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  schedule:
11  - cron: 47 1 * * 2
12  workflow_dispatch: {}
13jobs:
14  cron_unit_evals:
15    runs-on: namespace-profile-16x32-ubuntu-2204
16    strategy:
17      matrix:
18        model:
19        - anthropic/claude-sonnet-4-5-latest
20        - anthropic/claude-opus-4-5-latest
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      shell: bash -euxo pipefail {0}
32    - name: steps::cache_rust_dependencies_namespace
33      uses: namespacelabs/nscloud-cache-action@v1
34      with:
35        cache: rust
36    - name: steps::setup_linux
37      run: ./script/linux
38      shell: bash -euxo pipefail {0}
39    - name: steps::install_mold
40      run: ./script/install-mold
41      shell: bash -euxo pipefail {0}
42    - name: steps::download_wasi_sdk
43      run: ./script/download-wasi-sdk
44      shell: bash -euxo pipefail {0}
45    - name: steps::cargo_install_nextest
46      uses: taiki-e/install-action@nextest
47    - name: steps::clear_target_dir_if_large
48      run: ./script/clear-target-dir-if-larger-than 250
49      shell: bash -euxo pipefail {0}
50    - name: ./script/run-unit-evals
51      run: ./script/run-unit-evals
52      shell: bash -euxo pipefail {0}
53      env:
54        ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
55        OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
56        GOOGLE_AI_API_KEY: ${{ secrets.GOOGLE_AI_API_KEY }}
57        GOOGLE_CLOUD_PROJECT: ${{ secrets.GOOGLE_CLOUD_PROJECT }}
58        ZED_AGENT_MODEL: ${{ matrix.model }}
59    - name: steps::cleanup_cargo_config
60      if: always()
61      run: |
62        rm -rf ./../.cargo
63      shell: bash -euxo pipefail {0}
64    - name: run_agent_evals::cron_unit_evals::send_failure_to_slack
65      if: ${{ failure() }}
66      uses: slackapi/slack-github-action@b0fa283ad8fea605de13dc3f449259339835fc52
67      with:
68        method: chat.postMessage
69        token: ${{ secrets.SLACK_APP_ZED_UNIT_EVALS_BOT_TOKEN }}
70        payload: |
71          channel: C04UDRNNJFQ
72          text: "Unit Evals Failed: https://github.com/zed-industries/zed/actions/runs/${{ github.run_id }}"
73concurrency:
74  group: ${{ github.workflow }}-${{ github.ref_name }}-${{ github.ref_name == 'main' && github.sha || 'anysha' }}
75  cancel-in-progress: true