coderabbit-config-validation.yml

name: CodeRabbit Config Validation

# Called from pr.yml, so it is covered by the aggregate "PR Check Summary" required check.
# It deliberately has no `paths:` gate and no `dorny/paths-filter` step: pr.yml also fires on
# `merge_group` and `workflow_dispatch`, where a path filter has no PR context to diff against,
# and the summary job treats a skipped dependency as a failure. The whole job costs a checkout
# plus a ~10s CLI download, so it just always runs.
on:
  workflow_call:

permissions:
  contents: read

jobs:
  validate:
    name: Validate CodeRabbit config
    runs-on: ubuntu-latest
    timeout-minutes: 10

    steps:
      - name: Harden the runner (Audit all outbound calls)
        uses: step-security/harden-runner@e14015d583714f6e62063499dc959a02595150a1 # v2.21.1
        with:
          egress-policy: audit

      - name: Checkout repository
        uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5.0.0
        with:
          persist-credentials: false

      # Runs before the CLI: this is the check that would have caught the outage the config
      # comment in `.coderabbit.yaml` describes, it needs no network, and it takes
      # milliseconds. `coderabbit config validate` cannot catch it — a path_filters list that
      # silently reduces every review to two files is still perfectly schema-valid.
      - name: Check path_filters are exclusions only
        shell: bash
        run: |
          set -euo pipefail
          python3 - <<'PY'
          import sys

          import yaml

          CONFIG = ".coderabbit.yaml"

          with open(CONFIG, encoding="utf-8") as handle:
              config = yaml.safe_load(handle) or {}

          filters = (config.get("reviews") or {}).get("path_filters") or []
          if not isinstance(filters, list):
              print(f"::error file={CONFIG}::reviews.path_filters must be a list")
              sys.exit(1)

          positive = [entry for entry in filters if not str(entry).strip().startswith("!")]
          if positive:
              print(f"::error file={CONFIG}::reviews.path_filters contains a positive pattern")
              print(
                  "A single non-'!' entry in reviews.path_filters turns the list into an "
                  "allowlist: CodeRabbit then reviews only the files a positive pattern "
                  "matches and drops every other file as 'included by none'.",
                  file=sys.stderr,
              )
              print("Offending entries:", file=sys.stderr)
              for entry in positive:
                  print(f"  - {entry!r}", file=sys.stderr)
              print(
                  "\nExpress the intent with exclusions that do not match the files you want "
                  "reviewed, rather than excluding a directory and re-including a file from it.",
                  file=sys.stderr,
              )
              sys.exit(1)

          print(f"reviews.path_filters: {len(filters)} entries, all exclusions.")
          PY

      # Deliberately not `curl https://cli.coderabbit.ai/install.sh | sh`: that script is
      # mutable and fetches the release archive without verifying it, so piping it to a shell
      # would undo the pin below. Fetch the immutable versioned artifact instead and check it
      # against a recorded digest, which is the same posture as pinning an action to a SHA.
      # Bump CODERABBIT_VERSION and CODERABBIT_SHA256 together — a mismatch fails loudly,
      # which is the point of recording it.
      - name: Install the CodeRabbit CLI
        shell: bash
        env:
          CODERABBIT_VERSION: "0.7.2"
          CODERABBIT_SHA256: "32dd3a5a1238fa68eb7cfa95cc19cf6ad4980882589c616c0e300b920f1bb865"
          CODERABBIT_INSTALL_DIR: ${{ runner.temp }}/coderabbit-cli
        run: |
          set -euo pipefail

          archive="$(mktemp -d)/coderabbit-linux-x64.zip"
          curl -fsSL --retry 3 --retry-delay 2 -o "$archive" \
            "https://cli.coderabbit.ai/releases/${CODERABBIT_VERSION}/coderabbit-linux-x64.zip"

          echo "${CODERABBIT_SHA256}  ${archive}" | sha256sum --check --strict -

          mkdir -p "$CODERABBIT_INSTALL_DIR"
          unzip -q -o "$archive" coderabbit -d "$CODERABBIT_INSTALL_DIR"
          chmod +x "$CODERABBIT_INSTALL_DIR/coderabbit"
          echo "$CODERABBIT_INSTALL_DIR" >> "$GITHUB_PATH"

          # The digest only proves what was downloaded; this proves what will run.
          installed="$("$CODERABBIT_INSTALL_DIR/coderabbit" --version)"
          if [ "$installed" != "$CODERABBIT_VERSION" ]; then
            echo "::error::Expected CodeRabbit CLI ${CODERABBIT_VERSION}, got ${installed}"
            exit 1
          fi
          echo "CodeRabbit CLI ${installed} verified against the recorded digest."

      # Two-stage: YAML syntax, then the schema CodeRabbit publishes — the only authority on
      # which keys and values actually exist. Needs no authentication.
      - name: Validate against the published schema
        shell: bash
        run: |
          set -euo pipefail

          status=0
          output="$(coderabbit config validate .coderabbit.yaml 2>&1)" || status=$?
          printf '%s\n' "$output"

          if [ "$status" -eq 0 ]; then
            exit 0
          fi

          # The CLI exits 1 both for an invalid config and for its own inability to fetch the
          # schema. This check gates every merge, so a CodeRabbit outage must not block the
          # queue — but it must be loud rather than silently green.
          if printf '%s' "$output" | grep -qF "Could not load the current CodeRabbit configuration schema"; then
            echo "::warning file=.coderabbit.yaml::Could not reach the CodeRabbit schema; skipped schema validation. The path_filters check above still ran."
            exit 0
          fi

          echo "::error file=.coderabbit.yaml::CodeRabbit configuration is invalid"
          exit "$status"