run_cron_unit_evals.yml

 1# Generated from xtask::workflows::run_cron_unit_evals
 2# Rebuild with `cargo xtask workflows`.
 3name: run_cron_unit_evals
 4env:
 5  CARGO_TERM_COLOR: always
 6  CARGO_INCREMENTAL: '0'
 7  RUST_BACKTRACE: '1'
 8  ZED_CLIENT_CHECKSUM_SEED: ${{ secrets.ZED_CLIENT_CHECKSUM_SEED }}
 9on:
10  workflow_dispatch: {}
11jobs:
12  cron_unit_evals:
13    runs-on: namespace-profile-16x32-ubuntu-2204
14    strategy:
15      matrix:
16        model:
17        - anthropic/claude-sonnet-4-5-latest
18        - anthropic/claude-opus-4-5-latest
19        - google/gemini-3-pro
20        - openai/gpt-5
21      fail-fast: false
22    steps:
23    - name: steps::checkout_repo
24      uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
25      with:
26        clean: false
27    - name: steps::setup_cargo_config
28      run: |
29        mkdir -p ./../.cargo
30        cp ./.cargo/ci-config.toml ./../.cargo/config.toml
31    - name: steps::cache_rust_dependencies_namespace
32      uses: namespacelabs/nscloud-cache-action@v1
33      with:
34        cache: rust
35        path: ~/.rustup
36    - name: steps::setup_linux
37      run: ./script/linux
38    - name: steps::install_mold
39      run: ./script/install-mold
40    - name: steps::download_wasi_sdk
41      run: ./script/download-wasi-sdk
42    - name: steps::cargo_install_nextest
43      uses: taiki-e/install-action@nextest
44    - name: steps::clear_target_dir_if_large
45      run: ./script/clear-target-dir-if-larger-than 250
46    - name: ./script/run-unit-evals
47      run: ./script/run-unit-evals
48      env:
49        ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
50        OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
51        GOOGLE_AI_API_KEY: ${{ secrets.GOOGLE_AI_API_KEY }}
52        GOOGLE_CLOUD_PROJECT: ${{ secrets.GOOGLE_CLOUD_PROJECT }}
53        ZED_AGENT_MODEL: ${{ matrix.model }}
54    - name: steps::cleanup_cargo_config
55      if: always()
56      run: |
57        rm -rf ./../.cargo
58    - name: run_agent_evals::cron_unit_evals::send_failure_to_slack
59      if: ${{ failure() }}
60      uses: slackapi/slack-github-action@b0fa283ad8fea605de13dc3f449259339835fc52
61      with:
62        method: chat.postMessage
63        token: ${{ secrets.SLACK_APP_ZED_UNIT_EVALS_BOT_TOKEN }}
64        payload: |
65          channel: C04UDRNNJFQ
66          text: "Unit Evals Failed: https://github.com/zed-industries/zed/actions/runs/${{ github.run_id }}"
67concurrency:
68  group: ${{ github.workflow }}-${{ github.ref_name }}-${{ github.ref_name == 'main' && github.sha || 'anysha' }}
69  cancel-in-progress: true
70defaults:
71  run:
72    shell: bash -euxo pipefail {0}