Auto-fix ClickUp bugs with OpenAI planning and Cursor execution in n8n

ClickUp webhook → OpenAI analysis with scoring → optional Slack approval → Cursor background agent → GitHub PR creation → ClickUp status update

Who

DevOps leads or engineering managers who want to automate bug-fix pipelines for their teams

You get

When a ClickUp bug report arrives, automatically generate a fix plan, optionally require human approval, let Cursor implement and test the fix, open a GitHub PR, and update the task status

You need

n8n · slack · github

Steps

1. Configure webhook to receive bug reports

   Create a Webhook node in n8n with path clickup-bug-autofix, method POST, and responseMode set to responseNode. In ClickUp, configure an automation to send new bug tasks to this webhook URL, ensuring the payload includes taskId, taskName, description, stackTrace, repoUrl, repoBranch, priority, and listId.

2. Prepare normalized bug report data

   Use the Set node (Prepare Bug Report Data) to normalize the webhook payload into a configuration object. Key mappings: bugId ← taskId, taskName ← taskName/name (default 'Untitled bug'), description, stackTrace, repoUrl, repoBranch (default 'main'), priority (default 'normal'), reporter (default 'unknown'), clickupListId ← listId. Also set workflow constants: llmModel='gpt-4.1', autoFixConfidenceThreshold=0.75, cursorApiUrl='https://api.cursor.com/v0/background-agents', cursorApiModel='cursor-fast', cursorCallbackWebhookUrl (replace with actual wait-node resume URL), githubOwner, githubRepo, testCommand (default 'npm test'), slackChannel (default '#eng-bugfixes').

3. Analyze bug with OpenAI and generate a fix plan

   Connect the OpenAI Chat Model node (gpt-4.1) to the AI Agent node (Bug Fix Planner Agent). The AI Agent uses define mode with a system prompt that instructs the model to analyze root cause, affected files, fix approach, complexity, confidence, test plan, and risk flags, returning strict JSON only.

   Execution prompt · Source text。Bug Fix Planner Agent node text field (define mode)

   Copy prompt

   =You are a senior software engineer diagnosing a bug reported in ClickUp task {{ $json.bugId }} for repository {{ $json.repoUrl }} (branch {{ $json.repoBranch }}).
   
   Bug title: {{ $json.taskName }}
   Priority: {{ $json.priority }}
   
   Description:
   {{ $json.description }}
   
   Stack trace / logs (if any):
   {{ $json.stackTrace }}
   
   Analyze this bug and determine:
   - rootCause: the underlying cause of the bug, in plain language
   - affectedFiles: array of specific file paths in the repository most likely involved in the fix
   - fixApproach: a concrete description of the code change strategy to resolve it
   - fixComplexity: one of low, medium, high
   - confidence: a number from 0 to 1 estimating how confident you are that an automated coding agent could safely implement this fix without human intervention
   - testPlan: which existing tests should verify the fix, and any new test cases that should be added
   - riskFlags: array of short strings flagging any sensitive areas this fix touches (e.g. "auth", "payment", "database migration", "encryption", "user permissions") — empty array if none apply
   
   Return ONLY a JSON object (no markdown fences, no commentary) with exactly these keys: rootCause, affectedFiles, fixApproach, fixComplexity, confidence, testPlan, riskFlags.

4. Validate output and determine auto-fix eligibility

   Use the Code node (Analyze Auto-Fix Eligibility) to parse the AI output, strip markdown fences, validate JSON structure, normalize complexity (low/medium/high, default medium) and confidence (clamped 0–1). Extract risk flags and check against high-risk keywords (auth, payment, billing, migration, schema, secret, encryption, permission). Eligibility gate: confidence >= autoFixConfidenceThreshold (0.75) AND complexity !== 'high' AND no high-risk area touched AND affectedFiles.length > 0. Generate fixId = 'FIX-' + bugId + '-' + Date.now(), output full analysis with autoFixEligible boolean.

5. Approval branch: auto-pass or Slack manual review

   The If node checks autoFixEligible. If true, proceed directly to Cursor execution. If false, the Slack node sends a review message to the configured channel with bug summary, confidence, complexity, risk flags, root cause, fix approach, and affected files; then the Wait node pauses the workflow until a developer replies via webhook to resume.

   Execution prompt · Source text。Notify Dev on Slack for Review node text field

   Copy prompt

   =🛠️ *Bug needs manual review before auto-fix* — ClickUp task `{{ $json.bugId }}`: *{{ $json.taskName }}*
   
   Confidence: {{ $json.confidence }} | Complexity: {{ $json.fixComplexity }} | Risk flags: {{ $json.riskFlags.join(', ') || 'none' }}
   
   *Root cause:* {{ $json.rootCause }}
   *Proposed fix:* {{ $json.fixApproach }}
   *Affected files:* {{ $json.affectedFiles.join(', ') }}
   
   Reply to acknowledge and let the pipeline continue to Cursor, or handle this one manually instead.

6. Build Cursor task and trigger background agent

   The Code node (Create Fix Task Payload) merges the analysis with approval context: if auto-approved, reviewedBy is 'auto (confidence X >= threshold Y)'; if via Slack approval, reads reviewedBy and notes from the resume payload. Generates branch name autofix/{fixId}, builds the Cursor prompt (repository, bug, root cause, fix approach, affected files, test plan, branch and test command). The HTTP Request node POSTs JSON to cursorApiUrl with repository, baseBranch, branch, model, prompt, testCommand, callbackUrl, and metadata, using HTTP Header Auth credentials. Then the Wait node pauses for the Cursor callback.

7. Process Cursor result and create GitHub PR

   The Code node (Process Cursor Result for PR) parses the Cursor callback payload, extracting testsPassed, filesChanged, diffSummary, commitSha. If tests passed, status='ready_for_review'; else 'tests_failed'. Assembles PR title '[Auto-fix] {taskName} ({fixId})' and PR body (root cause, fix approach, files changed, reviewer, test results, diff summary). The GitHub node creates the PR (continueOnFail=true); if tests failed, the PR should be draft (verify draft parameter availability in your GitHub node version).

8. Update ClickUp and return response

   The ClickUp node updates the bug task status: 'in review' if tests passed, 'needs manual fix' if failed. The Respond to Webhook node returns a JSON summary with bugId, fixId, taskName, autoFixEligible, reviewedBy, branchName, status, testsPassed, prTitle, prUrl, and clickupStatus.

You are a senior DevOps engineer setting up the 'ClickUp Bug Auto-Fix' workflow from an n8n template. First confirm: 1) you have admin access to an n8n instance; 2) you have obtained credentials for OpenAI, Slack, ClickUp, GitHub, and the Cursor API; 3) ClickUp can configure webhooks pointing to your n8n instance. Then import workflow.json or rebuild nodes manually, paying special attention to constants in Prepare Bug Report Data, the prompt in Bug Fix Planner Agent, the threshold logic in Analyze Auto-Fix Eligibility, and the resume webhook URLs for both Wait nodes.