claude-fixer.yml

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
      id-token: write
    steps:
      - uses: actions/checkout@v4
        with:
          ref: ${{ github.event.workflow_run.head_branch }}
          token: ${{ secrets.GITHUB_TOKEN }}
          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: Configure git identity
        run: |
          git config user.name "github-actions[bot]"
          git config user.email "github-actions[bot]@users.noreply.github.com"

      - name: Generate fix branch name
        id: branch
        run: echo "name=fix/claude-auto-$(date +%s)" >> $GITHUB_OUTPUT

      - 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 }}"

          # Write all temp files to /tmp so the working tree stays clean for Claude's changes
          gh api "repos/${REPO}/actions/runs/${RUN_ID}/jobs" --paginate > /tmp/all_jobs.json
          FAILED_JOB_IDS=$(jq -r '.jobs[] | select(.conclusion == "failure") | .id' /tmp/all_jobs.json)

          : > /tmp/failed_logs.txt
          FAILED_TEST_FILES=""

          for job_id in $FAILED_JOB_IDS; do
            JOB_NAME=$(jq -r ".jobs[] | select(.id == $job_id) | .name" /tmp/all_jobs.json)
            echo "=== Failed Shard: ${JOB_NAME} ===" >> /tmp/failed_logs.txt

            gh api "repos/${REPO}/actions/jobs/${job_id}/logs" > /tmp/raw_log.txt 2>/dev/null || {
              echo "[Could not fetch logs for job ${job_id}]" >> /tmp/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=$(grep -n "Failed Tests" /tmp/raw_log.txt | tail -1 | cut -d: -f1)
            if [ -n "$START_LINE" ]; then
              CONTEXT_START=$((START_LINE > 5 ? START_LINE - 5 : 1))
              tail -n +"$CONTEXT_START" /tmp/raw_log.txt >> /tmp/failed_logs.txt
            else
              echo "[No 'Failed Tests' section found — showing last 100 lines]" >> /tmp/failed_logs.txt
              tail -100 /tmp/raw_log.txt >> /tmp/failed_logs.txt
            fi
            echo "" >> /tmp/failed_logs.txt

            NEW_FILES=$(grep " FAIL  test/" /tmp/raw_log.txt | grep -o "test/[^[:space:]>]*\.test\.ts" | sort -u | tr '\n' ' ')
            FAILED_TEST_FILES="${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: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}
          claude_args: "--allowedTools Bash,Read,Edit,Write --max-turns 100"
          prompt: |
            The "${{ 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 `/tmp/failed_logs.txt`. Read it first.

            Follow these steps carefully:

            1. **Diagnose the root cause**: Read `/tmp/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. **Stop once all tests pass** — do NOT run any git commands. The workflow will handle committing and opening the PR.

          additional_permissions: |
            actions: read

      - name: Restore git remote auth
        env:
          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
        run: git remote set-url origin https://x-access-token:${GITHUB_TOKEN}@github.com/${{ github.repository }}.git

      - name: Commit fixes
        id: auto-commit
        uses: stefanzweifel/git-auto-commit-action@v5
        with:
          commit_message: "fix: auto-fix test failures from ${{ github.event.workflow_run.name }}"
          branch: ${{ steps.branch.outputs.name }}
          create_branch: true

      - name: Open pull request and post failure logs
        if: steps.auto-commit.outputs.changes_detected == 'true'
        uses: actions/github-script@v7
        with:
          github-token: ${{ secrets.GITHUB_TOKEN }}
          script: |
            const { data: pr } = await github.rest.pulls.create({
              owner: context.repo.owner,
              repo: context.repo.repo,
              title: 'fix: auto-fix ${{ github.event.workflow_run.name }} failures on ${{ github.event.workflow_run.head_branch }}',
              body: [
                'Failures detected in `${{ github.event.workflow_run.name }}` on branch `${{ github.event.workflow_run.head_branch }}`.',
                '',
                'This PR was automatically generated by the Claude Test Fixer workflow.',
                '',
                '> ⚠️ AI-generated fix. Please review carefully before merging.',
              ].join('\n'),
              head: '${{ steps.branch.outputs.name }}',
              base: '${{ github.event.workflow_run.head_branch }}',
            });

            const fs = require('fs');
            let logs = fs.readFileSync('/tmp/failed_logs.txt', 'utf8');
            if (logs.length > 60000) {
              logs = logs.substring(0, 60000) + '\n... (truncated — see full logs in the failed workflow run)';
            }

            await github.rest.issues.createComment({
              owner: context.repo.owner,
              repo: context.repo.repo,
              issue_number: pr.number,
              body: [
                '## Test Failure Logs',
                '',
                '<details>',
                '<summary>Click to expand failure logs from the failed CI run</summary>',
                '',
                '```',
                logs,
                '```',
                '',
                '</details>',
              ].join('\n'),
            });