From 6fd9efacee276479273fc331690dc2132641f74f Mon Sep 17 00:00:00 2001 From: Matt Miller Date: Mon, 27 Jul 2026 19:00:23 -0700 Subject: [PATCH] chore(cursor-review): bump panel pins to claude-opus-5-thinking-max and kimi-k3-max (BE-4818) Refresh the review panel per the 2026-07-27 catalog re-audit. Anthropic slot moves Opus 4.8 -> Opus 5 thinking (max tier, ZDR-eligible, identical per-token pricing) and the judge default follows it; Moonshot moves K2.7 Code -> Kimi K3 (max tier). OpenAI (gpt-5.6-sol-max) and Google (gemini-3.1-pro) are unchanged. README model/inputs tables and the judge prompt's attribution example are kept in sync. --- .github/cursor-review/README.md | 6 +++--- .github/cursor-review/prompt-judge.md | 2 +- .github/workflows/cursor-review.yml | 24 +++++++++++++++--------- 3 files changed, 19 insertions(+), 13 deletions(-) diff --git a/.github/cursor-review/README.md b/.github/cursor-review/README.md index 7ee26ae..d8a1e7a 100644 --- a/.github/cursor-review/README.md +++ b/.github/cursor-review/README.md @@ -57,9 +57,9 @@ skipped if no Slack token is configured). | Lab | Model (Cursor catalog) | |---|---| | OpenAI | `gpt-5.6-sol-max` | -| Anthropic | `claude-opus-4-8-thinking-max` | +| Anthropic | `claude-opus-5-thinking-max` | | Google | `gemini-3.1-pro` | -| Moonshot | `kimi-k2.7-code` | +| Moonshot | `kimi-k3-max` | Each model runs **two review types**: @@ -190,7 +190,7 @@ descriptions live in the [workflow header](../workflows/cursor-review.yml). | Input | Default | What it does | |---|---|---| -| `judge_model` | `claude-opus-4-8-thinking-max` | Model that consolidates panel findings. | +| `judge_model` | `claude-opus-5-thinking-max` | Model that consolidates panel findings. | | `diff_size_cap` | `5000` | Max changed lines (after excludes); larger PRs are skipped. | | `review_label` | `cursor-review` | Label whose addition triggers the review. | | `diff_excludes` | lockfiles, `node_modules`, `dist`, `vendor`, minified/generated files | Pathspecs excluded from both the size count and the reviewed diff. | diff --git a/.github/cursor-review/prompt-judge.md b/.github/cursor-review/prompt-judge.md index dbe96e5..c222ef3 100644 --- a/.github/cursor-review/prompt-judge.md +++ b/.github/cursor-review/prompt-judge.md @@ -56,7 +56,7 @@ object with exactly: - "body": string — concise (1-3 sentences). Do NOT prefix the body with a severity word or emoji; the severity field drives the rendered badge. END with attribution like - `_Raised by 3 of 8 reviewers (gpt-5.6-sol-max adversarial, claude-opus-4-8-thinking-max edge-case, gemini-3.1-pro adversarial)._` + `_Raised by 3 of 8 reviewers (gpt-5.6-sol-max adversarial, claude-opus-5-thinking-max edge-case, gemini-3.1-pro adversarial)._` Order the array most-severe first. If no findings rise to the bar, return []. diff --git a/.github/workflows/cursor-review.yml b/.github/workflows/cursor-review.yml index 31cb62f..ae73294 100644 --- a/.github/workflows/cursor-review.yml +++ b/.github/workflows/cursor-review.yml @@ -53,7 +53,7 @@ on: the panel. type: string required: false - default: claude-opus-4-8-thinking-max + default: claude-opus-5-thinking-max diff_size_cap: description: >- Max changed lines (added + removed, after diff_excludes). PRs over @@ -317,19 +317,25 @@ jobs: # Keep in sync with the README model table. # # Pinned to each lab's newest highest-reasoning tier in Cursor's - # catalog (last checked 2026-07-14). OpenAI: GPT-5.6 Sol is the - # 5.6-family coding flagship (no 5.6 codex variant exists). Anthropic: - # Fable 5 tops the catalog but all Fable entries are marked NO ZDR in - # Cursor, and this panel reviews private-repo diffs, so it is excluded; - # Opus 4.8 thinking at its max tier is the top ZDR-eligible pick. - # Gemini (Pro line) and Kimi still ship a single reasoning tier each. + # catalog (last checked 2026-07-27). OpenAI: GPT-5.6 Sol is the + # 5.6-family coding flagship (Luna is the cost tier, Terra mid; the + # Codex line stops at 5.3). Anthropic: Fable 5 tops the catalog but + # all Fable entries are marked NO ZDR in Cursor, and this panel + # reviews private-repo diffs, so it is excluded; Opus 5 thinking at + # its max tier is ZDR-eligible (Cursor states this explicitly) and + # replaces Opus 4.8 at identical per-token pricing. Gemini: 3.1 Pro + # is still the only Pro-line entry (3.5/3.6 are Flash tier). Moonshot: + # Kimi K3 (max tier) supersedes K2.7 Code; no NO-ZDR marker in the + # catalog (same evidentiary bar as every other pin), ~3.2x/3.75x the + # per-token cost of K2.7 on this lab's two panel cells (~+10-15% + # total panel cost per PR). run: | cat > /tmp/models.json <<'JSON' [ "gpt-5.6-sol-max", - "claude-opus-4-8-thinking-max", + "claude-opus-5-thinking-max", "gemini-3.1-pro", - "kimi-k2.7-code" + "kimi-k3-max" ] JSON echo "models=$(jq -c . /tmp/models.json)" >> "$GITHUB_OUTPUT"