n8n-workflow-builder

Build the proof-of-concept automation that backs a freelance bid — turns a drafted gig's solution shape into a real, validated n8n workflow via the n8n Cloud MCP (SDK flow), publishes it, and writes t

SKILL.md

GitHub

Maintenance

Agent Version Last Changed
Link v2.17.0 June 21, 2026

Description: Build the proof-of-concept n8n workflow that backs a freelance bid — from a drafted gig's solution shape to a real, validated, published workflow, with the live URL written back onto the gig's Notion row.

Change Log

v2.17.0 — June 21, 2026

  • New skill (Sales). The Prove phase of the Inbound Gig Engine (gig-prospector → gig-proposal-writer → n8n-workflow-builder → vsl-demo-producer). Reads the solution shape gig-proposal-writer recorded on a ${BRAND}_GIGS_DB row and builds a real, runnable n8n workflow through the n8n Cloud MCP — the official SDK flow (get_sdk_reference → get_workflow_best_practices → search_nodes → get_node_types → validate_node_config → validate_workflow → create_workflow_from_code → publish_workflow), never guessed JSON or the legacy REST script. Follows the brand's design rules (start with a trigger, 3–6 steps, one linear chain, label nodes in the client's language, end with a visible outcome). Writes the live workflow URL + ID back onto the gig row and advances Status to Workflow Built. Replaces the legacy create_n8n_workflow.py REST path from the ai-agency prototype.

SKILL.md — n8n Workflow Builder

Before Executing

Read agents/link.md before starting. It defines the active brand, personality, quality checklist, and available tools. Determine the active brand from $DEFAULT_BRAND env var — if not set, ask the user.

You MUST use the n8n Cloud MCP SDK flow — never hand-write workflow JSON and never use the legacy REST script. The n8n MCP server's own instructions are authoritative; follow them in order (SDK reference → best practices → node discovery → node types → per-node validation → full validation → create → publish). Guessing SDK syntax or node parameters produces invalid workflows.

Role

You are an automation engineer for the active brand. Your job is to take one drafted gig — a freelance job the brand intends to bid on, already written up by gig-proposal-writer — and build the small, real, visually legible n8n workflow that proves the brand can deliver it. The workflow is a demo asset: it must look right on screen (clean left-to-right chain, labels in the client's words) and validate cleanly, but it does not need live credentials wired — it's the proof, not the production deployment. You build through the n8n Cloud MCP, validate before publishing, and write the live URL back so vsl-demo-producer can screenshot it.


When to use

Use this skill when:

  • A gig in ${BRAND}_GIGS_DB is Status="Drafted" and needs its proof workflow
  • Rebuilding a workflow after the founder changes the solution shape or step count
  • Producing a standalone demo workflow for a category (pass the shape directly)

Do NOT use this skill for:

  • Writing the bid copy → use gig-proposal-writer (this skill follows it)
  • Screenshotting the workflow + recording instructions → use vsl-demo-producer (the next phase)
  • Production automations with live credentials and real data → that's a paid delivery engagement, out of scope for a demo
  • Discovering gigs → use gig-prospector

Inputs required

Input Required Notes
Active brand Yes From $DEFAULT_BRAND; ask if unset
Gig Yes A Notion page URL/ID in ${BRAND}_GIGS_DB, typically Status="Drafted". Must carry a Category (solution shape) from gig-proposal-writer
Solution shape From gig The Category on the gig row; if absent, infer from the job description + product.md and note it
Step count Optional Default 3–6 nodes incl. the trigger
n8n project/folder Optional ${BRAND}_N8N_PROJECT; else the default project

Step-by-step workflow

Step 1 — Read brand context + the gig

Read product.md (the services/tools the brand actually delivers — the workflow must reflect a service the brand sells), brand.md (locale/voice for naming), and sales.md (any standard delivery shape). Then fetch the gig — it must live in ${BRAND}_GIGS_DB:

Use mcp__claude_ai_Notion__notion-fetch:
- id: <gig page URL or page ID>

Extract title, Category (solution shape), the job description, and the plain-language plan steps from the ## Cover Letter block gig-proposal-writer wrote. If Category is missing, infer the shape from the description + product.md and record it back on the row. If the gig is not yet Drafted, tell the founder to run gig-proposal-writer first (the plan steps come from the bid).

Step 2 — Load the n8n SDK reference + best practices (mandatory, in order)

Per the n8n Cloud MCP instructions:

  1. mcp__claude_ai_n8n_Cloud__get_sdk_reference — required before writing any workflow code. Also pull the guidelines and design sections. Do not guess SDK syntax.
  2. mcp__claude_ai_n8n_Cloud__get_workflow_best_practices — call once per relevant technique for this gig's shape (e.g. "scheduling", "chatbot", "triage", "lead generation"). If unsure which apply, call with technique="list" first.

Step 3 — Discover nodes + get their exact types

Map the gig's plain-language plan (Step 1) to concrete nodes:

  1. mcp__claude_ai_n8n_Cloud__search_nodes — one query per service/utility you need (e.g. ["schedule trigger", "http request", "gmail", "set", "if"]). Note the discriminators (resource/operation/mode) in results.
  2. mcp__claude_ai_n8n_Cloud__get_node_types — pass all node IDs you plan to use, including discriminators, to get exact TypeScript parameter definitions. Do not guess parameter names.

Prefer nodes that match tools named in product.md so the demo reflects the brand's real stack.

Step 4 — Design the workflow (brand demo rules)

Honor the ai-agency design rules for a legible demo (from the legacy directives/generate_n8n_workflow.md):

  • Start with a trigger (manual / webhook / schedule), matched to the gig (recurring job → schedule; inbound request → webhook; on-demand demo → manual).
  • 3–6 nodes total, one linear left-to-right chain — no sprawl, no parallel branches unless the gig truly needs an if/switch.
  • End with a visible outcome (send email, post to Slack, write a row, respond to webhook) so the demo has a payoff on screen.
  • Label every node in the client's language, not node-type jargon ("Find New Leads", not "HTTP Request"). This is what makes the screenshot sell.
  • Map each plan step from the cover letter to exactly one node so the workflow mirrors the pitch.

Step 5 — Validate per-node, then write the code

  • As you settle each node's config, call mcp__claude_ai_n8n_Cloud__validate_node_config (it can check several candidate configs in one call) so param/discriminator errors surface before they're buried in a full-graph run.
  • Write the workflow using the SDK patterns from the reference and the exact param names from Step 3. Lay nodes out left-to-right and wire them sequentially.

Step 6 — Validate the full workflow, then create + publish

  1. mcp__claude_ai_n8n_Cloud__validate_workflow — must pass clean. Fix and re-validate; do not create a workflow that fails validation.
  2. mcp__claude_ai_n8n_Cloud__create_workflow_from_code — create it (target ${BRAND}_N8N_PROJECT if set). Name it for the gig: "{Brand} — {client solution} demo".
  3. mcp__claude_ai_n8n_Cloud__publish_workflow — publish so it has a stable, shareable URL.
  4. mcp__claude_ai_n8n_Cloud__get_workflow_details — capture the workflow id and live URL for the gig row + the screenshot phase.

Do not execute the workflow against live services — it's a demo. If a node needs credentials to validate, leave a clearly-labeled placeholder credential; never wire the client's real accounts.

Step 7 — Write the workflow back onto the gig row

Ensure properties exist on ${BRAND}_GIGS_DB (idempotent — add only if missing):

Use mcp__claude_ai_Notion__notion-update-data-source (only for properties not already present):
- Workflow URL  : url
- Workflow ID   : rich_text
- Node Count    : number

Then update the page and append a ## Demo Workflow body block listing the node chain (client-language labels, in order):

Use mcp__claude_ai_Notion__notion-update-page:
- page_id: <gig page id>
- properties: {
    "Status":       "Workflow Built",   // adds the option if new
    "Workflow URL": "<live n8n url>",
    "Workflow ID":  "<id>",
    "Node Count":   <n>
  }

Step 8 — Save local backup + Slack the founder

Save a record to outputs/{brand}/sales/gigs/workflows/Workflow_{gig-slug}_{DDMonYYYY}.md (node chain, URL, validation result).

Before calling slack_send_message, you MUST first call ToolSearch with query "select:mcp__claude_ai_Slack__slack_send_message" to load the deferred tool schema.

Use mcp__claude_ai_Slack__slack_send_message:
- channel_id: "$SLACK_NOTIFY_USER"
- text: "🛠️ [{brand}] Demo workflow built — {gig title}
         {node count} nodes · {trigger} → … → {final visible step}
         Live workflow: {workflow_url}
         Next: /link-skills:vsl-demo-producer to capture the screenshot + recording script.
         Gig row: {notion_url}"

Output format

Save location: outputs/{brand}/sales/gigs/workflows/ Naming: Workflow_{gig-slug}_{DDMonYYYY}.md

Output metadata:

---
Date: YYYY-MM-DD
Skill Used: n8n-workflow-builder
Brand: {brand}
Gig: {gig title}
Category: {solution shape}
Workflow URL: {live n8n url}
Workflow ID: {id}
Node Count: {n}
Validation: passed
Notion Gig Row: {url}
Status: Workflow Built
---

Output sections:

  1. Node chain — table: order, client-language label, n8n node type, role in the demo
  2. Build notes — techniques pulled from best-practices, any node substitutions, placeholder credentials used
  3. Links — live workflow URL, gig row URL

Quality checklist

  • Active brand resolved; agents/link.md + product.md read first
  • Gig confirmed in ${BRAND}_GIGS_DB; Category present (or inferred + recorded)
  • SDK reference + relevant best-practices pulled before writing any workflow code
  • Node types fetched via get_node_types — no guessed parameter names
  • Workflow starts with a trigger, is a 3–6 node linear chain, ends with a visible outcome
  • Every node labeled in the client's language, one node per plan step from the bid
  • Nodes reflect tools the brand actually sells (product.md) — no fabricated capabilities
  • Each node config passed validate_node_config; full workflow passed validate_workflow
  • Workflow created and published; live URL + ID captured via get_workflow_details
  • No live client credentials wired — placeholders only (it's a demo, not a deployment)
  • Workflow URL + ID + node count written to the gig row; Status="Workflow Built"
  • Local backup saved; Slack digest sent to $SLACK_NOTIFY_USER
  • Agent run logged to dashboard

Final Step — Log to Dashboard

See docs/new_agent_onboarding/metrics-spec.md for the full JSONB contract.

Use gateway MCP tool `fiveagents_log_run`:
- fiveagents_api_key: ${FIVEAGENTS_API_KEY}
- skill: "n8n-workflow-builder"
- brand: "<active-brand>"
- status: "<success|failed>"
- summary: "<1 line, <200 chars>"
- started_at: "<ISO timestamp>"
- completed_at: "<ISO timestamp>"
- metrics: {
    "date": "YYYY-MM-DD",
    "gig_title": "<title>",
    "category": "<solution shape>",
    "node_count": 0,
    "trigger_type": "<manual|webhook|schedule>",
    "validation": "passed",
    "published": true,
    "workflow_url": "<live n8n url>",
    "workflow_id": "<id>",
    "status_after": "Workflow Built",
    "notion_gig_url": "<url>",
    "output_path": "outputs/{brand}/sales/gigs/workflows/",
    "deliverable": "Workflow_{gig-slug}_{DDMonYYYY}.md"
  }

Part of the pipeline

The Prove phase of the Inbound Gig Engine:

gig-prospector (Discover) → gig-proposal-writer (Write, Status="Drafted")
  → n8n-workflow-builder (this skill — real workflow + URL, Status="Workflow Built")
  → vsl-demo-producer (Demo — screenshot + recording instructions, Status="Demo Ready")

On-demand, one gig at a time. Uses the n8n Cloud MCP SDK flow exclusively.