Content Scorer Skill
Score any piece of marketing copy in seconds. Get a 0-100 resonance score, dimension-by-dimension breakdown, and specific rewrite suggestions — before you post, send, or publish.
Free vs Premium
Free tier (no API key needed):
- -
--demo — run a full score on built-in demo copy, zero external calls, see exactly what the output looks like - INLINECODE1 — fast forbidden word scan, runs locally, no API
- Score up to 3 pieces of copy/day using local MLX (if you have it running)
Premium tier (ANTHROPICAPIKEY):
- - Unlimited scoring via Claude Haiku (~$0.001 per score)
- INLINECODE2 — get improved copy alongside your score
- INLINECODE3 — A/B test multiple hooks side-by-side
- INLINECODE4 — pipe scores into your agent workflows
- Batch scoring for content calendars
The free compliance check alone is worth installing — catch forbidden words before they go live.
What this skill does
Analyzes marketing copy across 6 weighted dimensions and returns:
- - Content Resonance Score (0-100) — composite score calibrated against fMRI brain-response patterns (TRIBE v2 weight calibration)
- Per-dimension scores — hook strength, specificity, emotional resonance, NLP technique usage, CTA strength, compliance
- Rewrite suggestions — specific line-level changes to improve the weakest dimensions
- Platform fit check — flag copy that's too long/short for the target platform
- Compliance gate — detect forbidden words before they go live
Scoring dimensions
| Dimension | Weight | What it measures |
|---|
| Hook Strength | 25% | First line/sentence — does it grab attention in <3 seconds? |
| Emotional Resonance |
25% | Does it connect to the reader's real situation, fear, or desire? |
| NLP Technique Usage | 20% | Presuppositions, embedded commands, pacing/leading, reframes, future pacing |
| Specificity | 15% | Concrete numbers, outcomes, timeframes — no vague platitudes |
| CTA Strength | 10% | Clear, urgent next step with no exit ramp |
| Compliance | 5% | No forbidden words, MLO-safe language |
Why these weights: TRIBE v2 fMRI analysis found hook + emotional resonance drive 50% of cortical engagement in language and reward circuits. NLP technique presence activates anterior insula (urgency) and mPFC (social motivation). Specificity activates hippocampal encoding — specific claims are better remembered.
Input contract
Tell me:
- 1. The copy to score — paste it directly
- Platform (optional): email / linkedin / x / facebook / instagram / sms / ad / script / any
- Audience (optional): first-time buyers / investors / realtors / general
- Rewrite mode (optional):
--rewrite to get revised copy alongside the score
Example prompts:
- - "Score this LinkedIn post: [paste copy]"
- "Score for email, real estate investors: [paste copy]"
- "Score and rewrite: [paste copy] --rewrite"
- "Compliance check only: [paste copy]"
- "Score these 3 hooks and tell me which is strongest: [hook A] / [hook B] / [hook C]"
Output contract
Standard score output:
CODEBLOCK0
Rewrite output (with --rewrite):
CODEBLOCK1
Multi-hook comparison:
CODEBLOCK2
How the skill works
Uses score_content.py (in this directory). Local MLX first (LLM_BACKEND=local), Haiku fallback.
CODEBLOCK3
Core scoring implementation:
CODEBLOCK4
Calibration note — TRIBE v2
Dimension weights are calibrated against TRIBE v2 (Meta's fMRI brain-response prediction model, facebook/tribev2). Emma sales call transcripts were run through TRIBE to measure predicted neural activation in language (STG/IFG), reward (mPFC/precuneus), and urgency (ACC/anterior insula) circuits.
Calibration findings:
- - Hook + emotional resonance → 50% of language/reward activation (hence 25% + 25% weights)
- NLP techniques → anterior insula / urgency circuit activation (20% weight)
- Specificity → hippocampal encoding — concrete claims stick (15% weight)
- CTA framing → frontal-pole decisional activation (10% weight)
To recalibrate weights with fresh TRIBE data: see vault/learnings/2026-03-27-tribe-v2-colab-spec-task47.md.
Use cases by role
Sales copy (pre-send):
"Score this email sequence — I'm targeting homebuyers who browsed last week"
Social content (pre-post):
"Score this LinkedIn post and tell me if the hook is strong enough"
Hook A/B testing:
"Which of these 3 hooks will perform better and why?"
Compliance pre-check:
"Check this for forbidden words before I post it"
Training data QA:
"Score Turn 3 of this Emma call transcript for NLP technique usage"
Integration with agent infrastructure
CODEBLOCK5
Benchmark scores (reference)
| Copy type | Typical range | Notes |
|---|
| Generic real estate post | 40-55 | Vague, no hook, weak CTA |
| Good LinkedIn post |
60-75 | Decent hook, some specificity |
| Emma Turn 3 (post-R15) | 72-85 | Strong NLP, assume-the-close CTAs |
| Direct response ad (top 5%) | 85-92 | Hormozi-style, concrete, urgent |
| Perfect score territory | 93-100 | Rarely seen — Claude Sonnet 4.6 + expert copy review |
内容评分技能
在几秒钟内对任何营销文案进行评分。在发布、发送或出版之前，获得0-100的共鸣评分、逐维度细分以及具体的改写建议。
免费版与高级版
免费版（无需API密钥）：
- - --demo — 对内置演示文案运行完整评分，零外部调用，精确查看输出效果
- --compliance-only — 快速违禁词扫描，本地运行，无需API
- 使用本地MLX（如果已运行）每天最多评分3条文案
高级版（ANTHROPICAPIKEY）：
- - 通过Claude Haiku进行无限评分（每次评分约$0.001）
- --rewrite — 在评分的同时获取改进后的文案
- --compare — 并排A/B测试多个钩子
- --format=json — 将评分结果接入你的智能体工作流
- 内容日历的批量评分
仅免费合规检查一项就值得安装——在违禁词上线前将其捕获。
该技能的功能
从6个加权维度分析营销文案，并返回：
- - 内容共鸣评分（0-100） — 根据fMRI脑反应模式校准的综合评分（TRIBE v2权重校准）
- 各维度评分 — 钩子强度、具体性、情感共鸣、NLP技巧使用、CTA强度、合规性
- 改写建议 — 针对最弱维度的具体行级修改
- 平台适配检查 — 标记对于目标平台过长/过短的文案
- 合规门控 — 在违禁词上线前进行检测
评分维度
| 维度 | 权重 | 衡量内容 |
|---|
| 钩子强度 | 25% | 第一行/句——是否能在3秒内抓住注意力？ |
| 情感共鸣 |
25% | 是否与读者的真实处境、恐惧或欲望产生连接？ |
| NLP技巧使用 | 20% | 预设、嵌入式指令、节奏引导、重构、未来导向 |
| 具体性 | 15% | 具体数字、结果、时间框架——无模糊空话 |
| CTA强度 | 10% | 清晰、紧迫的下一步行动，无退出通道 |
| 合规性 | 5% | 无违禁词，MLO安全用语 |
为何采用这些权重： TRIBE v2 fMRI分析发现，钩子+情感共鸣驱动了语言和奖励回路中50%的皮层参与。NLP技巧的存在激活前岛叶（紧迫感）和mPFC（社会动机）。具体性激活海马体编码——具体的主张更容易被记住。
输入约定
告诉我：
- 1. 要评分的文案 — 直接粘贴
- 平台（可选）：电子邮件 / LinkedIn / X / Facebook / Instagram / 短信 / 广告 / 脚本 / 任意
- 受众（可选）：首次购买者 / 投资者 / 房地产经纪人 / 普通大众
- 改写模式（可选）：--rewrite 在评分的同时获取修改后的文案
示例提示：
- - 评分这条LinkedIn帖子：[粘贴文案]
- 为电子邮件评分，受众为房地产投资者：[粘贴文案]
- 评分并改写：[粘贴文案] --rewrite
- 仅合规检查：[粘贴文案]
- 评分这3个钩子并告诉我哪个最强：[钩子A] / [钩子B] / [钩子C]
输出约定
标准评分输出：
内容共鸣评分：74/100
维度细分：
钩子强度： 8/10 ✓ 强模式中断
情感共鸣： 7/10 ✓ 连接所有权渴望
NLP技巧： 6/10 → 存在节奏引导，无嵌入式指令
具体性： 8/10 ✓ 具体价格+时间线
CTA强度： 5/10 ⚠ 退出通道：如果你感兴趣
合规性： 10/10 ✓ 干净
最弱点：CTA退出通道——如果你感兴趣给了读者一条出路。
首要建议：将如果你感兴趣，私信我替换为在下方输入你的邮编——我会为你调取数据。
检测到的NLP：节奏引导（你所在区域的大多数买家现在...），未来导向（想象你自己...）
缺失：嵌入式指令——在陈述句中添加一个祈使句：...这就是为什么精明的买家现在正在锁定。
改写输出（使用--rewrite）：
[上述评分模块]
--- 改写 ---
[修改后的文案，修改处已高亮]
--- 改写结束 ---
所做的修改：
- 1. 钩子 → 更强的模式中断（删除了我将要分享...）
- CTA → 假设成交（在下方输入你的邮编替代如果你感兴趣）
- 在正文中添加了嵌入式指令（...精明的买家本周正在锁定）
多钩子对比：
钩子A：6/10 — 通用开场白，无模式中断
钩子B：9/10 — 强好奇心缺口+具体性（大多数买家不知道这让他们每月多花$340）
钩子C：7/10 — 有情感但模糊，缺乏具体性
胜出者：钩子B。将好奇心缺口与具体的损失框架相结合。
该技能的工作原理
使用scorecontent.py（位于此目录）。优先使用本地MLX（LLMBACKEND=local），Haiku作为后备。
bash
评分一条文案
python3 score_content.py 你的LinkedIn帖子文本 --platform=linkedin
评分+改写
python3 score_content.py 你的文案 --platform=email --rewrite
对比钩子
python3 score_content.py --compare 钩子A文本 钩子B文本 钩子C文本
仅合规检查（快速，无需API调用）
python3 score_content.py 你的文案 --compliance-only
JSON输出（用于智能体管道）
python3 score_content.py 你的文案 --format=json | jq .score
强制后端
LLM
BACKEND=local python3 scorecontent.py 文案 # Qwen3.5-9B（免费）
LLM
BACKEND=haiku python3 scorecontent.py 文案 # Claude Haiku（每次评分约$0.001）
核心评分实现：
python
SCORING_PROMPT = 你是一名直复营销文案分析师，接受过以下方面的训练：
- - Hormozi（价值堆叠、紧迫感、无需思考的报价）
- Belfort直线说服（语调、确定性、信任）
- Cardone 10X（大胆、假设性语言、承诺）
- NLP说服（预设、嵌入式指令、节奏引导、重构、未来导向）
对以下{platform}文案的每个维度进行0-10分评分。
要严格——10分意味着你见过的最好的直复营销文案。
要评分的文案：
{copy}
受众：{audience}
仅以以下JSON格式回复：
{{
hook_strength: {{ score: N, reason: ..., improvement: ... }},
emotional_resonance: {{ score: N, reason: ..., improvement: ... }},
nlp_technique: {{ score: N, detected: [technique1, ...], missing: ..., improvement: ... }},
specificity: {{ score: N, reason: ..., improvement: ... }},
cta_strength: {{ score: N, reason: ..., improvement: ... }},
compliance: {{ score: N, violations: [] }},
overall_comment: ...
}}
WEIGHTS = {
hook_strength: 0.25,
emotional_resonance: 0.25,
nlp_technique: 0.20,
specificity: 0.15,
cta_strength: 0.10,
compliance: 0.05,
}
FORBIDDEN_WORDS = [
pre-approval, pre-approved, pre-qualify, specialist,
mortgage, lending, rates, loan, showings, tours,
transfer, connect, team, agent, department,
qualify for, AWESOME
]
def compliance_check(copy: str) -> list[str]:
快速本地检查——无需API调用。
violations = []
copy_lower = copy.lower()
for word in FORBIDDEN_WORDS:
if word.lower() in copy_lower:
violations.append(word)
return violations
def composite_score(dimensions: dict) -> int:
total = sum(dimensions[k][score] * WEIGHTS[k] for k in WEIGHTS)
return round(total * 10) # 0-100
async def score(c
