[PR #28] [MERGED] ci(cursor-review): bump panel to latest per-lab reasoning tiers (GPT-5.6 Sol, Opus 4.8 max) #38

New issue

Closed

opened 2026-10-26 22:35:38 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-26 22:35:38 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/Comfy-Org/github-workflows/pull/28
Author: @mattmillerai
Created: 7/14/2026
Status: ✅ Merged
Merged: 7/19/2026
Merged by: @mattmillerai

Base: main ← Head: ci/bump-panel-models-2026-07

📝 Commits (1)

d6b585a ci(cursor-review): bump panel + judge to newest per-lab reasoning tiers

📊 Changes

3 files changed (+14 additions, -10 deletions)

View changed files

📝 .github/cursor-review/README.md (+3 -3)
📝 .github/cursor-review/prompt-judge.md (+1 -1)
📝 .github/workflows/cursor-review.yml (+10 -6)

📄 Description

ELI-5

The 4-model review panel was pinned to models from a few releases ago. This bumps each lane to the newest, highest-reasoning model its lab currently offers in Cursor's catalog, so PR reviews come from the best available reviewers.

What changed

Lane	Before	After
OpenAI	`gpt-5.3-codex-xhigh`	`gpt-5.6-sol-max`
Anthropic (panel + judge default)	`claude-opus-4-8-thinking-xhigh`	`claude-opus-4-8-thinking-max`
Google	`gemini-3.1-pro`	unchanged (still the only Pro-line entry)
Moonshot	`kimi-k2.7-code`	unchanged (still the only Kimi entry)

Also syncs the panel table in the cursor-review README and the model-name examples in the judge prompt, and rewrites the matrix comment with the selection rationale and a last-checked date.

Selection notes

GPT-5.6 Sol is OpenAI's stated coding flagship of the 5.6 family; the catalog has no 5.6 codex variant, so Sol max replaces Codex 5.3 xhigh as the top OpenAI reasoning tier.
Fable 5 deliberately excluded: it tops Cursor's Anthropic catalog, but every Fable entry is marked NO ZDR (no zero-data-retention). This panel reviews diffs from non-public repos, so lanes must stay on ZDR-eligible models. Opus 4.8 thinking at its new max tier is the top eligible pick.
max vs xhigh: the panel's contract is "highest reasoning tier available per lab." Max tiers think longer; if cells start brushing the 30-minute timeout on large diffs, the fallback is the xhigh tier one notch down.

Validation

actionlint clean.
Both new model IDs validated with live cursor-agent --model <id> -p calls (both respond correctly) — guarding against the pin-a-nonexistent-ID failure mode this file has hit before.
Merging this triggers the caller SHA-bump fan-out; the resulting bump PRs in consumer repos are expected.

🤖 Generated with Claude Code

_{🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.}

## 📋 Pull Request Information **Original PR:** https://github.com/Comfy-Org/github-workflows/pull/28 **Author:** [@mattmillerai](https://github.com/mattmillerai) **Created:** 7/14/2026 **Status:** ✅ Merged **Merged:** 7/19/2026 **Merged by:** [@mattmillerai](https://github.com/mattmillerai) **Base:** `main` ← **Head:** `ci/bump-panel-models-2026-07` --- ### 📝 Commits (1) - [`d6b585a`](https://github.com/Comfy-Org/github-workflows/commit/d6b585a4c0cdf03705a19d3cd921478436d3f38e) ci(cursor-review): bump panel + judge to newest per-lab reasoning tiers ### 📊 Changes **3 files changed** (+14 additions, -10 deletions) <details> <summary>View changed files</summary> 📝 `.github/cursor-review/README.md` (+3 -3) 📝 `.github/cursor-review/prompt-judge.md` (+1 -1) 📝 `.github/workflows/cursor-review.yml` (+10 -6) </details> ### 📄 Description ## ELI-5 The 4-model review panel was pinned to models from a few releases ago. This bumps each lane to the newest, highest-reasoning model its lab currently offers in Cursor's catalog, so PR reviews come from the best available reviewers. ## What changed | Lane | Before | After | |---|---|---| | OpenAI | `gpt-5.3-codex-xhigh` | `gpt-5.6-sol-max` | | Anthropic (panel + judge default) | `claude-opus-4-8-thinking-xhigh` | `claude-opus-4-8-thinking-max` | | Google | `gemini-3.1-pro` | unchanged (still the only Pro-line entry) | | Moonshot | `kimi-k2.7-code` | unchanged (still the only Kimi entry) | Also syncs the panel table in the cursor-review README and the model-name examples in the judge prompt, and rewrites the matrix comment with the selection rationale and a last-checked date. ## Selection notes 1. **GPT-5.6 Sol** is OpenAI's stated coding flagship of the 5.6 family; the catalog has no 5.6 codex variant, so Sol max replaces Codex 5.3 xhigh as the top OpenAI reasoning tier. 2. **Fable 5 deliberately excluded**: it tops Cursor's Anthropic catalog, but every Fable entry is marked **NO ZDR** (no zero-data-retention). This panel reviews diffs from non-public repos, so lanes must stay on ZDR-eligible models. Opus 4.8 thinking at its new max tier is the top eligible pick. 3. **max vs xhigh**: the panel's contract is "highest reasoning tier available per lab." Max tiers think longer; if cells start brushing the 30-minute timeout on large diffs, the fallback is the xhigh tier one notch down. ## Validation 1. `actionlint` clean. 2. Both new model IDs validated with live `cursor-agent --model <id> -p` calls (both respond correctly) — guarding against the pin-a-nonexistent-ID failure mode this file has hit before. 3. Merging this triggers the caller SHA-bump fan-out; the resulting bump PRs in consumer repos are expected. 🤖 Generated with [Claude Code](https://claude.com/claude-code) --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi

2026-10-26 22:35:38 +00:00