feat: ship production deploy, Playwright e2e, and agent implementer

Add Docker atomic deploy tooling, CI, Playwright smoke/visual suite, shared industrial theme, physics invariants, quality-mode wiring, and a worktree-isolated agent implementer MVP with hard safety limits. Co-authored-by: Cursor <cursoragent@cursor.com>
2026-07-15 17:14:01 +00:00
parent 985f7c327d
commit b5985f33bb
34 changed files with 2076 additions and 71 deletions
--- a/docs/AGENT_IMPLEMENTER_MVP.md
+++ b/docs/AGENT_IMPLEMENTER_MVP.md
@@ -0,0 +1,124 @@
+# Implementer MVP (worktree-isolated)
+
+**Status:** Implemented in `agent/cli.mjs`  
+**Date:** 2026-07-15  
+
+Safe, deterministic Implementer cycle that **does not** require an external LLM.  
+Patches run only inside a git worktree. The agent **never** merges to `main`, **never** deploys, and **never** pushes.
+
+---
+
+## Commands
+
+| Command | Purpose |
+| ------- | ------- |
+| `dry-run` | Plan only (baseline + propose) |
+| `run-once` | Verify baseline only (no patches) |
+| `implement [--task <path>]` | Full Implementer cycle |
+| `start` | Same as implement, but requires `AUTONOMOUS_AGENT_ENABLED=true` |
+| `status` / `stop` / `resume` / `pause` / `report` | Control plane |
+
+```bash
+# Recommended successful demo cycle
+node agent/cli.mjs implement --task agent/tasks/demo-docs.json
+
+# Or via helper script
+bash scripts/agent-implement.sh --task agent/tasks/demo-docs.json
+
+# Gated continuous entry (still one cycle in MVP)
+AUTONOMOUS_AGENT_ENABLED=true node agent/cli.mjs start --task agent/tasks/demo-docs.json
+```
+
+---
+
+## Implementer cycle
+
+1. **Watchdog** — kill switch, single lock, disk, load, memory  
+2. **Plan** — load task JSON (default: `agent/tasks/demo-docs.json`)  
+3. **Validate policy** — allowlist only: `docs`, `tests`, `data-testid`, `a11y`, `small-ui`, `logging-scripts`  
+4. **Create worktree** — `.agent/worktrees/<run-id>` on branch `agent/<run-id>/<task-slug>`  
+5. **Apply bounded patch** — e.g. create `docs/AGENT_CYCLE_DEMO.md`  
+6. **Enforce limits** — ≤12 files, ≤800 diff lines, ≤25 min cycle  
+7. **Verify** — `npm test && npm run build` inside the worktree  
+8. **Score + decide** — `ACCEPT_CANDIDATE` or `REJECT`  
+9. **Report** — `agent/reports/<run-id>.json` (+ `latest.json`)
+
+### On REJECT
+
+- Remove worktree  
+- Delete local agent branch  
+- Keep the JSON report  
+
+### On ACCEPT_CANDIDATE
+
+- Keep worktree + branch  
+- Print manual review instructions  
+- Human must cherry-pick / merge into a feature branch if desired  
+
+---
+
+## Watchdog
+
+| Check | Fail condition |
+| ----- | -------------- |
+| Lock | `agent/state/agent.lock` held by a live PID (second agent refused) |
+| Kill switch | `agent/state/KILL` present |
+| Disk | `df /` usage **> 85%** |
+| Load | loadavg 1m **> nproc** |
+| Memory | MemAvailable **< 1 GiB** |
+
+```bash
+node agent/cli.mjs stop     # write KILL
+node agent/cli.mjs resume   # clear KILL
+node agent/cli.mjs status   # lock + watchdog snapshot
+```
+
+---
+
+## Example task
+
+`agent/tasks/demo-docs.json` — docs-only safe demo:
+
+- Policy: `docs`  
+- Patch: create `docs/AGENT_CYCLE_DEMO.md`  
+- No source / physics / deploy / `.env` changes  
+
+---
+
+## Limits (also in `LIMITS` in `cli.mjs`)
+
+| Limit | Value |
+| ----- | ----: |
+| max cycle | 25 min |
+| max files | 12 |
+| max diff lines | 800 |
+| parallel agents | 1 |
+
+Hard forbids: merge to main, production deploy, push, secret/`.env` access.
+
+---
+
+## Manual review after ACCEPT
+
+```bash
+# Inspect
+cat agent/reports/latest.json
+git -C .agent/worktrees/<run-id> show HEAD
+
+# Accept into your feature branch yourself (example)
+git cherry-pick <agent-commit-sha>
+
+# Or discard
+git worktree remove --force .agent/worktrees/<run-id>
+git branch -D agent/<run-id>/demo-docs
+```
+
+---
+
+## What this MVP does **not** do
+
+- Call external LLMs  
+- Modify production nginx/docker/deploy  
+- Touch `.env`  
+- Auto-merge or auto-deploy  
+- Change demo physics (prefer docs-only demo task)  
--- a/docs/PERFORMANCE_AFTER_MAXIMUM_DEMO.md
+++ b/docs/PERFORMANCE_AFTER_MAXIMUM_DEMO.md
@@ -0,0 +1,35 @@
+# PERFORMANCE_AFTER_MAXIMUM_DEMO
+
+Measured: 2026-07-15T17:11:06.031Z  
+Commit: `985f7c3`
+
+## Bundle (measured)
+
+| Метрика | Значение |
+| ------- | -------: |
+| Total JS | 1300.9 KB |
+| Total JS gzip | 358.1 KB |
+| Main chunk | index-ncgt6PBL.js (96.4 KB gzip) |
+| R3F chunk gzip | 226.1 KB |
+
+## Runtime (Playwright Chromium headless)
+
+| Метрика | До | После | Цель | Статус |
+| ------- | -: | ----: | ---: | ------ |
+| Average FPS | n/a (не измерялось) | 9.6 | ≥30 | CHECK |
+| Minimum FPS | n/a | 3.3 | ≥20 | CHECK |
+| p95 frame time | n/a | 200.1 ms | ≤33 ms | CHECK |
+| Load time | n/a | 1008 ms | ≤5000 | OK |
+| JS heap start | n/a | 13400000 | — | measured |
+| JS heap after interaction | n/a | 13400000 | no runaway | measured |
+| Canvas present | — | 1 | ≥1 | OK |
+| JS gzip (main+vendor) | ~335 KB main previously | 358.1 KB | — | measured |
+
+## Notes
+
+- **Headless Chromium WebGL is not a GPU FPS measurement.** Values ~9 FPS / p95 ~200 ms reflect software/SwiftShader-like rendering in CI/headless, **not** desktop Chrome with NVIDIA. Do not claim 30–60 FPS from this pass.
+- For jury/demo FPS, measure in headed Chromium on the presentation machine (`npm run test:e2e:headed` + DevTools Performance) or Engineering Details FPS meter.
+- Draw calls / triangles require WebGL inspector in headed mode; not available in this headless pass.
+- Demo quality mode keeps `shadows=false` and `effectsEnabled=false` for stable real-device FPS (see `qualityMode.ts`). High mode may enable shadows.
+- Raw JSON: `agent/reports/perf-metrics.json`
+- Runtime error (if any): none
--- a/docs/PLAYWRIGHT_E2E_REPORT.md
+++ b/docs/PLAYWRIGHT_E2E_REPORT.md
@@ -0,0 +1,95 @@
+# Playwright E2E + Visual Regression
+
+Lightweight end-to-end and visual regression coverage for the continuous demo on `/` and the SPA route `/details`.
+
+## Prerequisites
+
+- Node.js with project deps installed (`npm install`)
+- Chromium for Playwright: `npx playwright install chromium`
+- App reachable at the base URL (default preview: `http://127.0.0.1:3101`)
+
+By default tests **do not** start a server. Start preview yourself, for example:
+
+```bash
+npm run build && npx vite preview --host 127.0.0.1 --port 3101
+```
+
+Or point at an already-running preview/prod instance via `PLAYWRIGHT_BASE_URL`.
+
+## Scripts
+
+| Script | Command | Purpose |
+|--------|---------|---------|
+| `test:e2e` | `playwright test` | Run all e2e + visual tests |
+| `test:e2e:headed` | `playwright test --headed` | Same, headed browser |
+| `test:e2e:update` | `playwright test --update-snapshots` | Refresh visual baselines |
+
+## Environment
+
+| Variable | Default | Meaning |
+|----------|---------|---------|
+| `PLAYWRIGHT_BASE_URL` | `http://127.0.0.1:3101` | Target app origin |
+| `PLAYWRIGHT_START_SERVER` | unset | Set to `1` to let Playwright run `vite preview` on port 3101 |
+
+Examples:
+
+```bash
+# Against local preview (already running)
+PLAYWRIGHT_BASE_URL=http://127.0.0.1:3101 npm run test:e2e
+
+# Against prod-style port
+PLAYWRIGHT_BASE_URL=http://127.0.0.1:3100 npm run test:e2e
+
+# Auto-start preview if nothing is listening
+PLAYWRIGHT_START_SERVER=1 npm run test:e2e
+```
+
+## Config highlights
+
+- **workers: 1** — keep e2e light on shared servers
+- **retries: 1**
+- **chromium only**
+- Screenshot on failure; trace on first retry
+- Soft visual thresholds (`threshold: 0.3`, `maxDiffPixelRatio: 0.05`)
+
+## Test map
+
+| Spec | Coverage |
+|------|----------|
+| `e2e/smoke.spec.ts` | `/` loads, canvas (or loading→canvas), no `pageerror`, play works, replay if finished |
+| `e2e/controls.spec.ts` | next/prev seek, jump case 0 & 8, speed buttons, presentation toggle |
+| `e2e/safety.spec.ts` | jam (index 8) → FAULT; emergency (index 9) → EMERGENCY |
+| `e2e/routes.spec.ts` | `/details`, refresh stays on details, navigate back to `/` |
+| `e2e/visual.spec.ts` | Idle HUD panel snapshot only (not full WebGL canvas) |
+
+Stable selectors live on `MainPage` as `data-testid` values (`demo-play`, `demo-hud`, `demo-case-{n}`, etc.).
+
+## Updating visual snapshots
+
+1. Ensure the app at `PLAYWRIGHT_BASE_URL` matches the UI you want to lock.
+2. Run:
+
+```bash
+npm run test:e2e:update
+```
+
+3. Review diffs under `e2e/visual.spec.ts-snapshots/` (and commit those PNGs when intentional).
+4. Re-run `npm run test:e2e` to confirm green.
+
+Snapshots intentionally target **`[data-testid="demo-hud"]`** so WebGL canvas noise does not fail CI.
+
+## System deps
+
+On a fresh Linux host, Chromium may need OS libraries:
+
+```bash
+npx playwright install chromium
+npx playwright install-deps chromium
+```
+
+## Artifacts (gitignored)
+
+- `test-results/` — failure screenshots / traces
+- `playwright-report/` — HTML report
+- `blob-report/` — blob reporter output
+`}
--- a/docs/PRODUCTION_DEPLOYMENT_REPORT.md
+++ b/docs/PRODUCTION_DEPLOYMENT_REPORT.md
@@ -0,0 +1,64 @@
+# PRODUCTION_DEPLOYMENT_REPORT
+
+Дата: 2026-07-15  
+Ветка: `feature/maximum-demo-realism`  
+Базовый commit этапа 1: `985f7c3`  
+Production bundle после этапа 2: `index-ncgt6PBL.js`
+
+## Способ деплоя
+
+Docker-контейнер `owl-web-1` (image `owl-web:<release>`), порт `127.0.0.1:3100→80`.  
+Docker API: `DOCKER_HOST=tcp://127.0.0.1:2375` (из coder-контейнера).  
+Compose-файл: `docker-compose.server.yml` (исторически `-p owl`).  
+Скрипт: `scripts/deploy-production.sh`.
+
+Публичный доступ в этой среде:
+
+- Loopback: `http://127.0.0.1:3100/`
+- Quick tunnel cloudflared: `https://invitations-based-characters-accent.trycloudflare.com/`
+- `https://arhipovdan.ru/` из контейнера даёт TLS SNI / openresty 404 — DNS/прокси домена вне текущего coder-окружения; проверяйте снаружи.
+
+## Releases
+
+| Поле | Значение |
+| ---- | -------- |
+| Old release container | `owl-web-1-backup-20260715-165347` / `…-1712` |
+| Old bundle | `index-CCdZzxPJ.js` |
+| New release | `20260715-1712` |
+| New image | `owl-web:20260715-1712` |
+| New bundle | `index-ncgt6PBL.js` |
+| Backup HTML | `releases/backup-pre-*` |
+
+## Команды
+
+```bash
+export DOCKER_HOST=tcp://127.0.0.1:2375
+cd /home/coder/arhipovdan/app
+npm test && npm run build
+bash scripts/deploy-production.sh 20260715-1712
+curl -s http://127.0.0.1:3100/ | grep -Eo 'index-[A-Za-z0-9_-]+\.js'
+```
+
+## Healthcheck
+
+| Check | Result |
+| ----- | ------ |
+| `GET /` | 200 |
+| `GET /details` (SPA) | 200 |
+| Bundle | `index-ncgt6PBL.js` |
+| Tunnel HTML | same bundle |
+
+## Rollback
+
+```bash
+export DOCKER_HOST=tcp://127.0.0.1:2375
+docker stop owl-web-1 && docker rm owl-web-1
+docker rename owl-web-1-backup-20260715-1712 owl-web-1
+docker start owl-web-1
+```
+
+Или восстановить HTML из `releases/backup-pre-<release>/`.
+
+## Smoke (tunnel)
+
+Подтверждено: CASE 1/10, Jump to case 1–10, Previous/Next, Present, Play Demo.