[PR #371] [MERGED] feat(cursor-review): move the panel's Anthropic cell to Opus 5.5 at -xhigh #373

New issue

Closed

opened 2026-10-26 22:39:09 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-26 22:39:09 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/Comfy-Org/github-workflows/pull/371
Author: @mattmillerai
Created: 10/7/2026
Status: ✅ Merged
Merged: 10/7/2026
Merged by: @mattmillerai

Base: main ← Head: matt/panel-opus-5-5

📝 Commits (1)

7fb2256 feat(cursor-review): move the panel's Anthropic cell to Opus 5.5 at -xhigh

📊 Changes

4 files changed (+38 additions, -21 deletions)

View changed files

📝 .github/cursor-review/README.md (+13 -3)
📝 .github/cursor-review/prompt-judge.md (+1 -1)
📝 .github/workflows/cursor-review.yml (+20 -13)
📝 docs/callers/cursor-review.md (+4 -4)

📄 Description

What

  "gpt-5.6-sol-max",
- "claude-opus-5-thinking-max",
+ "claude-opus-5-5-xhigh",
  "kimi-k3-high"

The Anthropic cell was on claude-opus-5-thinking-max — that is Opus 5, not 5.5. This moves it to Opus 5.5 one reasoning notch down, on the same spend reasoning as the Kimi -max→-high and judge -max→-xhigh moves from the 2026-08-27 review.

Verified against the live catalog, not from memory

Read out of the preflight job log of run 37688671671, which prints cursor-agent models in full:

claude-opus-5-5-low / -medium / -high / -xhigh / -max   (+ -fast variants)

No claude-opus-5-5-* entry carries a NO ZDR marker, so it is eligible for a panel that reviews private-repo diffs. (Fable remains the NO-ZDR family and stays excluded.)

The catalog quirk this surfaced

Opus 5.5 ships no -thinking- variants. Opus 5 has both a plain and a -thinking- line; 5.5 has only claude-opus-5-5-<tier>. So the old pin has no literal counterpart and the nearest equivalent is a plain tier id — exactly the "the same tier of a newer version may not exist" case the caller docs warn about. Now documented in the heredoc comment and the README so the next person bumping this doesn't go looking for claude-opus-5-5-thinking-xhigh.

Two axes, on purpose

This moves version and tier together. That's fine for a default change and would be wrong for an attribution experiment. The comment says so explicitly, so it isn't later mistaken for one.

Kept in sync

The heredoc is the single source of truth, but several things quote it and would have drifted:

README panel table + the rationale prose
the last checked audit date catalog-drift.py reads (verified it still extracts: panel, judge and date all parse)
the worked panel_models example in the caller docs, which otherwise proposed what is now the default
the attribution-format example in prompt-judge.md

Deliberately not in scope

The judge stays on claude-opus-5-thinking-xhigh (Opus 5). Its default is replicated across cursor-axis-base.yml and the five axis-*.yml cursor-approve workflows — a different system, mid-change in #369. Worth doing, in its own PR.

Testing

python3 -m unittest discover -s .github/cursor-review/tests — 1147 passed
check_workflow_pins.py — pass
catalog-drift.py extraction re-run against the edited workflow: panel ['gpt-5.6-sol-max', 'claude-opus-5-5-xhigh', 'kimi-k3-high'], judge intact, last checked → 2026-10-07

Reaching cloud

Merging this trips bump-cursor-review-callers.yml, which opens the SHA-bump PR on each caller. Cloud picks up Opus 5.5 when that bump PR merges there — no variable and no org admin needed. Independent of #370, which adds the variable knob for future experiments.

🤖 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/371 **Author:** [@mattmillerai](https://github.com/mattmillerai) **Created:** 10/7/2026 **Status:** ✅ Merged **Merged:** 10/7/2026 **Merged by:** [@mattmillerai](https://github.com/mattmillerai) **Base:** `main` ← **Head:** `matt/panel-opus-5-5` --- ### 📝 Commits (1) - [`7fb2256`](https://github.com/Comfy-Org/github-workflows/commit/7fb2256a1a565c4b1f002407932eca87847db949) feat(cursor-review): move the panel's Anthropic cell to Opus 5.5 at -xhigh ### 📊 Changes **4 files changed** (+38 additions, -21 deletions) <details> <summary>View changed files</summary> 📝 `.github/cursor-review/README.md` (+13 -3) 📝 `.github/cursor-review/prompt-judge.md` (+1 -1) 📝 `.github/workflows/cursor-review.yml` (+20 -13) 📝 `docs/callers/cursor-review.md` (+4 -4) </details> ### 📄 Description ## What ```diff "gpt-5.6-sol-max", - "claude-opus-5-thinking-max", + "claude-opus-5-5-xhigh", "kimi-k3-high" ``` The Anthropic cell was on `claude-opus-5-thinking-max` — that is Opus **5**, not 5.5. This moves it to Opus 5.5 one reasoning notch down, on the same spend reasoning as the Kimi `-max`→`-high` and judge `-max`→`-xhigh` moves from the 2026-08-27 review. ## Verified against the live catalog, not from memory Read out of the preflight job log of run `37688671671`, which prints `cursor-agent models` in full: ``` claude-opus-5-5-low / -medium / -high / -xhigh / -max (+ -fast variants) ``` No `claude-opus-5-5-*` entry carries a `NO ZDR` marker, so it is eligible for a panel that reviews private-repo diffs. (Fable remains the NO-ZDR family and stays excluded.) ## The catalog quirk this surfaced **Opus 5.5 ships no `-thinking-` variants.** Opus 5 has both a plain and a `-thinking-` line; 5.5 has only `claude-opus-5-5-<tier>`. So the old pin has no literal counterpart and the nearest equivalent is a plain tier id — exactly the "the same tier of a newer version may not exist" case the caller docs warn about. Now documented in the heredoc comment and the README so the next person bumping this doesn't go looking for `claude-opus-5-5-thinking-xhigh`. ## Two axes, on purpose This moves **version and tier together**. That's fine for a default change and would be wrong for an attribution experiment. The comment says so explicitly, so it isn't later mistaken for one. ## Kept in sync The heredoc is the single source of truth, but several things quote it and would have drifted: - README panel table + the rationale prose - the `last checked` audit date `catalog-drift.py` reads (verified it still extracts: panel, judge and date all parse) - the worked `panel_models` example in the caller docs, which otherwise proposed what is now the default - the attribution-format example in `prompt-judge.md` ## Deliberately not in scope **The judge stays on `claude-opus-5-thinking-xhigh`** (Opus 5). Its default is replicated across `cursor-axis-base.yml` and the five `axis-*.yml` cursor-approve workflows — a different system, mid-change in #369. Worth doing, in its own PR. ## Testing - `python3 -m unittest discover -s .github/cursor-review/tests` — 1147 passed - `check_workflow_pins.py` — pass - `catalog-drift.py` extraction re-run against the edited workflow: panel `['gpt-5.6-sol-max', 'claude-opus-5-5-xhigh', 'kimi-k3-high']`, judge intact, `last checked` → `2026-10-07` ## Reaching cloud Merging this trips `bump-cursor-review-callers.yml`, which opens the SHA-bump PR on each caller. Cloud picks up Opus 5.5 when that bump PR merges there — no variable and no org admin needed. Independent of #370, which adds the variable knob for future experiments. 🤖 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 22:39:09 +00:00