/n8n — automation-stack health & repair

🧒 When reporting to Anton end with a child-simple "In plain words" recap. (memory eli5-always)

Anton's operational nervous system = self-hosted n8n at https://n8n.example.com (v2.14.2, 89 workflows). n8n = event bus, NOT the brain. Full state + every gotcha = memory [[n8n-stack]] (read it first). Workflows fail SILENTLY → this skill makes failure visible and fixes it safely.

Access: REST https://n8n.example.com/api/v1/, header X-N8N-API-KEY; key in $USERPROFILE/.claude/secrets/n8n.env ⚠️ expires 2026-07-15 (recreate non-expiring; warn if near). API needs a browser User-Agent header or returns 403 (WAF). The n8n MCP (mcp__n8n__*) is wired BUT can drop mid-session (it disconnected 2026-07-04) — do NOT depend on it.

MCP-down fallback = build/edit straight over REST (proven 2026-07-04): $IMPORTS_ROOT/n8n\n8n_build.py — create_workflow(name,nodes,connections), post_webhook(path,payload) (fire a webhook for /tt), set_active(id,on); and n8n_edit.py — get_workflow/update_workflow (backup-before-PUT), list_workflows, _req. Both reuse the WAF User-Agent. The resilience workflows built this way — WATCHDOG IdSRNxfOT9Mo7Qse (dead-man's-switch), ERROR CATCHER 5X7xsGlwz44AO8Wf (set as settings.errorWorkflow to make a workflow's failures scream in chat 03), INBOUND DOOR 5e8JGjnVSvnkrRSX, bus-bridge U96y7qLDLufGpbUf — plus their builders (watchdog_build.py/catcher_build.py/inbound_build.py) live in _imports\n8n\. Full map = memory [[n8n-watchdog]].

Step 0 — key liveness (run BEFORE anything else)

First check the API key is still alive: compare today's date to N8N_KEY_EXPIRES in n8n.env (see Access above for the date). If today is past it the key is DEAD → every call silently 403s, so recreate the key first (n8n UI → Settings → API → new key, paste into n8n.env, bump N8N_KEY_EXPIRES) — a Tier-2 write, so pause for Anton's "+" (see Safety). Within ~7 days → warn but proceed.

Modes

• health (default, READ-ONLY) — mcp__n8n__n8n_health_check for liveness, then scan recent executions for errors (mcp__n8n__n8n_executions status=error, or python $IMPORTS_ROOT/n8n/fetch_n8n.py). Report RED (persistent code-level failures), FLAKY (transient 503/aborted → just enable node auto-retry), and IDLE-but-active (cron not firing). ALWAYS prove a failure from a real execution (resultData.runData[node].error) before naming a cause — the failing node may differ from the first suspect.
• dashboard — python $IMPORTS_ROOT/n8n/build_dashboard.py → open _Dashboards\n8n-Automation-Audit.html (Anton works by eye). Re-run fetch_n8n.py → enrich_n8n.py → build_dashboard.py to refresh (all 0 LLM tokens; cluster summaries on FREE Sonnet subagents per [[model-routing-sonnet-grunt]]).
• fix <id> — repair ONE workflow. BACKUP FIRST always: python $IMPORTS_ROOT/n8n/n8n_edit.py backup-all (→ raw/backup_<ts>/; per-edit before/after to raw/edits/, restore via restore_workflow). Show Anton the diff (BEFORE→AFTER), get his "+", then apply. PUT needs ONLY {name,nodes,connections,settings} (strip read-only fields).

Known scars (verify still true before acting — memory may be stale)

• Events Posting — was name '_input' is not defined (Python task-runner bug); FIXED 2026-06-16 by converting Code nodes Python→JavaScript. Rule: when a Python code node mysteriously fails on _input, convert to JS rather than debugging Pyodide.
• Gitbook Pavel — was 11% err on empty-sessionId; FIXED with an additive guard Filter before the memory chain.
• Codex cron / aaaZeroInbound — abandoned/standalone (awaiting Anton's archive call).

Safety

• Health = read-only, run freely. Editing/activating a workflow, recreating the API key, deleting workflows = pause + confirm (Tier-2, [[operating-agreement]]).
• Backup BEFORE every edit (backup-all); never edit without the snapshot.

Output

RED/FLAKY/IDLE counts + the worst offender with its proven error + suggested fix. Then 🧒 recap. Related: [[platinum-crm-import]] (CHARM CRM engine), [[automation-inventory]] (that = LOCAL jobs; this = REMOTE n8n).

<!--kit-footer-->

---

Like this skill? It is part of second-brain-starter-kit: the second brain we built for ourselves and run every day at Palo Alto AI Research Lab. Install the whole set with npx skills add tonydzi/second-brain-starter-kit. Everything is open source and free, so take what you need.

Flagships worth a look on their own: secondop-panel (a second opinion from a panel of external models), claude-memory-tidy (stop your agent's memory from rotting), telegram-mcp-kit (your own Telegram over MCP in about 15 minutes).

Author: Anton Dziatkovskii, Palo Alto AI Research Lab. Telegram @tonydzi - WhatsApp +1 341 222 9178 - X @Tony_Stef_

Engineers: want to test-drive this setup? Message me. I hand out free starter seeds to engineers who test and report back, and custom skill requests are welcome.