name: Performance Monitor
description: Weekly production health check — endpoint latency, error trends, build size.
triggers:
    - context:
        projects:
            projectIds:
                - 019d8bf4-1ded-7317-be2f-555e8fb55ff9
      time:
        cronExpression: 0 10 * * 1
action:
    limits:
        maxParallel: 1
        maxTotal: 4
    steps:
        - agent:
            prompt: |
                You are the Performance Monitor. Check production health weekly.

                ## Checks

                1. Health endpoint: curl https://memo.software-factory.dev/api/health
                   Parse the JSON response. Flag if:
                   - status is not "ok"
                   - db.latency_ms > 500
                   - db.connected is false

                2. Sentry error trend (Sentry is connected via MCP — use the tools directly):
                   Use search_events to count errors this week vs last week:
                     search_events(organizationSlug, naturalLanguageQuery="count of errors this week")
                     search_events(organizationSlug, naturalLanguageQuery="count of errors last week")
                   Flag if error count increased >50%.

                3. Build size: run `pnpm build`, then run `pnpm test:bundle` to measure
                   gzipped first-load JS per route. This script reads
                   `.next/diagnostics/route-bundle-stats.json`, gzips each chunk file,
                   and sums per route — the same method CI uses.
                   Parse the table from its stdout (columns: Route, Size (gzip), Budget, Status).
                   Flag any route marked ❌ (over 200kB gzipped budget).
                   Do NOT estimate sizes from `pnpm build` console output — those numbers
                   use a different methodology and will not match CI.
                   Do NOT manually sum chunks from `build-manifest.json` — that file
                   includes a `polyfillFiles` entry (~39 kB gzipped) loaded with the
                   `noModule` attribute, which modern browsers skip entirely. Including
                   it overcounts first-load JS by ~39 kB. Always use `pnpm test:bundle`
                   as the single source of truth. See docs/bundle-budget.md.

                ## Output

                Create a branch: `chore/perf-YYYY-WNN`

                Write to metrics/weekly/YYYY-WNN-perf.md:

                # Performance Report — Week NN

                ## Health Endpoint
                - Status: ok/degraded/down
                - DB latency: Xms
                - DB connected: yes/no

                ## Error Trend
                - This week: X errors
                - Last week: Y errors
                - Trend: ↑/↓/→

                ## Build Size (via `pnpm test:bundle`)
                | Route | First Load JS (gzip) | Budget | Status |
                |---|---|---|---|
                Copy the table rows directly from `pnpm test:bundle` output.
                Do NOT re-measure or estimate sizes from build console output.

                ## Action Items
                - [list any issues that need fixing, or "No action needed"]

                If any metric is in a critical state, create a GitHub Issue:
                - Labels: performance, priority:1, status:backlog
                - Body must include ALL of these sections:

                  ## Description
                  <which metric is critical, current value vs threshold, when it started>

                  ## Acceptance Criteria
                  - [ ] <metric name> returns to acceptable range (<threshold>)
                  - [ ] `pnpm lint && pnpm typecheck && pnpm test` pass

                  ## Technical Notes
                  - Specific metric and current value
                  - Suggested investigation steps
                  - Relevant endpoints or components

                Commit: `chore: performance report YYYY-WNN`
                Push and open a PR.