Content Gen

Multi-format content generation with built-in neural quality scoring — every piece is scored against neuromarketing dimensions before delivery. Content that...

Security Scan

The skill's requirements and instructions align with its stated purpose: it uses the Anthropic API to generate and re-score content; nothing requested or instructed is disproportionate or unrelated.

Purpose & Capability

Name/description (content generation with neural scoring) match the declared primary credential (ANTHROPIC_API_KEY) and the install of the @anthropic-ai/sdk package. Requesting a node or python3 runtime is plausible for executing score/regenerate logic.

Instruction Scope

SKILL.md is an instruction-only spec describing generation, scoring, and auto-rewrite logic. It does not instruct the agent to read unrelated files, access other environment variables, or transmit data to unexpected endpoints beyond implied use of the Anthropic API.

Install Mechanism

Install spec pulls a single npm package (@anthropic-ai/sdk). This is proportionate to calling the Anthropic API, but note the skill has no code files bundled — the runtime will need to fetch and use that SDK when invoked.

Credentials

Only ANTHROPIC_API_KEY is required and identified as the primary credential. No unrelated secrets or config paths are requested.

Persistence & Privilege

The skill does not request always:true or any system-wide privileges. It is user-invocable and can run autonomously (platform default), which is appropriate for this integration.

Scan Findings in Context

[no_findings] expected: No code files present and the regex scanner found nothing to analyze. That is consistent with an instruction-only skill that relies on an external SDK at runtime.

Assessment

This skill appears coherent: it uses the Anthropic API to generate and re-score content and requests only the Anthropic API key. Before installing, confirm you are comfortable granting the skill access to your ANTHROPIC_API_KEY (it will be used to call the Anthropic service). Ensure you have a suitable runtime (node or python3) since the install lists the @anthropic-ai/sdk package. Understand that the 'neural_score' is an internal metric — review outputs before publishing (models can hallucinate or introduce factual errors), and verify any customer-facing 'score badge' claims are appropriate for your use case and compliance requirements.

Content Gen Skill — v1.2 (Quality-Gated)

Generate a full month of content in 2 hours, or get a single piece in 30 seconds — with a neural quality score on every output.

What Changed in v1.2

Previous versions generated content and stopped. v1.2 adds a content-scorer quality gate:

1. Content is generated
2. Output is scored against 8 neuromarketing dimensions (0-100)
3. If score is below your threshold (default: 70), the skill identifies the weakest dimensions and regenerates with specific improvement instructions
4. Final output includes the neural score as a deliverable proof point

This means you never receive content that hasn't passed a quality check — and your customers see the score.

---

What This Skill Does

Bulk or single-piece content for:

• 30-day social calendars — full month of posts across X, LinkedIn, Facebook, Instagram
• Video scripts — short-form (30s/60s) and long-form (3-5 min)
• SEO articles — keyword-targeted blog posts
• Listing copy — property descriptions from address or specs
• Agent partnership decks — talking points, one-pagers, pitch scripts
• Objection handling scripts — responses for common buyer/seller objections
• Market update newsletters — weekly/monthly digest for lead nurture lists

---

Input Contract

{
  "mode": "calendar|script|article|listing|partnership|objection|newsletter",
  "topic": "specific topic or paste raw notes",
  "platform": "x|linkedin|facebook|instagram|all",
  "length": "30 days | 500 words | 3 variations",
  "brand_voice": "urgent|educational|conversational|luxury|first-time-buyer",
  "score_threshold": 70,
  "auto_rewrite": true,
  "max_rewrites": 2,
  "output_score": true
}

---

Output Contract

All outputs now include a neural_score block:

{
  "content": "...",
  "neural_score": {
    "total": 79,
    "pass": true,
    "threshold_used": 70,
    "rewrites_needed": 1,
    "dimensions": {
      "second_person_density": 18,
      "sentence_complexity": 14,
      "concrete_noun_density": 12,
      "gain_vs_loss_framing": 11,
      "urgency_signals": 8,
      "identity_alignment": 7,
      "social_proof": 5,
      "specificity": 4
    },
    "improvement_applied": "Increased second-person framing and added concrete dollar amounts. Specificity improved by adding timeline (90 days) and quantified outcome."
  },
  "score_badge": "This content scored 79/100 on neural signal strength — above the 60/100 average for mortgage marketing content.",
  "structured_output": { "...mode-specific JSON..." }
}

---

The Quality Gate — How It Works

Neural Scoring Dimensions (8 factors, total 100pts)

Dimension	Max Points	What It Measures
Second-person density	20	"you/your" frequency — speaks directly to reader
Sentence complexity	18	Readability — shorter sentences score higher
Concrete noun density	16	Real numbers, named things, specific amounts
Gain vs loss framing	14	Positive outcomes weighted vs fear framing
Urgency signals	12	Time-bound language, scarcity, action pressure
Identity alignment	10	ICP match — does this speak to the right person
Social proof	6	Specificity of evidence (numbers, outcomes, names)
Specificity	4	Vagueness penalty — "great results" scores 0

Auto-Rewrite Logic

Generate content
→ Score against 8 dimensions
→ If total ≥ threshold: return with score
→ If total < threshold:
     Identify bottom 3 dimensions
     Generate improvement instructions
     Regenerate with explicit fixes
     Re-score
     If still < threshold and max_rewrites > 1: repeat
     Return best version with score + note on what improved

Default threshold: 70/100 (above industry average of 60) Recommended threshold for paid products: 75+ Max rewrites: 2 (prevents infinite loops)

---

Modes

Calendar:

{
  "week": 1,
  "posts": [
    {
      "day": "Mon", "platform": "LinkedIn",
      "hook": "...", "body": "...", "cta": "...",
      "char_count": 284, "hashtags": ["#RealEstate"],
      "neural_score": 77
    }
  ],
  "calendar_avg_score": 74,
  "low_score_posts": ["Day 8 (62) — rewrote: +9pts"]
}

Script:

[0:00-0:04] HOOK — Neural score: 81
"If you've served and you're renting right now — pay attention."
Score drivers: identity_alignment (9/10), urgency (8/10)

[0:04-0:12] PROBLEM
...
[Script score summary: 78/100]

Article: Full markdown with meta description, keyword density note, and per-section score.

All other modes: Markdown with score badge appended.

---

Score Badge (Use as Social Proof)

Every output includes a ready-to-use score badge. Add it to your Gumroad product page, Fiverr delivery, or social post:

"This content scored 79/100 on neural signal strength — above the 60/100 industry average for mortgage marketing content."

Buyers don't need to understand the methodology. They see proof it's been tested.

---

Example — Quality Gate in Action

Input: "Write a LinkedIn post about VA loans for veterans who are renting"

Draft 1 score: 61 (below 70 threshold)

• Weak: second-person density (8/20) — uses "veterans" not "you"
• Weak: specificity (1/4) — no dollar amounts or timelines
• Auto-improvement: switch to direct second-person, add "$0 down" and "no PMI" specifics

Draft 2 score: 78 (above threshold, delivered)

• "If you served and you're renting right now — you may be leaving a real advantage on the table. VA loans let eligible veterans buy with zero down and no PMI — saving $150–$300/month compared to a conventional loan on the same home."

Output includes: final content + 78/100 score + "1 rewrite needed" note

---

Backend

Uses generate.py (shared backend). Local MLX first, Haiku fallback.

LLM_BACKEND=local openclaw run content-gen "30-day calendar"   # free
LLM_BACKEND=haiku openclaw run content-gen "SEO article"       # ~$0.30

Note: Scoring runs locally (no additional API cost). Only generation uses LLM.

---

Tiers

Free — Single-piece generation only, no auto-rewrite, score reported but no gate Standard ($12/mo) — All modes, quality gate active, up to 1 rewrite, score badge included Pro ($37/mo) — All modes, up to 2 rewrites, threshold configurable, bulk calendar with per-post scores, structured JSON export

---

Integration

# Via Telegram
@openclaw content-gen calendar "30 days, mortgage, threshold 75"

# Via Claude Code
openclaw run content-gen "video script, 60s, VA loan, score gate 75"

# Direct
openclaw run content-gen --mode=article --topic="first-time buyer myths" --threshold=72

---

v1.2.0 — Added content-scorer quality gate. Every output now scored against 8 neuromarketing dimensions. Auto-rewrite if below threshold. Score badge included in all deliverables. Upgraded from depth 2 (template generator) to depth 3 (quality-validated generator).