bulk-humanize

name	bulk-humanize
description	Strip AI writing patterns from multiple services, guides, or categories in parallel. Use when asked to "humanize all", "bulk humanize", "de-AI all content", "humanize all categories", or clean up writing across many items at once.
argument-hint	'service all' or 'guide all' or 'category all' or 'service <name1>, <name2>' or 'category <slug1>, <slug2>'

Bulk Humanize Skill

Humanize content for multiple services, guides, or categories in parallel by dispatching one subagent per item. Routes to the appropriate per-type skill: /humanize-service, /humanize-guide, or /humanize-category.

Argument parsing

Argument	Meaning
service ProtonMail, Tutanota	Humanize these specific services
service all	Humanize all services that have content
service category Email	Humanize all services in the Email category
guide all	Humanize all guides that have content
guide gmail-to-protonmail, outlook-to-tutanota	Humanize specific guides by slug
category all	Humanize all categories that have content
category social-media, email	Humanize these specific categories

Process

Step 1: Build the item list

For service all:

mcp__payload__findServices: {"limit": 100, "depth": 0}

Filter client-side for services where content is not null/empty.

For guide all:

mcp__payload__findGuides: {"limit": 100, "depth": 0}

Filter for guides where intro is not null/empty.

Step 2: Confirm with user

Show the list and count. Ask for confirmation.

Step 3: Dispatch parallel humanize agents

Agent prompt template for services:

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

Use the /humanize skill. Process:

1. Fetch the service:
   mcp__payload__findServices: {"where": "{\"name\": {\"contains\": \"{SERVICE_NAME}\"}}", "limit": 1}
   Get: description, content, features, issues

2. Run the AI Pattern Detection on all text fields. Check for these 18 patterns:

   Structural: rule of three lists, parallel sentence openings, symmetric paragraphs, formulaic transitions, setup-punchline
   Word-level: inflated vocab (utilize→use, leverage→use, comprehensive→[specifics]), hedging clusters, vague intensifiers, AI-favorite words (delve, landscape, paradigm, ecosystem, seamless, cutting-edge), promotional adjectives
   Tone: excessive enthusiasm, false certainty, anthropomorphization, reader flattery, empty empathy
   Punctuation: em dashes (zero allowed), semicolons (zero allowed), ellipses for drama

3. First pass: Rewrite all flagged text. Shorter sentences. Specific details. Varied sentence length. Natural contractions.

4. Second pass: Audit the rewrite. Fix remaining AI-sounding patterns. Add one short paragraph if all paragraphs are same length. Vary sentence structure.

5. Save changed fields via mcp__payload__updateServices.

Return: service name, patterns found count by category, fields updated.

Agent prompt template for guides:

You are humanizing the guide "{GUIDE_SLUG}" on switch-to.eu.

Use the /humanize skill. Same 18-pattern detection and two-pass rewrite process.

1. Fetch via mcp__payload__findGuides: {"where": "{\"slug\": {\"equals\": \"{GUIDE_SLUG}\"}}", "limit": 1}
   Get: intro, beforeYouStart, steps (each step's content), troubleshooting, outro

2. Apply pattern detection and two-pass rewrite to every text field.

3. Save changed fields via mcp__payload__updateGuides.

Return: guide slug, patterns found count by category, fields updated.

Step 4: Collect and report

## Bulk Humanize Complete

Humanized: X/Y items

| Item | Patterns Found | Patterns Fixed | Remaining |
|------|---------------|----------------|-----------|
| ProtonMail | 8 | 8 | 0 |
| Tutanota | 5 | 4 | 1 (review manually) |
| ... | ... | ... | ... |

Most common patterns across all items:
1. [pattern] — found X times
2. [pattern] — found X times

Guardrails

• Max parallel agents: 10 at a time.
• Content required: Skip items without content. Report them.
• Only update changed fields. Don't save a field if it didn't change.