| name | general-scams-article |
| description | Produce a single deeply-researched, audited, deployed article in /scams/everywhere/ from the existing Reddit research corpus. Implements the workflow we used for the pig-butchering page (PR #1187): research → plan → build → 8-pass audit → hub-update → deploy.
Trigger on: "/general-scams-article <slug>", "produce a general scams article", "write a /scams/everywhere page", or any request to author a new general (non-travel) scam page.
|
| type | project-skill |
| invocation | /general-scams-article <slug> |
General-Scams Article Producer
What this skill does
Produces a single article at /scams/everywhere/<slug>/index.html plus all
required supporting changes (hub update, sitemap regen, LLMs.txt entry,
internal cross-links, schema validation, deploy + merge). The output target
is a 10/10 stellar, well-researched, accurate, non-hallucinating article
with maximum SEO/AI-search visibility.
Origin
This skill encodes the workflow that produced PR #1187 (pig-butchering article)
and PR #1190 (nav link),
audited 3x for execution / quality / SEO and iterated until each dimension
passed 9+/10.
Inputs
A slug from the Coming-Soon list on /scams/everywhere/. Currently
supported (per corpus-mapping.json):
ai-voice-clone-scamsbank-impersonation-and-zelletech-support-scamspackage-text-scams (USPS / FedEx / Amazon)real-estate-wire-fraudfake-job-offersmarketplace-scams-fb-craigslistmedicare-and-elder-scamsrecovery-scamsgift-card-scamsconcert-and-event-ticket-scamssextortiondoor-to-door-contractor-scams
Add more by editing corpus-mapping.json and source-mapping.json.
Outputs (atomic, single PR)
/scams/everywhere/<slug>/index.html — flagship 6,000–8,500-word page/scams/everywhere/index.html — hub updated (slug moved Coming Soon → Live Guides)sitemap.xml — regeneratedllms.txt — entry added- Other
/scams/everywhere/<sibling>/index.html pages — Related Reading sections updated to cross-link to the new page
sources.md sidecar at tmp/scam-skill/<slug>/sources.md (gitignored audit trail)- Series-nav neighbors — when a new page joins the corpus, the prev/next siblings need their
.series-nav updated (see "Canonical series order" in page-anatomy.yaml)
Editorial + UI lessons from the 2026-04-29 corpus audit
Two follow-on audits (Tier 2 editorial + Tier 4 UI/UX) reshaped the skill.
Read these BEFORE drafting a new page:
voice-rules.md § "Banned cross-page formula tics" — the sentence-shapes
caught when 5 pages in the same template were reviewed in one sitting.voice-rules.md § "Variant paragraph-3 opener rotation" — defenses must
not all open with "The defense is..." Use the rotation set.voice-rules.md § "Hook ↔ Variant #1 decoupling" — the hook story and
V1's first scene must be different cases.page-anatomy.yaml § page_meta_row — the new severity replacement.
Don't ship "5 High Risk" pills; use 🎯 Target + 📈 Median loss pills.page-anatomy.yaml § book_mid_cta — required mid-page CTA placed
directly after the action_grid (highest-intent moment).page-anatomy.yaml § series_nav — required prev/next sibling links
above the legal disclaimer; circular order is canonical.page-anatomy.yaml § CSS / inline-style policy — do NOT add a per-page
<style> block; use the consolidated styles in /assets/scams.css
under body.editorial-v2.- Pass 8 hard-fails the formula tics now; run the helper before commit.
Hard checkpoints (the skill stops and waits for you here)
- After Stage 1 — research summary + variant cluster proposal
- After Stage 2 — final plan with audit scores
- After Stage 4 — finished page + 8-pass audit report
- Before merge — explicit user authorization to merge
How to invoke
/general-scams-article ai-voice-clone-scams
The skill will read this directory, walk through the 8 stages, and stop at
each checkpoint for your approval.
File map
| File | Purpose |
|---|
SKILL.md | This file (entrypoint) |
workflow.md | The 8 stages in detail with exact commands |
page-anatomy.yaml | The 15-section page template with slot definitions |
variant-template.yaml | Per-variant micro-template (7 elements) |
voice-rules.md | Voice + tone rules + AI-tic anti-patterns |
audit-checklists/pass-1-clarity.md | Pass 1 checks |
audit-checklists/pass-2-voice.md | Pass 2 checks |
audit-checklists/pass-3-design.md | Pass 3 checks |
audit-checklists/pass-4-authority.md | Pass 4 checks |
audit-checklists/pass-5-grammar.md | Pass 5 checks |
audit-checklists/pass-6-source-reread.md | Pass 6 checks (re-fetch sources) |
audit-checklists/pass-7-accessibility.md | Pass 7 WCAG AA checks |
audit-checklists/pass-8-ai-tics.md | Pass 8 anti-AI-pattern grep |
corpus-mapping.json | slug → corpus-keys (in tmp/scam_research/corpus.json) |
source-mapping.json | slug → required federal/NGO/industry/press sources |
helpers/extract_top_threads.py | Pull top-N threads from corpus, sorted by relevance |
helpers/update_hub.py | Move slug from Coming Soon → Live Guides on hub |
helpers/archive_urls.py | Submit cited URLs to archive.org for permanence |
helpers/verify_anti_tics.py | Pass 8 automated grep |
README.md | Quickstart for first-time use |
State persistence (resume)
Per-build state is written to tmp/scam-skill/<slug>/state.json after each
stage. Resuming is automatic if the state file exists.
Source-of-truth corpus
The Reddit research corpus is at tmp/scam_research/corpus.json (39 categories,
276 threads, 3,769 comments — produced by the original research run).
The skill reads from this corpus; it does not perform fresh Reddit searches
unless the user explicitly requests an update.
