pipeline

name	pipeline
description	Run the full content pipeline (write + humanize + SEO check) on one or more services, guides, or categories. Use when asked to "run the pipeline", "write and polish", "full content workflow", or "write + humanize + SEO". Runs steps sequentially per item, items in parallel.
argument-hint	'service <name>' or 'service all' or 'guide all' or 'category all' or 'category <slug1>, <slug2>'

Content Pipeline Skill

Run write → humanize → seo-check in sequence on one or more items. When processing multiple items, each item's pipeline runs as a parallel subagent.

What it does

For each item, the pipeline runs three steps in order:

1. Write — Generate content from research data (/write-service, /write-guide, or /write-category)
2. Humanize — Strip AI patterns from the written content (/humanize-service, /humanize-guide, or /humanize-category)
3. Seo-check — Audit and score the final content (/seo-check-service, /seo-check-guide, or /seo-check-category)

Each step depends on the previous one completing. But different items are independent and run in parallel.

Argument parsing

Argument	Meaning
service ProtonMail	Full pipeline for one service
service ProtonMail, Tutanota, Mailbox.org	Full pipeline for these services in parallel
service all	Full pipeline for all services with completed research
service unwritten	Pipeline for researched services that have no content yet
service category Email	Pipeline for all services in the Email category
guide Gmail to ProtonMail	Full pipeline for one guide
guide all	Full pipeline for all guides
category all	Full pipeline for all categories
category social-media, email	Full pipeline for these categories

Process

Step 1: Build the item list

Same query logic as the bulk skills:

For service all / service unwritten:

mcp__payload__findServices: {"where": "{\"researchStatus\": {\"equals\": \"complete\"}}", "limit": 100, "depth": 0}

For unwritten, filter for services where content is null/empty.

For specific names: Split comma-separated list, look up each.

For guide all:

mcp__payload__findGuides: {"limit": 100, "depth": 1}

Step 2: Confirm with user

Show:

• Number of items
• List of names
• What will happen: "Each item gets: write → humanize → seo-check (sequential). Items run in parallel."
• Estimated scope (3 MCP updates per item)

Ask for confirmation.

Step 3: Update pipeline status

For each service, update the contentPipelineStatus to "in-progress":

mcp__payload__updateServices: {"contentPipelineStatus": "in-progress"}

Step 4: Dispatch parallel pipeline agents

Launch one subagent per item. Each agent runs the three steps sequentially within itself.

Agent prompt template for services:

You are running the full content pipeline for service "{SERVICE_NAME}" on switch-to.eu.

Run these three steps IN ORDER. Each step must complete before the next starts.

## Step 1: Write (/write)

1. Fetch the service:
   mcp__payload__findServices: {"where": "{\"name\": {\"contains\": \"{SERVICE_NAME}\"}}", "limit": 1, "depth": 1}

2. Verify researchStatus is "complete". If not, return "SKIPPED: research not complete".

3. Write these fields using research data:
   - description: 1-2 sentences, under 200 chars
   - features: 4-6 short tags, 2-4 words each
   - content: 150-250 words Lexical richText JSON (opening → what's different → worth knowing h2)
   - pricingTiers: structured array from pricing research
   - For non-EU: issues array with factual privacy concerns

   Voice: knowledgeable friend. Short sentences. Active voice. No em dashes or semicolons.
   Consumer-friendly: "only you can read your emails" not "zero-access encryption".
   Always include "Worth knowing" section.

4. Save via mcp__payload__updateServices with _status: "draft".

## Step 2: Humanize (/humanize)

1. Re-fetch the service to get the just-written content.

2. Check all text fields for 18 AI patterns:
   Structural: rule of three, parallel openings, symmetric paragraphs, formulaic transitions, setup-punchline
   Word-level: inflated vocab, hedging, vague intensifiers, AI-favorite words, promotional adjectives
   Tone: excessive enthusiasm, false certainty, anthropomorphization, reader flattery, empty empathy
   Punctuation: em dashes (zero), semicolons (zero), ellipses

3. First pass: rewrite flagged text. Shorter sentences, specific details, varied length, natural contractions.

4. Second pass: audit for remaining AI feel. Fix.

5. Save changed fields via mcp__payload__updateServices.

## Step 3: SEO Check (/seo-check)

1. Re-fetch the service to get humanized content.

2. Run 10-point checklist (each: PASS 10, PARTIAL 5, FAIL 0):
   Meta Title, Meta Description, Keywords, Heading Structure, Content Length,
   Internal Context, Open Graph, Image, Readability, Freshness

3. Calculate score (max 100).

4. Save: seoScore, seoNotes, lastSeoReviewAt. Auto-populate empty metaTitle/metaDescription.

5. Update contentPipelineStatus to "complete".

## Return

Return a summary:
- Service name
- Write: word count, pricing tiers count
- Humanize: patterns found/fixed
- SEO: score and rating
- Any issues or gaps

Agent prompt template for guides:

You are running the full content pipeline for guide "{GUIDE_TITLE}" on switch-to.eu.

Run these three steps IN ORDER:

## Step 1: Write
Fetch both source and target services. Write: title, description, difficulty, timeRequired,
intro, beforeYouStart, steps (4-8), troubleshooting, outro, missingFeatures.
Save with _status: "draft".

## Step 2: Humanize
Re-fetch guide. Apply 18-pattern detection + two-pass rewrite on all text fields
(intro, beforeYouStart, step contents, troubleshooting, outro). Save changed fields.

## Step 3: SEO Check
Re-fetch guide. Run 10-point checklist. Save seoScore, seoNotes, lastSeoReviewAt.

Return: guide title, step count, patterns found/fixed, SEO score.

Step 5: Collect and report

## Content Pipeline Complete

Processed: X/Y items (write → humanize → seo-check)

| Item | Type | Words | AI Patterns Fixed | SEO Score | Status |
|------|------|-------|-------------------|-----------|--------|
| ProtonMail | Service | 220 | 6 | 85/100 | Complete |
| Tutanota | Service | 195 | 3 | 78/100 | Complete |
| ... | ... | ... | ... | ... | ... |

Average SEO score: XX/100
Total AI patterns fixed: XX

Skipped (no research): [list]
Failed: [list with reasons]

Next steps:
- Review all drafts in Payload admin (/admin)
- Run /bulk-translate to localize completed content
- Publish approved content

Guardrails

• Max parallel agents: 10 at a time. Each agent does 3 sequential steps, so they're heavier than single-skill agents.
• Research required: Services without completed research are skipped.
• Sequential within each agent: Write MUST complete before humanize. Humanize MUST complete before SEO check. This is enforced by the agent running steps in order.
• Draft status: Everything saves as draft. No auto-publishing.
• Idempotent: Running the pipeline twice on the same service overwrites the previous content. Confirm with user if content already exists.
• Pipeline status tracking: The contentPipelineStatus field on services tracks progress. Values: not-started → in-progress → complete.