Migrate CI to --from-obproject-toml, add CLAUDE.md

.github/workflows/deploy.yml

@@ -3,7 +3,6 @@ on:
   push:
     branches:
     - main
-    - eddie/test-01-26-2026
   pull_request:
     branches:
     - main
@@ -29,30 +28,17 @@ jobs:
         fetch-depth: 0
 
     - name: Set up Python
-      uses: actions/setup-python@v1
+      uses: actions/setup-python@v5
       with:
-        python-version: 3.12
+        python-version: "3.12"
 
     - name: Install dependencies
-      run: |
-        python3 -m pip install -U requests
-        python3 -m pip install outerbounds pyyaml
-        python3 -m pip install -U ob-project-utils
+      run: pip install outerbounds ob-project-utils
+
     - name: Configure Outerbounds
       run: |
-        PROJECT_NAME=$(yq .project obproject.toml)
-        DEFAULT_CICD_USER="${PROJECT_NAME//_/-}-cicd"
-        PLATFORM=$(yq .platform obproject.toml)
-        CICD_USER=$(yq ".cicd_user // \"$DEFAULT_CICD_USER\"")
-        PERIMETER="default"
-        echo "🏗️ Deployment target:"
-        echo "  Platform: $PLATFORM"
-        echo "  CI/CD User: $CICD_USER"
-        echo "  Perimeter: $PERIMETER"
         outerbounds service-principal-configure \
-          --name $CICD_USER \
-          --deployment-domain $PLATFORM \
-          --perimeter $PERIMETER \
+          --from-obproject-toml \
           --github-actions
 
     - name: Deploy Project
@@ -62,4 +48,4 @@ jobs:
         GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
         COMMENTS_URL: ${{ github.event.pull_request.comments_url }}
         PYTHONUNBUFFERED: 1
-      run: obproject-deploy
+      run: obproject-deploy

CLAUDE.md

@@ -0,0 +1,39 @@
+# Flow Chaining Example (flow_chaining_demo)
+
+Minimal example demonstrating two patterns for chaining flows:
+`@project_trigger(event=...)` for event-based triggering with payload,
+and `@trigger_on_finish(flow=...)` for direct flow completion chaining.
+
+## Platform Features Used
+
+- **Events**: `ProjectEvent("start_training")` published from PreprocessFlow, consumed by TrainFlow via `@project_trigger`
+- **Apps**: Trigger app in `deployments/trigger-app/` for firing events from a UI
+
+## Flows
+
+| Flow | Trigger | Purpose |
+|------|---------|---------|
+| PreprocessFlow | Manual | Parallel dataset processing via `foreach`, publishes `start_training` event |
+| TrainFlow | @project_trigger(start_training) | Receives event payload as Parameters, runs training |
+
+## Key Pattern
+
+PreprocessFlow publishes a `ProjectEvent("start_training")` with payload.
+TrainFlow's `@project_trigger(event="start_training")` maps payload keys to
+`Parameter` names (e.g., `processed_paths`). This is the recommended approach
+over `@trigger_on_finish` when you need to pass data via event payload.
+
+## Run Locally
+
+```bash
+cd flow-chaining-example
+python flows/preprocess/flow.py run --datasets "path1,path2,path3"
+python flows/train/flow.py run --learning_rate 0.05 --n_estimators 200
+```
+
+## Common Pitfalls
+
+- Event payload values map to Parameters by name; `processed_paths` is sent as JSON string and parsed in the flow
+- `@project_trigger` is branch-scoped (events only trigger flows on the same branch)
+- TrainFlow cannot access upstream artifacts when triggered by event (unlike `@trigger_on_finish` which exposes `current.trigger.run.data`)
+- The README describes `@trigger_on_finish` but the actual code uses `@project_trigger` -- the event-based pattern was adopted later