A team adds an evaluation workflow for agent-generated pull requests. The workflow should preserve enough evidence for error analysis even when the agent passes some checks but fails others. Evaluate the following statements. 1. checks.json provides quantitative workflow status evidence. 2. review-signals.json alone proves the task is successful. 3. if: always() helps retain evidence after failed steps.

name: agent-evaluation

on:
  pull_request:
    branches:
      - main

permissions:
  contents: read
  pull-requests: read
  checks: read

jobs:
  evaluate:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4

         - name: Collect test output
           run: |
             mkdir -p evaluation
             npm ci
             npm test 2>&1 | tee evaluation/test-output.log

         - name: Collect check state
           env:
             GH_TOKEN: ${{ github.token }}
           run: |
             gh pr checks ${{ github.event.pull_request.number }} \
                --json name,state,conclusion \
                > evaluation/checks.json

         - name: Collect review signals
           env:
             GH_TOKEN: ${{ github.token }}
           run: |
             gh pr view ${{ github.event.pull_request.number }} \
                --json reviews,comments \
                > evaluation/review-signals.json

         - name: Upload evaluation evidence
           if: always()
           uses: actions/upload-artifact@v4
           with:
             name: agent-evaluation
             path: evaluation/