Strip TTS, Wikipedia RAG, activity-page agents, and OTEL for workshop starter

This rebuilds main as the workshop-starter version of the app: same
architecture, infra, art-style picker, and docs as all-features, but
with the four 'guide-able' features removed so workshop attendees can
build them back in following the docs in docs/04..07.

Removed:
- TTS: backend/app/tts.py, /api/tts route, AZURE_SPEECH_* settings,
  azure-cognitiveservices-speech dep, frontend listen/play buttons.
- Wikipedia RAG: backend/app/wikipedia.py, wikipedia_topic /
  wikipedia_mode StoryRequest fields, orchestrator integration,
  prompt block, and frontend Wikipedia mode UI.
- Activity-page agents: look_and_find, character_glossary,
  approval_gateway, final_assembly executors and the bonus fan-out
  branch in workflow.py; activity-page models, prompts, frontend
  pages, ProgressTracker rows, and demo-story event entries.
- OTEL observability: backend/app/telemetry.py, telemetry import in
  main.py, opentelemetry-* deps, OTEL spans/record_llm_usage calls
  in agents, and OTEL/ENABLE_INSTRUMENTATION settings.

Workflow now reverts to the core sequential chain:
Orchestrator -> StoryArchitect -> ArtDirector -> [StoryReviewer?] -> Decision

Decision yields the StoryResponse directly to the workflow output
(no FinalAssembly downstream).

Speech-related Bicep params kept as optional in infra/ so users
following the TTS guide can 'azd env set AZURE_SPEECH_RESOURCE_ID …'
+ 'azd provision' without re-editing Bicep.

Demo stories: stripped look_and_find / character_glossary keys from
each story.json and removed bonus-agent + wikipedia events from the
matching events.json so they replay cleanly with the new tracker.

Also: ignore backend/demo_stories/_drafts/ (per-session draft images
created by the storage backend at runtime, only promoted on save).

Verified:
- python -m compileall -q backend/app: clean
- 'from app.main import app; build_story_workflow(StoryRequest())': OK
- frontend npm run lint: 131 errors (down from 167 baseline)
- frontend npm run build: clean
- az bicep build infra/main.bicep: clean
- end-to-end SSE smoke test (uvicorn + curl): event: complete fires;
  final story has only the new StoryResponse keys (no look_and_find,
  no character_glossary); only the 5 expected executors emit progress

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Author: asofio

.gitignore

@@ -22,6 +22,11 @@ env/
 # Per-developer azd env config (only the .example.sh template is committed)
 infra/scripts/azd-env.local.sh
 
+# Per-session draft images created by the storage backend at runtime.
+# Drafts are promoted into a permanent demo-story folder when the user clicks
+# "Save Story"; otherwise they remain here as ephemeral local artifacts.
+backend/demo_stories/_drafts/
+
 # Node
 node_modules/
 frontend/dist/

backend/.env.example

@@ -19,26 +19,3 @@ FOUNDRY_IMAGE_MODEL_DEPLOYMENT_NAME=gpt-image-1.5
 # CORS — URL of the React dev server. Change if you run frontend on a different port.
 CORS_ORIGIN=http://localhost:5173
 
-# Azure Speech Service (Text-to-Speech)
-# Region where your Speech resource is deployed (e.g. eastus, westus2)
-# Authentication uses DefaultAzureCredential — ensure your identity has
-# the "Cognitive Services Speech User" role on the Speech resource.
-AZURE_SPEECH_REGION=[REGION]
-AZURE_SPEECH_ENDPOINT=https://[REGION].tts.speech.microsoft.com
-
-# Full ARM resource ID of your Speech resource (required for AAD token auth)
-# Format: /subscriptions/<sub-id>/resourceGroups/<rg>/providers/Microsoft.CognitiveServices/accounts/<name>
-AZURE_SPEECH_RESOURCE_ID=/subscriptions/[SUBSCRIPTION_ID]/resourceGroups/[RESOURCE_GROUP]/providers/Microsoft.CognitiveServices/accounts/[ACCOUNT_NAME]
-
-# ── Agent Framework Observability ──────────────────────────────────────────────
-# Activates Agent Framework's built-in instrumentation (spans for agent
-# invocations, LLM chat calls, and tool executions).
-ENABLE_INSTRUMENTATION=true
-
-# Includes prompts, completions, function arguments and results in span
-# attributes. WARNING: only enable in dev/test — may expose PII in traces.
-ENABLE_SENSITIVE_DATA=true
-
-# ── Standard OpenTelemetry ─────────────────────────────────────────────────────
-OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:4317
-OTEL_SERVICE_NAME=children-story-studio

backend/app/agents/approval_gateway.py

@@ -1,20 +1,18 @@
 """
-DecisionExecutor — Routing node in the workflow.
+DecisionExecutor — Routing & terminal node in the workflow.
 
 Receives the ReviewResult and decides:
-  - If approved OR max revision cycles reached → assemble StoryResponse and send_message
-    downstream to FinalAssemblyExecutor (or fan-out to bonus agents)
-  - If rejected AND revision budget remains → send RevisionSignal back to OrchestratorExecutor
-
-NOTE: This node no longer calls ctx.yield_output() directly. The terminal yield_output
-call has moved to FinalAssemblyExecutor, which is always the last node in the graph.
+  - If approved OR max revision cycles reached → assemble StoryResponse and
+    yield_output to end the workflow
+  - If rejected AND revision budget remains → send RevisionSignal back to
+    OrchestratorExecutor (full regen) or ImageRevisionSignal back to
+    ArtDirectorExecutor (selective image regen)
 """
 
 import logging
 
 from agent_framework import Executor, WorkflowContext, handler
 
-from ..config import settings
 from ..events import ProgressDetailEvent
 from ..models import ReviewResult, StoryDraft, StoryResponse
 from ..signals import ImageRevisionSignal, RevisionSignal
@@ -26,7 +24,7 @@
 
 class DecisionExecutor(Executor):
     """
-    Routes the workflow: approve & emit final story, or loop back for revision.
+    Routes the workflow: approve & yield final story, or loop back for revision.
     """
 
     def __init__(self) -> None:
@@ -75,10 +73,7 @@ async def handle_review(
                 )
 
             story_response = await self._assemble_story(review, revision_count, ctx)
-            # Persist the approved StoryResponse so FinalAssemblyExecutor can always
-            # read it from shared state, regardless of which bonus agents executed.
-            ctx.set_state("approved_story", story_response.model_dump_json())
-            await ctx.send_message(story_response)
+            await ctx.yield_output(story_response)
 
         elif (
             review.revision_scope == "images_only"
@@ -160,3 +155,4 @@ async def _assemble_story(
             review_notes=issue_summary if issue_summary else "Story approved with no issues.",
             revision_rounds=revision_rounds,
         )
+