09-29 and it is a supported Codex model.

_{seidroid review · decision approve · session 9ebbe45d0f8b408199ec3249d21e48ff · turn resp_claude_0060ab2fc60494bb7fd51db2b6b08d09 · item a84cab1600bd5ce396b689e66b92d787}

Findings: 0 blocking | 0 non-blocking | 0 posted inline

[bdchatham]

@seidroid review

[seidroid]

New since my last review: ecr-server.yml's deploy gate now also asserts reasoning_effort == 'high' for the codex scout, matching the PR gate. Nothing blocks; of codex's two findings, the model-id claim is dropped again (OpenAI released gpt-6.1-sol on 2026-09-29 and it is a supported Codex model), and its low-severity point on ecr-server.yml:128 holds and is kept below as a nit.

1 nit, not posted on the code

• .github/workflows/ecr-server.yml:128 — (Raised by codex.) The comment at line 100 says None means a bundle sets no effort, but want_effort is not None skips the check for those bundles, so an effort added to any of them would still pass. Either assert None for them too, or reword the comment to say None means not checked.

_{seidroid review · decision approve · session 9ebbe45d0f8b408199ec3249d21e48ff · turn resp_claude_6c1d379ffa8ea3852bd048b782813e97 · item ac7352086a92598484bee4ab0c137f5d}

Findings: 0 blocking | 0 non-blocking | 0 posted inline

[bdchatham]

@seidroid review

[seidroid]

The only change since my last review rewords the ecr-server.yml comment to say None skips the effort check, which settles the nit I carried from codex; nothing else moved, and nothing blocks. Codex's one finding this round, that gpt-6.1-sol is not a supported model id, is dropped again because OpenAI released gpt-6.1-sol on 2026-09-29 and it is a supported Codex model.

_{seidroid review · decision approve · session 9ebbe45d0f8b408199ec3249d21e48ff · turn resp_claude_78d76ba04dcf554df7e7c79b34be8ed0 · item 6057bf25d1d55835a773a6984cf52f17}

Findings: 0 blocking | 0 non-blocking | 0 posted inline