Skip to content

Claude Test Fixer

Claude Test Fixer #1

Workflow file for this run

name: Claude Test Fixer
on:
workflow_run:
workflows: ["Test all (Linux)", "Test all (MacOS)"]
types: [completed]
jobs:
determine-runner:
runs-on: ubuntu-latest
if: ${{ github.event.workflow_run.conclusion == 'failure' }}
outputs:
runner: ${{ steps.pick.outputs.runner }}
steps:
- id: pick
run: |
if [[ "${{ github.event.workflow_run.name }}" == *"MacOS"* ]]; then
echo "runner=macos-latest" >> $GITHUB_OUTPUT
else
echo "runner=ubuntu-latest" >> $GITHUB_OUTPUT
fi
fix-on-failure:
needs: determine-runner
runs-on: ${{ needs.determine-runner.outputs.runner }}
permissions:
contents: write
pull-requests: write
actions: read
steps:
- uses: actions/checkout@v4
with:
ref: ${{ github.event.workflow_run.head_branch }}
fetch-depth: 0
- name: Use Node.js 24
uses: actions/setup-node@v4
with:
node-version: '24.x'
cache: 'npm'
- name: Enable linger (Linux only)
if: runner.os == 'Linux'
run: loginctl enable-linger $(whoami)
- run: npm ci
- name: Fetch and filter failure logs from all shards
id: fetch-logs
env:
GH_TOKEN: ${{ github.token }}
run: |
REPO="${GITHUB_REPOSITORY}"
RUN_ID="${{ github.event.workflow_run.id }}"
gh api "repos/${REPO}/actions/runs/${RUN_ID}/jobs" --paginate > all_jobs.json
FAILED_JOB_IDS=$(jq -r '.jobs[] | select(.conclusion == "failure") | .id' all_jobs.json)
: > failed_logs.txt
FAILED_TEST_FILES=""
for job_id in $FAILED_JOB_IDS; do
JOB_NAME=$(jq -r ".jobs[] | select(.id == $job_id) | .name" all_jobs.json)
echo "=== Failed Shard: ${JOB_NAME} ===" >> failed_logs.txt
gh api "repos/${REPO}/actions/jobs/${job_id}/logs" > raw_log.txt 2>/dev/null || {
echo "[Could not fetch logs for job ${job_id}]" >> failed_logs.txt
continue
}
# "Failed Tests" is always the last section in vitest output —
# take from its line to EOF so we capture all failures without a hard limit.
START_LINE=$(grep -n "Failed Tests" raw_log.txt | tail -1 | cut -d: -f1)
if [ -n "$START_LINE" ]; then
CONTEXT_START=$((START_LINE > 5 ? START_LINE - 5 : 1))
tail -n +"$CONTEXT_START" raw_log.txt >> failed_logs.txt
else
echo "[No 'Failed Tests' section found — showing last 100 lines]" >> failed_logs.txt
tail -100 raw_log.txt >> failed_logs.txt
fi
echo "" >> failed_logs.txt
NEW_FILES=$(grep " FAIL test/" raw_log.txt | grep -o "test/[^[:space:]>]*\.test\.ts" | sort -u | tr '\n' ' ')
FAILED_TEST_FILES="${FAILED_TEST_FILES}${NEW_FILES}"
done
FAILED_TEST_FILES=$(echo "${FAILED_TEST_FILES}" | tr ' ' '\n' | grep -v '^$' | sort -u | tr '\n' ' ' | xargs)
echo "failed_test_files=${FAILED_TEST_FILES}" >> "$GITHUB_OUTPUT"
- name: Claude Fix Failed Tests
uses: anthropics/claude-code-action@v1
with:
claude_code_oauth_token: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}
prompt: |
The "${{ github.event.workflow_run.name }}" workflow failed on branch "${{ github.event.workflow_run.head_branch }}".
Failing test files: ${{ steps.fetch-logs.outputs.failed_test_files }}
Filtered failure logs from all failed shards are in `failed_logs.txt` in the working directory. Read it first.
Follow these steps carefully:
1. **Diagnose the root cause**: Read `failed_logs.txt` thoroughly. Before writing a single line of code, understand WHY the tests are failing — not just the symptom. Read the relevant source files to understand the existing architecture and patterns.
2. **Implement an architecturally sound fix**: The fix must address the underlying root cause, match existing code patterns in this codebase, and introduce no workarounds or hacks. Do not mask symptoms.
3. **Verify each fix**: For each failing test file, run:
```
npm run test -- <test_file_path> --no-file-parallelism --disable-console-intercept
```
On macOS, prepend `CI=true` and ensure `/opt/homebrew/bin` is in PATH.
4. **Iterate until all tests pass**: If tests still fail after a fix, re-read the output, deepen your understanding of the root cause, and refine the fix. Do not give up and do not apply increasingly speculative patches. Each iteration must be grounded in analysis.
5. **Create a PR only after all failing tests pass**:
- Create a new branch from "${{ github.event.workflow_run.head_branch }}" (e.g. `fix/claude-auto-<short-description>`)
- Commit the changes with a descriptive message explaining the root cause and fix
- Open a pull request targeting "${{ github.event.workflow_run.head_branch }}"
additional_permissions: |
actions: read