⛔ This skill MUST be read in full — not skimmed. Formal review gates depend on its workflow.
Skipping steps silently bypasses quality checks. Missing gates = undetected breakages.

Content Verification — Content-Domain Gate

Overview

Ensures every Content-domain issue reaches an actual verification skill. Thin orchestrator — dispatches to the existing content pipeline v2 reviewers and aggregates their output into a single gate verdict.

Announce at start: "I'm using the content-verification skill to run content verification."

Invoked by test-routing when domain=content is detected. Also the Content-domain verification trigger referenced by research-verification (content-pipeline research deferral) and issue-creation (Content domain row in the complexity table).

When to Use

• Editorial, guide, deal, or refresh content changes (issue has content/editorial/deal labels)
• Content written by any content writer skill that needs the post-writing review cycle
• Any routing that lands on the Content domain

When NOT to Use

• Pure code changes with no content → routed to code verification
• Research investigations → research-verification (unless content-pipeline keyword/SERP research → this skill)
• Trivial copy/label changes with no editorial substance → skip

Inputs Required

Input	Source	Notes
Draft content	Writer skill output	intro, body, FAQs, meta fields (page_type-specific)
Research brief	Content research (fact-checked)	Verified claims, ✓/~/✗ tags (breadth/depth source; deal mode uses the Deal Research Brief with confidence tags)
Deal Research Brief	deal-content-writer Step 0.5 (deal mode only)	specialty_facts, practical_tips, confidence tags — source for breadth B-D1–B-D4 / depth D-D1–D-D3
deal_id	Orchestrator (deal mode only)	resolves the deal DB record — used ONLY by fact-checker D1–D5
page_type	Orchestrator	editorial | guide | deal | refresh
Language	Orchestrator	es/en — determines audience consistency checks

Process

Step 1 — Read Domain & Inputs

Confirm the routing is Content-domain from the test-routing verification plan:

Domain: content
page_type: editorial | guide | deal | refresh

Gather the draft content, research brief, page_type, and language. If page_type = deal, run deal-mode checks: breadth B-D1–B-D4 and depth D-D1–D-D3 read the Deal Research Brief (specialty_facts, practical_tips, confidence tags) — they do NOT need a DB audit; only fact-checker D1–D5 reads the deal DB record (value_mechanism, access_conditions, business.name, category).

Step 2 — Fan-Out Breadth + Depth Reviewers (Parallel)

Dispatch the two post-writing reviewers in parallel (per content pipeline v2 — they run concurrently):

1. content-reviewer-breadth — coverage: neighborhoods, cuisine/experience types, price tiers, partner completeness (BR1–BR7 / B-D1–B-D4)
2. content-reviewer-depth — local depth, unique findings, audience consistency (DPT1–DPT5 / D-D1–D-D3)

Pass to each: draft content, research brief, page_type, language. Collect their structured ISSUE blocks (P0/P1/P2 with location/problem/fix) and summary counts.

Step 3 — Review Cycle Until Clean

The reviewers are gates, not one-shot checks:

1. Merge issues from both reviewers, dedupe by location+problem
2. If issues exist → feed the merged list back to the writer for targeted fixes (fix ONLY flagged claims, no full rewrites)
3. Re-dispatch reviewers on revised content
4. Repeat until both return 0 issues or max iterations reached (cap 10 refinement rounds)

After max iterations: P0 remaining → gate stays closed, surface for human review. P1 → address before delivery. P2 → advisory, proceed with note. Never silently drop content.

Step 4 — Dispatch Fact-Checker (Writing)

Once the reviewer cycle is clean (is_clean: true), dispatch content-fact-checker-writing:

• Editorial/guide/refresh: extract claims → map to verified research brief (FC1+ / FCE1+ for guide deal-embed UUIDs) → P0/P1/P2 issue list
• Deal: run D1–D5 against the deal DB record (discount vs value_mechanism, business name, access framing, invented details, meta accuracy)

Fact-check is itself a gate — run its refinement loop until 0 issues or the 10-round cap. P0 after cap blocks delivery.

Step 5 — Return Verification Report

## Content Verification Report

**Issue:** #N
**page_type:** editorial
**Language:** es
**Reviewers dispatched:** breadth, depth (parallel), fact-checker-writing

### Issues
| # | Reviewer | Severity | Location | Problem | Fix |
|---|----------|----------|----------|---------|-----|
| 1 | breadth | P1 | body §3 | Missing neighborhood: Playa Norte | Add one-sentence mention... |

**Summary:**
- P0 issues: X (block delivery)
- P1 issues: Y (should fix before delivery)
- P2 issues: Z (optional)
- Unique research findings used: X/Y

**Verdict:**
- ✓ PASS — all gates clean, content verified
- ✗ NEEDS FIXES — [count] issues require correction

Pipeline Handoff

Invoked by: test-routing (domain=content), research-verification (content-pipeline research deferral), issue-creation (Content verification trigger)
Dispatches to: content-reviewer-breadth, content-reviewer-depth (parallel), content-fact-checker-writing
Consumed by: verification-gate checks in test-routing (Content domain dispatch) and research-verification / issue-creation references.

What Fails If You Skip

Skip	Consequence
Breadth review	Drafts publish with dropped neighborhoods, cuisines, or partners — incomplete pages
Depth review	Generic travel-blog content — no local insight, no audience consistency
Fact-check gate	Invented facts, prices, or superlatives reach production content
Deal-mode checks	Wrong discount %, wrong access framing, or invented deal details go live
Parallel fan-out	Serial dispatch doubles review wall-clock time for no quality gain

---

Continue following the workflow as mandated by this skill. Do not skip steps.