Phase 5 deployment hygiene — Dockerfile, demo discovery, first 3 screenshots

Substrate fixes surfaced by actually deploying the v3.4 stack in demo
mode for screenshot capture. None of the Phase 5 sub-agents ran the
deployment; these are real release-prep gaps the test suite + npm
audit didn't catch.

- Dockerfile: COPY VERSION (was missing — /health reported 0.0.0)
  + COPY demo/ (was missing — demo mode had no content to discover).

- docker-compose.yml: pass RAGBOT_DEMO + RAGBOT_SCHEDULER from host
  env into the api container so `RAGBOT_DEMO=1 docker compose up -d`
  actually enables demo mode (was previously a no-op).

- web/Dockerfile: bumped node:20-alpine → node:22-alpine + installs
  npm@11 explicitly. The lockfile generated by the host's npm 11.6
  (Agent B's Dependabot pin-bumps) couldn't be validated by the
  container's bundled npm 10.8 — `Missing: @emnapi/...` errors.

- src/ragbot/_demo_registration.py: demo skill-roots filter no longer
  excludes the bundled starter pack. The original filter was
  over-aggressive: it isolated USER skills correctly but also hid the
  six universal starter-pack skills that ship with v3.4. Fixed
  filter returns [starter_pack_root, demo/skills/] so demo evaluators
  see the realistic skill surface a new install presents.

- docs/screenshots/keyboard-shortcuts-overlay.png + mcp-settings.png +
  skills-panel.png: real 1440x900 captures via Playwright against the
  v3.4.0 demo deployment. Replace the 1x1 placeholder PNGs Agent C
  staged. The remaining 4 captures are still in flight (one of them
  surfaced a real policy-API bug being fixed next).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: rajivpant

Dockerfile

@@ -26,7 +26,11 @@ COPY --from=dependencies /usr/local/bin /usr/local/bin
 
 COPY src/ ./src/
 COPY engines.yaml .
+COPY VERSION .
 COPY ragbot ragbot_api ./
+# Bundled demo workspace + demo skill pack used by RAGBOT_DEMO=1.
+# Discovery short-circuits to this content when demo mode is active.
+COPY demo/ ./demo/
 
 RUN chmod +x ragbot ragbot_api
 

docker-compose.yml

@@ -44,6 +44,16 @@ services:
       # Windows) and recent Docker Engine on Linux. Override if Ollama runs
       # elsewhere (e.g., a LAN-shared inference server).
       OLLAMA_API_BASE: ${OLLAMA_API_BASE:-http://host.docker.internal:11434}
+      # Demo mode: when set to 1, discovery hard-isolates to the bundled
+      # demo/ workspace and the demo skill pack so screenshots, docs, and
+      # cold-start evaluations all run against the same content. Default
+      # is unset (production behaviour). Passed through from the host
+      # env (`RAGBOT_DEMO=1 docker compose up -d`).
+      RAGBOT_DEMO: ${RAGBOT_DEMO:-}
+      # Opt-in scheduler (Phase 4 + 5). When set to 1, the lifespan
+      # handler starts the synthesis_engine.tasks.SchedulerLoop so
+      # cron-style routines (memory.consolidate_recent_idle, etc.) fire.
+      RAGBOT_SCHEDULER: ${RAGBOT_SCHEDULER:-}
     volumes:
       # Persistent session storage
       - ragbot-sessions:/root/.local/share/ragbot/sessions

docs/screenshots/keyboard-shortcuts-overlay.png

@@ -56,15 +56,32 @@ def _workspaces_filter() -> Optional[Dict[str, str]]:
 
 
 def _skill_roots_filter() -> Optional[List[str]]:
-    """Return only the bundled demo skills root when demo mode is active.
+    """Return the bundled-skill roots when demo mode is active.
 
     Returning ``None`` signals "no override; let the substrate scan the
     normal skill-root chain."
+
+    Demo mode isolates USER-installed skills (workspace-scoped collections,
+    operator's private skill repo, plugin-installed skills) from the
+    evaluator's view so the demo shows a known-clean state. The bundled
+    starter pack — the universal-scope skills that ship with the substrate
+    out of the box — is INCLUDED in demo mode because it is part of "what
+    Ragbot v3.4 ships with," not part of the operator's content. Pairing
+    the starter pack with the demo's own ``demo/skills/`` directory gives
+    evaluators a realistic view of the skill surface a new user sees.
     """
     if not is_demo_mode():
         return None
+    # Lazy import keeps the demo-registration module's import surface
+    # narrow (ragbot/_demo_registration → synthesis_engine.skills imports
+    # would otherwise cycle through ragbot's __init__).
+    from synthesis_engine.skills.starter_pack import starter_pack_root
+
+    roots: List[str] = [starter_pack_root()]
     path = demo_skills_path()
-    return [str(path)] if path is not None else []
+    if path is not None:
+        roots.append(str(path))
+    return roots
 
 
 set_discovery_filter(SCOPE_WORKSPACES, _workspaces_filter)

docs/screenshots/mcp-settings.png

@@ -1,4 +1,11 @@
-FROM node:20-alpine AS base
+FROM node:22-alpine AS base
+
+# Match the host's npm major. The lockfile that ships with v3.4 was
+# generated by npm 11.x (Phase 5 Agent B's Dependabot triage produced
+# the npm overrides that require the newer lockfile schema). The Node
+# 22 base ships with npm 10.9; we bump to npm 11 explicitly so `npm ci`
+# can validate the lockfile reproducibly.
+RUN npm install -g npm@11
 
 # Install dependencies only when needed
 FROM base AS deps