seo-auditor

SEO Auditor

⚠️ PARKED — NOT WIRED (2026-08-12 wiring audit). No Stage() dispatches this agent; its only skill-layer mention is a ## See also in skills/phase-optimize/SKILL.md. Nothing produces seo-audit.json, and its only would-be reader is the equally-parked agents/publisher.md. Decision (operator, 2026-08-12): tombstone, not wire. Its job was absorbed piecemeal: run_quality_gates.py (the quality-gates BASH stage) runs the CORE-EEAT / CITE / AI-Slop scorers, and the deterministic lint stages (render-lint, keyword-density, visual-density, stat-grid, paa-alignment, locale-spelling, brand-fact) cover the rest with zero LLM cost. The scripts this file uniquely invoked — scripts/validate/word_count.py, scripts/lint/markdown_structure_check.py, scripts/lint/evidence_density_check.py — are therefore ORPHANED as pipeline checks; treat any claim that they run per-article as false until a Stage() names them. "Quality Gate 1 of 4" in the description below is the RETIRED numbering; the live gate set is quality-gates → geo-content-optimizer → independent-reviewer → pre-publish-gate.

The aggregator of deterministic SEO scoring. Runs every script in scripts/validate/ + scripts/lint/ that produces an SEO signal, combines into a unified score.

When invoked

• After humanizer complete (Stage: humanizer-complete)
• Before editor-in-chief review (Stage: optimize-complete)
• Triggered by L2 phase-optimize as Gate 1 of 4
• Manually via /seo-check memory/workspace/{task_id}/draft.md

Inputs

• memory/workspace/{task_id}/draft.md
• memory/workspace/{task_id}/meta.json
• memory/workspace/{task_id}/state.json
• references/seo/seo-checklist-2026.md

Tool whitelist

• Read — load draft + meta + references
• Bash — execute scoring scripts
• Write — output seo-audit.json

Forbidden: Edit, Task, WebFetch. Auditor inspects, does not modify or fetch.

Workflow

Step 1: Run all SEO scripts

# Word count + structure
python -m scripts.validate.word_count memory/workspace/{task}/draft.md --json
python -m scripts.lint.markdown_structure_check memory/workspace/{task}/draft.md --json

# Title validation (register-aware; validates the short seo_title vs the H1)
python -m scripts.validate.title_validator "{meta.seo_title}" --primary "{kw}" --register "{meta.register}" --h1 "{meta.title}" --json

# Keyword density
python -m scripts.lint.keyword_density --input draft.md --primary "{kw}" --json

# Citation Capsule lint (per-H2 self-contained block check)
python -m scripts.lint.citation_capsule_lint memory/workspace/{task}/draft.md --json

# Information gain — plain prose signals (bracket markers RETIRED 2026-07-14; O01/O02)
python -m scripts.validate.core_eeat_scorer memory/workspace/{task}/draft.md --project-slug {project} --json

# Evidence density (≥2 tables + ≥3 stats with FLOW Triple + ≥1 Tier-1 source)
python -m scripts.lint.evidence_density_check memory/workspace/{task}/draft.md \
    --format {format} --json
# (if YMYL: add --ymyl flag for stricter thresholds)

# Link resolution (all internal + external links 200 OK)
python -m scripts.validate.link_resolver memory/workspace/{task}/draft.md --json

# Schema validation (no deprecated types)
python -m scripts.validate.schema_validator memory/workspace/{task}/schema.json --json

# APA citation format
python -m scripts.validate.apa_format_validator memory/workspace/{task}/citations.json --json