ruvnet--ruflo/.github/workflows/oia-audit-weekly.yml

# Weekly composite audit (ADR-150 Phase 2 — iter 7).
#
# Runs `harness oia-audit` against the ruflo repo every Sunday at
# 04:17 UTC (off-peak, off-the-hour to avoid the global :00 thundering
# herd; see CronCreate "Avoid the :00 and :30 minute marks").
#
# The audit bundles three orthogonal MetaHarness static-analysis
# surfaces — oia-manifest + threat-model + mcp-scan — into one
# timestamped record uploaded as a CI artifact. Failure threshold:
# composite worst severity >= HIGH fails the workflow.
#
# Accumulated artifacts (retained 90 days) enable drift detection
# over time without needing a persistent memory store in CI.
#
# ADR-150 graceful degradation: when metaharness is unavailable, the
# script emits a degraded payload and exits 0 — the workflow
# continues, and the artifact records the degraded state.
name: oia-audit-weekly

on:
  schedule:
    - cron: '17 4 * * 0'  # Sundays at 04:17 UTC
  workflow_dispatch:        # manual trigger for ad-hoc audits
    # iter 109 — parameterize the policy thresholds so ad-hoc runs can
    # explore stricter/looser gates without editing YAML.
    inputs:
      threshold:
        description: 'Structural similarity threshold (default 0.85 — scheduled). Lower = stricter drift alert.'
        type: string
        default: '0.85'
        required: false
      alert_on_new_severity:
        description: 'Alert on any introduced finding ≥ this severity'
        type: choice
        options:
          - info
          - low
          - medium
          - warn
          - high
          - error
          - critical
        default: high
        required: false
  push:
    branches: [main]
    paths:
      - 'plugins/ruflo-metaharness/scripts/oia-audit.mjs'
      - 'plugins/ruflo-metaharness/scripts/_harness.mjs'
      - '.github/workflows/oia-audit-weekly.yml'

jobs:
  audit:
    runs-on: ubuntu-latest
    timeout-minutes: 10
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-node@v4
        with:
          node-version: '20'

      - name: Run composite audit (alert on HIGH)
        # --dry-run because CI has no persistent memory store; the
        # artifact upload below is the durability mechanism.
        # --alert-on-worst high fails the job on any composite HIGH
        # finding; medium/low/clean pass.
        run: |
          node plugins/ruflo-metaharness/scripts/oia-audit.mjs \
            --path . \
            --dry-run \
            --alert-on-worst high \
            --format json \
            > /tmp/oia-audit.json
          cat /tmp/oia-audit.json | head -80
          # Extract composite severity for the workflow summary
          WORST=$(node -e "
            const j = JSON.parse(require('fs').readFileSync('/tmp/oia-audit.json'));
            console.log(j.composite?.worst || 'unknown');
          ")
          echo "## Composite worst severity: \`$WORST\`" >> $GITHUB_STEP_SUMMARY
          echo "Artifact uploaded under \`oia-audit-$(date -u +%Y-%m-%d)\`." >> $GITHUB_STEP_SUMMARY

      - name: Upload audit artifact (90-day retention for drift tracking)
        if: always()
        uses: actions/upload-artifact@v4
        with:
          name: oia-audit-${{ github.run_id }}
          path: /tmp/oia-audit.json
          retention-days: 90

      # iter 69 — close the loop: the weekly cron has always claimed
      # "Accumulated artifacts enable drift detection over time" (line 13)
      # but the diff step was never wired. Iter-67's --baseline-file
      # fastpath made it cheap (~1.4s). Now every Sunday's run diffs
      # against the most recent prior weekly artifact and surfaces the
      # structural distance + alert.
      # iter 70 — `if: always()` so drift detection fires EVEN when the
      # audit step exits non-zero (HIGH alert). The pre-iter-70 default
      # conditional skipped drift exactly when it was most valuable:
      # the weeks where something was breaking.
      - name: Download prior week's audit artifact (if any)
        id: prior-artifact
        if: always()
        continue-on-error: true
        run: |
          set +e
          # List recent successful runs of this workflow, pick the most
          # recent one BEFORE this run, then download its artifact.
          PREV_RUN=$(gh run list \
            --workflow=oia-audit-weekly.yml \
            --status=success \
            --limit=10 \
            --json databaseId,headSha \
            --jq ".[] | select(.databaseId != ${{ github.run_id }}) | .databaseId" \
            | head -1)
          if [ -z "$PREV_RUN" ]; then
            echo "No prior successful run found — first weekly audit. Skipping drift step."
            echo "has_prior=false" >> $GITHUB_OUTPUT
            exit 0
          fi
          echo "Downloading prior audit from run $PREV_RUN..."
          gh run download "$PREV_RUN" \
            --name "oia-audit-${PREV_RUN}" \
            --dir /tmp/prior \
            && echo "has_prior=true" >> $GITHUB_OUTPUT \
            && ls -la /tmp/prior/
        env:
          GH_TOKEN: ${{ github.token }}

      - name: Compute structural drift vs prior week (iter 69)
        # iter 70 — always() AND has-prior: fires on the failure path too
        if: always() && steps.prior-artifact.outputs.has_prior == 'true'
        run: |
          set -e
          # iter-67 fastest path: --baseline-file skips audit-list + memory roundtrip
          # iter 79 — also gate on new HIGH-severity findings. Catches the
          # case where structure stayed similar but a security regression
          # appeared (e.g., new mcp-scan HIGH finding). Either gate can fire.
          # iter 109 — parameterized policy thresholds via workflow_dispatch
          # inputs. Scheduled runs fall back to 0.85 / high.
          THRESHOLD="${{ inputs.threshold || '0.85' }}"
          ALERT_SEV="${{ inputs.alert_on_new_severity || 'high' }}"
          node plugins/ruflo-metaharness/scripts/drift-from-history.mjs \
            --baseline-file /tmp/prior/oia-audit.json \
            --dry-run \
            --threshold "$THRESHOLD" \
            --alert-on-new-severity "$ALERT_SEV" \
            --format json \
            > /tmp/drift-trend.json || true
          # Extract verdict for the workflow summary
          VERDICT=$(node -e "
            const j = JSON.parse(require('fs').readFileSync('/tmp/drift-trend.json'));
            console.log(j.drift?.structuralDistance?.verdict || 'unavailable');
          ")
          OVERALL=$(node -e "
            const j = JSON.parse(require('fs').readFileSync('/tmp/drift-trend.json'));
            console.log(j.drift?.structuralDistance?.overall ?? 'n/a');
          ")
          ALERT=$(node -e "
            const j = JSON.parse(require('fs').readFileSync('/tmp/drift-trend.json'));
            console.log(j.alert?.triggered ? 'TRIGGERED' : 'OK');
          ")
          # iter 97 — also surface the fast-path label so the Actions UI
          # shows which iter-66/67 path executed. Confirms cron used the
          # baseline-file path (expected ~1.4s) and not slow path (~26s).
          PATH_TAKEN=$(node -e "
            const j = JSON.parse(require('fs').readFileSync('/tmp/drift-trend.json'));
            const t = j.timing || {};
            console.log((t.path || 'unknown') + ' (wall ' + (t.parallelWallMs ?? '?') + 'ms)');
          ")
          echo "## Drift vs prior week: \`$VERDICT\` (similarity=$OVERALL)" >> $GITHUB_STEP_SUMMARY
          echo "Alert status: \`$ALERT\`" >> $GITHUB_STEP_SUMMARY
          echo "Path: \`$PATH_TAKEN\`" >> $GITHUB_STEP_SUMMARY
          # iter 108 — fail the workflow if the cron accidentally fell
          # through to the slow path. The cron's --baseline-file flag
          # is the load-bearing invariant for cron-budget correctness;
          # if it disappears, the audit takes ~26s instead of ~1.4s
          # and the iter-7 weekly-cron budget exhausts on slower runners.
          PATH_LABEL=$(node -e "
            const j = JSON.parse(require('fs').readFileSync('/tmp/drift-trend.json'));
            console.log(j.timing?.path || 'unknown');
          ")
          if [ "$PATH_LABEL" != "file" ]; then
            echo "::error::Cron drift step expected timing.path='file' but got '$PATH_LABEL' — the iter-67 fastpath flag may have regressed"
            exit 1
          fi

      - name: Upload drift trend artifact
        # iter 70 — same always() pattern; the trend artifact is most
        # useful for forensics when the workflow has failed.
        if: always() && steps.prior-artifact.outputs.has_prior == 'true'
        uses: actions/upload-artifact@v4
        with:
          name: drift-trend-${{ github.run_id }}
          path: /tmp/drift-trend.json
          retention-days: 90