Skip to content

Update Benchmarks

Update Benchmarks #1

name: Update Benchmarks
on:
schedule:
- cron: '30 2 * * 1'
workflow_dispatch:
concurrency:
group: update-benchmarks
cancel-in-progress: false
permissions:
contents: write
pull-requests: write
jobs:
update-benchmarks:
runs-on: ubuntu-latest
steps:
- name: Checkout repository
uses: actions/checkout@v7
- name: Setup pnpm
uses: ./.github/actions/setup-pnpm
- name: Refresh exact benchmark mappings
run: pnpm fetch:benchmarks
- name: Regenerate manifest modules
run: pnpm generate:manifests
- name: Validate updated data
run: |
pnpm test:validate
pnpm data-health:check
pnpm format:check
pnpm biome:check
git diff --check
- name: Check for changes
id: git-check
run: |
git diff --exit-code manifests src/lib/generated || echo "changes=true" >> "$GITHUB_OUTPUT"
- name: Create pull request
if: steps.git-check.outputs.changes == 'true'
uses: peter-evans/create-pull-request@v8
with:
token: ${{ secrets.GITHUB_TOKEN }}
commit-message: 'chore(data): refresh benchmark scores'
title: 'chore(data): refresh benchmark scores'
body: |
## Summary
Automated refresh from exact entries in the official SWE-bench leaderboard data.
Each model manifest declares the upstream result ID and exact displayed label. The
refresh fails rather than applying fuzzy model-name matching.
## Validation
- Manifest schemas
- Generated manifest modules
- Data health
branch: auto-update-benchmarks
delete-branch: true
labels: |
automated