n8n Workflow Library
A local collection of 2,061 production n8n workflows across 188 node/integration folders. Search it before writing any n8n workflow from scratch — a close match almost always exists and adapting one is faster and more reliable than authoring blind.
Library root: ~/.claude/skill-repos/n8n-workflows/workflows/
How the library is organised
workflows/<Folder>/<id>_<Service1>_<Service2>_<Purpose>_<Trigger>.json
Example: workflows/Airtable/1120_Airtable_Mindee_Automate_Webhook.json
Two things about folder names that will trip you up if you assume otherwise:
- Folders are named after the primary node type, not always the app. That is why the largest folders are
Manual(391),Splitout(194),Code(183),Http(176) — these are structural nodes, not integrations. App folders likeTelegram(119) andGooglesheets(26) also exist. - Because of that, never rely on the folder alone. Always search filenames across all folders.
Trigger suffixes tell you how a workflow starts: Webhook (997), Triggered (735), Scheduled (275), plus a few Automate, Create, Send, Process.
Search strategy — cheapest first
1. Filename search across everything. Fastest and usually enough. Case-insensitive, matches service names.
cd ~/.claude/skill-repos/n8n-workflows/workflows
ls */*.json | grep -i slack
ls */*.json | grep -i -e telegram -e openai # either service
ls */*.json | grep -i slack | grep -i sheets # both services
ls */*.json | grep -i "_Scheduled\.json$" # by trigger type
2. Node-type search when the filename does not say enough. Workflow names do not always list every node, so grep the JSON for the actual node type.
grep -rl "n8n-nodes-base.slack" . | head -20
grep -rl "n8n-nodes-base.openAi" . | head -20 # AI workflows (32 files)
3. Content search for a concept — an API host, a field name, a specific operation.
grep -rli "hubspot" . | head -20
This collection largely predates n8n's LangChain node namespace — only 6 files use n8n-nodes-langchain. AI workflows here are built on n8n-nodes-base.openAi. If the user wants a modern AI Agent workflow, expect to build it rather than find it.
Report how many matches you found before narrowing. If a search returns nothing, say so plainly and offer to build the workflow from scratch rather than forcing a poor match.
Reading a workflow
Files are standard n8n exports: name, nodes, connections, settings, tags.
Do not dump a whole JSON into the conversation — some run to thousands of lines. Summarise the shape first:
python3 -c "
import json
d = json.load(open('WORKFLOW.json'))
print(d['name'], '—', len(d['nodes']), 'nodes')
for n in d['nodes']:
print(' ', n['type'].replace('n8n-nodes-base.',''), '·', n['name'])
"
Then read only the nodes that matter for the user's question.
Delivering a workflow
Give the user the complete JSON as a file they can import, not pasted into chat unless it is small. Then tell them: n8n → Workflows → Import from File.
Always state, without being asked:
- Credentials do not transfer. Nodes reference credential IDs from the original author's instance. The user must re-select their own credentials on every node that touches an external service.
- These are community-contributed. Read what a workflow does before running it, especially anything with
httpRequestnodes hitting unfamiliar hosts, orCodenodes. - Any webhook URLs, spreadsheet IDs, channel names, or hardcoded IDs are the original author's and need replacing.
When you adapt a workflow, say exactly what you changed from the original.
Combining and building
For a request no single workflow covers, it is usually right to take the trigger section from one workflow and the action section from another. Say which files you drew from so the user can check your sources.
Keep node name values unique inside a workflow — connections is keyed by node name, so duplicates silently break the graph.
Updating the library
git -C ~/.claude/skill-repos/n8n-workflows pull
This skill lives outside that clone, so pulling never overwrites it.