feat: add n8n-workflow-manager skill for autonomous workflow fix/validate/test loop

.agent/skills/n8n-workflow-manager/SKILL.md

+---
+name: n8n-workflow-manager
+description: "Autonomous n8n workflow management: validate, auto-fix, deploy, and test workflows using n8n MCP tools in a self-healing loop. Handles connection fixing, node version compatibility, credential mapping, and execution testing. Use when: n8n workflow, fix workflow, deploy workflow, test workflow, n8n connections, n8n validation."
+---
+
+# n8n Workflow Manager
+
+**Role**: Autonomous n8n Workflow Operations Engineer
+
+You are an expert in managing n8n workflows programmatically using the n8n MCP tools. You can validate, fix, deploy, and test workflows autonomously, running in a loop until everything passes. You understand n8n's node types, connection types (main, ai_languageModel, ai_tool, ai_memory), version compatibility, and credential management.
+
+## Core Loop: Fix → Validate → Test → Repeat
+
+When asked to manage, fix, or deploy an n8n workflow, follow this autonomous loop:
+
+```
+┌─────────────┐
+│  1. INSPECT  │ ← Get workflow structure
+└──────┬──────┘
+       ▼
+┌─────────────┐
+│ 2. VALIDATE │ ← Run validation, identify issues
+└──────┬──────┘
+       ▼
+┌─────────────┐
+│  3. FIX     │ ← Auto-fix connections, types, versions
+└──────┬──────┘
+       ▼
+┌─────────────┐
+│ 4. VERIFY   │ ← Re-validate, confirm fixes
+└──────┬──────┘
+       ▼
+┌─────────────┐
+│  5. TEST    │ ← Execute workflow, check results
+└──────┬──────┘
+       ▼
+    Pass? ──No──→ Go to step 2
+       │
+      Yes
+       ▼
+    ✅ DONE
+```
+
+## Step 1: Inspect Workflow
+
+Always start by getting the full workflow structure:
+
+```
+# Get workflow overview
+n8n_get_workflow(id="WORKFLOW_ID", mode="structure")
+
+# For full details including parameters
+n8n_get_workflow(id="WORKFLOW_ID", mode="full")
+```
+
+**Checklist when inspecting:**
+- [ ] Count total nodes and connections
+- [ ] Identify AI nodes (agent, chainLlm, lmChatOpenAi, lmChatGoogleGemini)
+- [ ] Identify Tool nodes (httpRequestTool, toolHttpRequest, toolCode)
+- [ ] Check for missing connections (especially ai_languageModel, ai_tool)
+- [ ] Verify node typeVersions match the target n8n version
+- [ ] Check credential assignments
+
+## Step 2: Validate Workflow
+
+```
+# Validate by workflow ID
+n8n_validate_workflow(id="WORKFLOW_ID")
+
+# Or validate workflow JSON directly
+validate_workflow(workflow={...})
+```
+
+**Common validation errors:**
+| Error               | Cause                          | Fix                                     |
+| ------------------- | ------------------------------ | --------------------------------------- |
+| Node not installed  | typeVersion too new for server | Downgrade typeVersion                   |
+| Missing connection  | AI node orphaned               | Add ai_languageModel/ai_tool connection |
+| Invalid structure   | Node type changed              | Swap node type                          |
+| Missing credentials | Import/deploy                  | Remap credentials in UI                 |
+
+## Step 3: Auto-Fix Patterns
+
+### Fix Missing AI Connections
+
+AI connections are the #1 issue when importing/deploying workflows. They get lost because n8n's paste/import only preserves `main` connections.
+
+```
+# Connect language model to AI Agent
+n8n_update_partial_workflow(
+  id="WORKFLOW_ID",
+  operations=[
+    {type: "addConnection", source: "OpenAI Chat Model", target: "AI Agent", sourceOutput: "ai_languageModel"},
+    {type: "addConnection", source: "Tool Node", target: "AI Agent", sourceOutput: "ai_tool"}
+  ]
+)
+```
+
+**Connection types for AI nodes:**
+| Source Node Type                           | Connection Type    | Target          |
+| ------------------------------------------ | ------------------ | --------------- |
+| lmChatOpenAi, lmChatGoogleGemini           | `ai_languageModel` | agent, chainLlm |
+| httpRequestTool, toolHttpRequest, toolCode | `ai_tool`          | agent           |
+| memoryBufferWindow                         | `ai_memory`        | agent           |
+| outputParserStructured                     | `ai_outputParser`  | agent           |
+
+### Fix Version Compatibility
+
+When deploying from newer n8n to older, node types and typeVersions must be downgraded:
+
+**n8n v2.x → v1.x mappings:**
+| Node              | v2.x Type                                    | v1.x Type                                       |
+| ----------------- | -------------------------------------------- | ----------------------------------------------- |
+| HTTP Request Tool | `n8n-nodes-base.httpRequestTool` v4.4        | `@n8n/n8n-nodes-langchain.toolHttpRequest` v1.1 |
+| AI Agent          | `@n8n/n8n-nodes-langchain.agent` v3.1        | `@n8n/n8n-nodes-langchain.agent` v1.6           |
+| Chain LLM         | `@n8n/n8n-nodes-langchain.chainLlm` v1.9     | `@n8n/n8n-nodes-langchain.chainLlm` v1.4        |
+| OpenAI Chat       | `@n8n/n8n-nodes-langchain.lmChatOpenAi` v1.3 | v1.2                                            |
+| HTTP Request      | `n8n-nodes-base.httpRequest` v4.2            | v4.1                                            |
+| Google Sheets     | `n8n-nodes-base.googleSheets` v4.7           | v4.5                                            |
+| Set               | `n8n-nodes-base.set` v3.4                    | v3.3                                            |
+
+```
+# Downgrade node typeVersions
+n8n_update_partial_workflow(
+  id="WORKFLOW_ID",
+  intent="Downgrade node typeVersions for v1.x compatibility",
+  operations=[
+    {type: "updateNode", nodeName: "AI Agent", updates: {typeVersion: 1.6}},
+    {type: "updateNode", nodeName: "Tool Node", updates: {type: "@n8n/n8n-nodes-langchain.toolHttpRequest", typeVersion: 1.1}}
+  ]
+)
+```
+
+### Fix Missing Main Connections
+
+```
+# Add standard flow connections
+n8n_update_partial_workflow(
+  id="WORKFLOW_ID",
+  operations=[
+    {type: "addConnection", source: "Trigger", target: "Process"},
+    {type: "addConnection", source: "Process", target: "Output"}
+  ]
+)
+```
+
+### Fix via Autofix
+
+```
+# Let n8n MCP auto-detect and fix common issues
+n8n_autofix_workflow(id="WORKFLOW_ID", applyFixes=true)
+```
+
+### Clean Stale Connections
+
+```
+# Remove broken refs after node deletions/renames
+n8n_update_partial_workflow(
+  id="WORKFLOW_ID",
+  operations=[{type: "cleanStaleConnections"}]
+)
+```
+
+## Step 4: Verify Fixes
+
+After fixing, always re-validate:
+
+```
+# Re-validate
+n8n_validate_workflow(id="WORKFLOW_ID")
+
+# Check structure
+n8n_get_workflow(id="WORKFLOW_ID", mode="structure")
+```
+
+**Verify checklist:**
+- [ ] All AI nodes have their language model connected
+- [ ] All Tool nodes connect to AI Agent via ai_tool
+- [ ] All main flow connections are intact (trigger → process → output)
+- [ ] No "?" nodes in UI (check via browser if available)
+- [ ] Node typeVersions compatible with target n8n version
+
+## Step 5: Test Execution
+
+```
+# Test via webhook trigger
+n8n_test_workflow(workflowId="WORKFLOW_ID", triggerType="webhook", data={...})
+
+# Check execution results
+n8n_executions(action="list", workflowId="WORKFLOW_ID", status="error", limit=5)
+
+# Get error details
+n8n_executions(action="get", id="EXECUTION_ID", mode="error")
+```
+
+## Deployment Checklist
+
+When deploying a workflow to a new n8n instance:
+
+1. **Export** workflow JSON from source
+2. **Check n8n version** of target (health check or UI)
+3. **Import** workflow (paste JSON or use API)
+4. **Fix connections** — AI connections ALWAYS get lost on import
+5. **Fix node versions** — if source is newer than target
+6. **Map credentials** — credential IDs differ between instances
+7. **Update URLs** — `localhost` → production URLs (e.g. `host.docker.internal`)
+8. **Validate** — run full validation
+9. **Test** — execute and verify results
+10. **Activate** — enable schedule triggers for production
+
+## REST API Fallback
+
+When n8n MCP doesn't connect to the target instance, use direct REST API calls:
+
+```python
+import json, urllib.request, ssl
+
+API_KEY = "your-api-key"
+BASE_URL = "https://your-n8n.com/api/v1"
+WORKFLOW_ID = "your-workflow-id"
+
+ctx = ssl.create_default_context()
+
+# GET workflow
+req = urllib.request.Request(
+    f"{BASE_URL}/workflows/{WORKFLOW_ID}",
+    headers={"X-N8N-API-KEY": API_KEY}
+)
+wf = json.loads(urllib.request.urlopen(req, context=ctx).read())
+
+# Modify connections
+wf['connections']['OpenAI Chat Model'] = {
+    'ai_languageModel': [[{'index': 0, 'node': 'AI Agent', 'type': 'ai_languageModel'}]]
+}
+
+# PUT updated workflow
+data = json.dumps({
+    'name': wf['name'], 'nodes': wf['nodes'],
+    'connections': wf['connections'], 'settings': wf.get('settings', {})
+}, ensure_ascii=False).encode()
+
+req2 = urllib.request.Request(
+    f"{BASE_URL}/workflows/{WORKFLOW_ID}",
+    data=data, method="PUT",
+    headers={"X-N8N-API-KEY": API_KEY, "Content-Type": "application/json"}
+)
+urllib.request.urlopen(req2, context=ctx)
+```
+
+## Anti-Patterns
+
+### ❌ Import and Assume It Works
+**Why bad**: AI connections ALWAYS break on import. Tool nodes may use incompatible types.
+**Instead**: Always run the Fix → Validate → Test loop after every import.
+
+### ❌ Manual UI Connection Dragging for AI Nodes
+**Why bad**: Fragile, error-prone, doesn't scale.
+**Instead**: Use `n8n_update_partial_workflow` with `addConnection` operations.
+
+### ❌ Ignoring typeVersion Mismatches
+**Why bad**: Nodes show "?" and workflow won't execute.
+**Instead**: Always check target n8n version and downgrade typeVersions if needed.
+
+### ❌ Hardcoding Credential IDs
+**Why bad**: Credential IDs differ between n8n instances.
+**Instead**: Let n8n match credentials by name, or remap in UI after deploy.
+
+## MCP Tools Quick Reference
+
+| Task            | Tool                          | Key Params                   |
+| --------------- | ----------------------------- | ---------------------------- |
+| List workflows  | `n8n_list_workflows`          | `active`, `tags`             |
+| Get workflow    | `n8n_get_workflow`            | `id`, `mode`                 |
+| Validate        | `n8n_validate_workflow`       | `id`                         |
+| Auto-fix        | `n8n_autofix_workflow`        | `id`, `applyFixes`           |
+| Partial update  | `n8n_update_partial_workflow` | `id`, `operations`           |
+| Full update     | `n8n_update_full_workflow`    | `id`, `nodes`, `connections` |
+| Test/trigger    | `n8n_test_workflow`           | `workflowId`, `triggerType`  |
+| Executions      | `n8n_executions`              | `action`, `workflowId`       |
+| Search nodes    | `search_nodes`                | `query`                      |
+| Node info       | `get_node`                    | `nodeType`                   |
+| Deploy template | `n8n_deploy_template`         | `templateId`                 |
+| Version history | `n8n_workflow_versions`       | `mode`, `workflowId`         |
+
+## Related Skills
+
+Works well with: `ai-agents-architect`, `fastapi-pro`, `langgraph`