Add bootstrap and worker metadata for zero-discovery campaign initialization

.github/aw/bootstrap-agentic-campaign.md

@@ -0,0 +1,154 @@
+# Bootstrap Instructions (Phase 0)
+
+This phase runs when discovery returns zero work items, providing initial work for the campaign to begin.
+
+{{ if .BootstrapMode }}
+## Bootstrap Strategy: {{ .BootstrapMode }}
+
+{{ if eq .BootstrapMode "seeder-worker" }}
+### Seeder Worker Dispatch
+
+When no work items are discovered, dispatch a seeder/scanner worker to discover initial work:
+
+**Worker ID**: `{{ .SeederWorkerID }}`
+
+**Payload**:
+```json
+{{ .SeederPayload }}
+```
+
+{{ if gt .SeederMaxItems 0 }}
+**Max Items**: {{ .SeederMaxItems }} (limit how many items the seeder returns)
+{{ end }}
+
+**Implementation Steps**:
+
+1. Check if discovery returned zero items (no worker outputs found)
+2. If zero items:
+   - Use the `dispatch_workflow` safe output to trigger the seeder worker
+   - Pass the campaign_id and configured payload
+   - Wait for the seeder worker to complete and create initial work items
+3. On next orchestrator run, the discovery step will find the seeder's outputs
+
+**Seeder Worker Contract**:
+- MUST accept `campaign_id` and `payload` inputs (standard worker contract)
+- MUST create discoverable outputs (issues, PRs, or discussions)
+- MUST apply the tracker label: `campaign:{{ .CampaignID }}`
+- SHOULD limit output count to configured max-items if provided
+- SHOULD use deterministic work item keys for idempotency
+
+{{ else if eq .BootstrapMode "project-todos" }}
+### Project Board Todo Items
+
+When no work items are discovered, read from the Project board's "{{ .TodoValue }}" column:
+
+{{ if .StatusField }}
+**Status Field**: `{{ .StatusField }}`
+{{ else }}
+**Status Field**: `Status` (default)
+{{ end }}
+
+{{ if .TodoValue }}
+**Todo Value**: `{{ .TodoValue }}`
+{{ else }}
+**Todo Value**: `Todo` (default)
+{{ end }}
+
+{{ if gt .TodoMaxItems 0 }}
+**Max Items**: {{ .TodoMaxItems }} (limit how many Todo items to process)
+{{ end }}
+
+{{ if .RequireFields }}
+**Required Fields**: {{ range $index, $field := .RequireFields }}{{ if $index }}, {{ end }}`{{ $field }}`{{ end }}
+  - Skip Todo items where any of these fields are empty
+{{ end }}
+
+**Implementation Steps**:
+
+1. Check if discovery returned zero items (no worker outputs found)
+2. If zero items:
+   - Query the Project board at `{{ .ProjectURL }}`
+   - Filter items where Status = "{{ .TodoValue }}"
+   {{ if .RequireFields }}- Skip items missing required fields{{ end }}
+   {{ if gt .TodoMaxItems 0 }}- Limit to {{ .TodoMaxItems }} items{{ end }}
+   - Select workers based on item metadata (see Worker Selection below)
+   - Dispatch appropriate worker workflows for each Todo item
+3. Update Project status to "In Progress" for dispatched items
+4. On next orchestrator run, the discovery step will find the worker outputs
+
+**Project Item to Payload Mapping**:
+- Read Project field values from the Todo item
+- Map to worker payload schema based on worker metadata
+- Include campaign_id in every worker dispatch
+- Use the item's URL or number as the work_item_id
+
+{{ else if eq .BootstrapMode "manual" }}
+### Manual Bootstrap
+
+No automatic bootstrap configured. Wait for manual work item creation:
+
+- Work items should be created manually (issues, PRs, or discussions)
+- All items MUST have the tracker label: `campaign:{{ .CampaignID }}`
+- Items MUST follow the worker output labeling contract
+- Once items exist, the orchestrator will discover them normally
+
+{{ end }}
+{{ end }}
+
+---
+
+## Worker Selection
+
+{{ if .WorkerMetadata }}
+When dispatching workers during bootstrap, use deterministic selection:
+
+{{ range $index, $worker := .WorkerMetadata }}
+### Worker {{ add1 $index }}: {{ .ID }}
+
+**Capabilities**: {{ range $capIndex, $cap := .Capabilities }}{{ if $capIndex }}, {{ end }}`{{ $cap }}`{{ end }}
+
+**Payload Schema**:
+{{ range $fieldName, $fieldDef := .PayloadSchema }}- `{{ $fieldName }}` ({{ .Type }}{{ if .Required }}, required{{ end }}): {{ .Description }}
+{{ end }}
+
+**Output Labeling**:
+{{ if .OutputLabeling.Labels }}- Labels: {{ range $labelIndex, $label := .OutputLabeling.Labels }}{{ if $labelIndex }}, {{ end }}`{{ $label }}`{{ end }}
+{{ end }}- Key in Title: {{ .OutputLabeling.KeyInTitle }}
+{{ if .OutputLabeling.KeyFormat }}- Key Format: `{{ .OutputLabeling.KeyFormat }}`
+{{ end }}
+- Campaign tracker label applied automatically: `campaign:{{ $.CampaignID }}`
+
+**Idempotency Strategy**: {{ .IdempotencyStrategy }}
+
+{{ if .Priority }}**Priority**: {{ .Priority }} (higher = preferred when multiple workers match)
+{{ end }}
+
+{{ end }}
+
+**Selection Algorithm**:
+
+1. For each work item, check which workers can handle it:
+   - Match work item type/metadata to worker capabilities
+   - Check if worker's payload schema requirements can be satisfied
+2. If multiple workers match:
+   - Select the worker with highest priority
+   - If priorities are equal, select first alphabetically by ID
+3. Build payload from work item metadata according to worker's payload schema
+4. Dispatch worker with campaign_id and constructed payload
+
+{{ else }}
+**Note**: No worker metadata configured. Use workflow IDs from campaign spec for dispatch.
+{{ end }}
+
+---
+
+## Bootstrap Success Criteria
+
+After bootstrap completes:
+
+1. ✅ At least one worker workflow dispatched (or manual items created)
+2. ✅ All dispatched items will have proper tracker labels
+3. ✅ Next orchestrator run will discover >= 1 work item
+4. ✅ Campaign transitions from bootstrap phase to normal operation
+
+**Idempotency**: Bootstrap only runs when discovery = 0. Once work items exist, normal orchestration takes over.