n8n-Audit-Workflow

KO | EN

gitlite — search

#python #ai #android #machine-learning #linux #docker #nodejs #windows #awesome #cli #awesome-list #vue #ios #mcp

n8n-Audit-Workflow

(★ 14)

A reusable auditing tool for n8n that automatically analyses workflows for security issues, performance risks, error handling, readability, and AI usage.

n8n-Audit-Workflow Latest Version Download

Download Latest Version (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# n8n Workflow Auditor

A reusable auditing tool for n8n that automatically analyses workflows for security issues, performance risks, AI usage, error handling gaps, and naming clarity.

It generates a **Markdown or PDF report** that can be consumed directly, published (e.g., GitHub Gist, Slack, Confluence), or enriched by an LLM for narrative summarisation.

## 👀 Example Output
- [Report Example 1 (AI Enabled in Workflow)](https://github.com/christinec-dev/n8n-Audit-Workflow/blob/main/audit_report_using_ai.md)  
- [Report Example 2 (AI Disabled in Workflow)](https://github.com/christinec-dev/n8n-Audit-Workflow/blob/main/audit_report_no_ai.md)

---

## 🌟 Features
- **Security Analysis**
  - Detects leaked secrets, API keys, bearer tokens, and weak passwords  
  - Flags company identifiers configured in the `Configuration` node  
  - Reports on credential usage (hardcoded, OAuth, predefined, generic)

- **Performance Analysis**
  - Identifies **bottlenecks** (slow nodes vs workflow average)  
  - Flags **unstable nodes** (high runtime variance)  
  - Detects **disabled** and **never-executed** nodes  
  - Reports per-node execution stats (avg runtime, stdev, samples, dominant share)

- **AI/LLM Usage Tracking**
  - Audits all AI nodes (`@n8n/n8n-nodes-langchain`)  
  - Extracts model, preview of prompts, token estimates, execution times  
  - Provides counts: active, disabled, and total AI nodes

- **Error Handling Analysis**
  - Flags nodes with potential runtime failures (APIs, DB queries, LLM calls, custom code)
  - Adds per-node recommendations (retry on failure, continue on error, fallback models, polling)
  - Includes global golden rules for error handling (Error Workflows, retries, notifications, failover strategies)
 
- **Node Naming Audit**
  - Detects nodes left with default names (e.g., Node1, Node2) or unclear short names
  - Recommends descriptive, consistent naming to improve maintainability and debugging

- **Reporting**
  - Outputs **Markdown** (`audit_report.md`)  
  - Optionally generates **PDF** (`audit_report.pdf`)  
  - Can publish reports as **GitHub Gists** (default) or send to Slack/Confluence/etc.  
  - Supports **AI summarisation** for narrative reports (via OpenAI)

---

## ⚙️ Setup

### View visual setup guide [here](https://github.com/christinec-dev/n8n-Audit-Workflow/blob/main/SETUP.md).

### 1. Import Workflow
Load the provided `n8n Workflow Auditor.json` into your n8n instance. Do this in its own workflow.

### 2. Configure Settings
Open the **Configuration** node and adjust:
- `maxExecutions` → number of past runs to analyse (default: `5`)  
- `bottleneckThreshold` → multiplier over average to flag bottlenecks (default: `1.5`, should be fine)  
- `unstableThreshold` → stdev/avg ratio to flag instability (default: `0.5`, should be fine)  
- `shouldAuditSubflows` → include subflows (`true`/`false`)  
- `companyIdentifiers` → comma-separated sensitive terms you want flagged (e.g. `AcmeCorp,ProjectX`)  
- `shouldUseAIReporting` → `true` = LLM summarisation · `false` = scripted Markdown

### 3. Connect Credentials
Ensure the following credentials are configured in n8n:
- **n8nApi** → access your instance settings and generate an API key (used by `Get Executions`)  
- **openAiApi** (or another LLM) → for AI summarisation (if enabled)  
- **httpCustomAuth** + **githubApi** → for publishing reports as GitHub Gists (optional)

(You can replace this step with Slack, Confluence, or ConvertAPI.)

### 4. Select Target Workflow
In the **Get Latest Successful Executions of Workflow** node, set the `workflowId` of the workflow you want to audit.  
Optionally configure **Get Latest Successful Executions of Subworkflows** if analysing subflows.

### 5. Run the Audit
Trigger the workflow manually (**Test workflow**) to run the audit:
1. Fetches the last *N* executions  
2. Runs **security, performance, and AI audits**  
3. Generates a **Markdown/PDF report**

### 6. View Reports
- Default: uploaded as a private **GitHub Gist**  
- Alternative: sent to Slack/Confluence/ConvertAPI (replace output nodes)  
- PDF generation is included if enabled in the final stage

---

## 📜 Report Contents
Each report includes:
- **Glossary** of audit terms  
- **Global Summary** (workflows analysed, issues by severity)  
- **Per-workflow breakdown**:
  - Security Findings (tables of issues)  
  - Performance Findings (slow nodes, bottlenecks, disabled/unused)  
  - API & Credential Usage (node + credential type)  
  - AI Audit Findings (model, prompt preview, token estimate, runtime)
  - Error Handling Findings (node-level risks + actionable fixes)  
  - Recommendations  
- **Overall Recommendations** (security, performance, AI risks)

---

## ⚖️ Scaling Considerations

Running the **n8n Workflow Auditor** involves fetching past executions, parsing node data, and generating large Markdown/AI reports. This can be **resource intensive** depending on how many executions you analyse and the size of your workflows.
 
### Instance Sizing (if using AWS)
- **t2.micro / t2.small (1 GB RAM, 1 vCPU)** → *Not recommended*. Workflows with many executions or AI summarisation will likely crash or hang due to memory/CPU limits.  
- **t3.small (2 GB RAM, 2 vCPU burstable)** → Minimum viable size for small audits.  
- **t3.medium (4 GB RAM, 2 vCPU)** → Recommended for production use with AI summarisation enabled.  
- **t3.large+ (8 GB RAM, 2+ vCPU)** → For teams auditing multiple workflows or large histories.

### Execution Limits
- Use the `maxExecutions` setting in the **Configuration** node to cap how many runs are analysed (default: `5`).  
- Best practice is **20–50 executions per audit**. Avoid loading hundreds or thousands at once.

### Memory Optimisation
- Drop unnecessary fields (avoid storing full execution logs).  
- Aggregate metadata only (node name, runtime, errors).  
- Slice arrays (`executions.slice(0, N)`) when iterating in custom code.

### AI Summarisation
- AI prompts can grow large when embedding full JSON results.  
- Keep JSON payloads lean (remove fields not needed for reporting).  
- If using AI reporting, ensure at least **t3.medium (4 GB RAM)** to prevent out-of-memory crashes.

### Stability Enhancements
- Enable **swap space** on smaller instances as a buffer (2 GB recommended).  
- Separate heavy audits into dedicated workflows, isolating them from production automations.  
- Consider offloading summarisation (AI Writer) to a serverless function or bigger host if costs allow.

---

## 🛠️ Troubleshooting

Running audits can sometimes stress small EC2 instances or misconfigured n8n environments.  
Here are common issues and fixes:

### 1. Workflow Crashes / EC2 Instance Freezes
**Symptom:** The whole n8n process or EC2 instance stops responding during an audit.  
**Cause:** Out-of-memory (OOM) kill on small instances (`t2.micro`, `t2.small`). The auditor loads too many executions into memory.  
**Fix:**
- Upgrade to at least `t3.small` (2 GB RAM) or `t3.medium` (4 GB RAM).  
- Limit executions in **Configuration → `maxExecutions`** (20–50 recommended).  
- Add swap space (2 GB) for stability buffer.

---

### 2. Workflow Hangs at AI Writer
**Symptom:** The audit gets stuck when generating the AI summary.  
**Cause:** Prompt too large (full JSON > 100s of KB) or instance CPU throttling.  
**Fix:**
- Reduce execution history (`maxExecutions`).  
- Trim unneeded fields from JSON before passing to AI.  
- Ensure sufficient memory (4 GB+).

---

### 3. Database Bloat
**Symptom:** SQLite/Postgres DB grows very large, queries slow down.  
**Cause:** n8n stores full workflow execution logs, including audit payloads.  
**Fix:**
- Use **SQLite vacuum** or Postgres `VACUUM FULL`.  
- Configure n8n to prune old executions:  
  ```env
  EXECUTIONS_DATA_PRUNE=true
  EXECUTIONS_DATA_MAX_AGE=336  # keep 14 days (hours)
  EXECUTIONS_DATA_PRUNE_HARD_DELETE=true

---

## 🔮 Roadmap
- Add inline PDF charts for runtime distribution  
- Expand AI audit with cost estimation (tokens × pricing)  
- Extend destinations: Jira, Notion, Google Drive
- Add performance issues time blocks to identify when performance dips during morning, evening, or afternoon
- Subflows identification within sublows. Currently only main workflow + subflows of main workflow is being audited
- Improved report formatting, for both manual and AI generated reports

n8n-Audit-Workflow

n8n-Audit-Workflow Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

n8n-Audit-Workflow

n8n-Audit-Workflow Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

Download README