topic-angle-selector
Within the chosen format, generates 6 title candidates across ≥3 angles, validates each via title_validator.py, scores CTR potential, picks top-1. Required step before outline-architect. Use whenever planning to write — happens after format-selector and before outline.
XuanRanL · 已有静态检查记录
安装与使用
运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。
npx skills add XuanRanL/loamwright-SEO-Skill --skill topic-angle-selector技能说明
name: topic-angle-selector description: Within the chosen format, generates 6 title candidates across ≥3 angles, validates each via title_validator.py, scores CTR potential, picks top-1. Required step before outline-architect. Use whenever planning to write — happens after format-selector and before outline. allowed-tools: [Read, Write, Bash]
Topic Angle Selector
Given a chosen format_id, generate 6 strong title candidates, validate them, pick the winner.
Inputs
state.brief.primary_keyword+secondary_keywordsangle.json.format_id(from format-selector)research.json(top competitor titles, content gaps)references/seo/power-words.mdreferences/seo/angle-catalog.md(12 angles)references/seo/micro-copy-tactics.mdprojects/{slug}/brand-config.json(target_locale, banned_competitors)projects/{slug}/business-context.json :: voice_default→ register (see below)
Two title fields (2026-06-16) — generate BOTH
Stop conflating one string. Produce two aligned fields (full rationale:
docs/title-optimization-plan-2026-06-16.md):
| Field | Job | Constraint |
|---|---|---|
seo_title | The indexed <title> / rank_math_title / og:title. Survives Google's rewrite (Google rewrites ~76% of titles; >65 chars → ~99.9%). | 51–60 chars (hard-fail >65 / <30), primary keyword once + front-loaded, sentence case, no bracketed/parenthesized year, register-compliant. |
h1 | The on-page H1 / WordPress post title. Carries the full human thesis + nuance + the year if wanted. | Longer OK (≤ ~90 chars). MUST share the entity + primary keyword + every number that is in seo_title (Google preserves a number in the displayed title 97.3% of the time only when it is in BOTH). |
The old single title field still maps to h1 for back-compat (publisher routes
meta.title → WP post title, meta.seo_title → rank_math_title).
Register (pick before generating)
Read business-context.json :: voice_default and resolve a register — or call
scripts.validate.title_validator.infer_register(voice_default):
| Register | Projects | Digit | Power words | Hard-banned |
|---|---|---|---|---|
b2b_technical | project-charlie, project-juliet | real metric/count, in H1 too | discouraged (warn) | hype: amazing/revolutionary/game-changer |
b2b_procurement | project-kilo | economics numbers | discouraged | hype |
dtc_celebration | project-hotel (living-pet/gift) | optional, soft | warmth ok | hype, urgency |
dtc_grief | project-hotel (pet-loss/memorial) | default none; soft if used | BANNED | hype, urgency, best/top/proven/ranked/guaranteed/#1, exclamation |
ecommerce | project-echo (after /init) | product specs | optional | — |
default | unknown | optional | optional | — |
project-hotel spans two registers — pick dtc_grief for pet-loss/memorial/cremation
content and dtc_celebration for living-pet/gift content, per the article's intent. Do
NOT use one project-wide register for it.
Output
Updates workspace/{task_id}/angle.json with:
seo_title(winning, validated short<title>)h1(aligned long display title — also written totitlefor back-compat)register(the resolved register key)slug_draftangle(one of 12)hookpromisepower_word(optional now — empty string is valid)digit(optional now — empty string is valid)personaalternative_titles_considered[](5 backups for repair Round 5; each an{seo_title, h1}pair)
Workflow
Stage 1: Generate 6 candidates (LLM)
Each candidate is an {seo_title, h1} pair. Constraints (changed 2026-06-16 — evidence in
docs/title-optimization-plan-2026-06-16.md):
- 6 candidates spanning ≥3 different angles (from
references/seo/angle-catalog.md) - Each
seo_titleis 51–60 chars (hard ceiling 65), primary_keyword once, front-loaded - Each
h1carries the full thesis; shares the entity + keyword + every number in itsseo_title - A power word is OPTIONAL, not required — and is BANNED in the
dtc_griefregister. Lead with specificity (a defining metric, a contrast, the article's actual finding), not a power word. - A digit is OPTIONAL, not required — include one only when it is a real list count or metric
AND it also appears in the
h1. For informational / myth-buster / grief intents, omit it. - Each in sentence case (NOT Title Case)
- No bracketed/parenthesized year in
seo_title((2026)/[2026]are rewrite-bait — put the year in theh1instead) - Respect the register's hard-banned terms (see table above)
- Vary openings: entity-first / contrast-first / verb-first / question-first (only for true question intent — questions give NO SERP-CTR premium per Backlinko 2025)
Suggested distribution per 6 candidates:
- 2 from intent-matching angle
- 2 from adjacent angles
- 2 wild-card (myths / case-study / trends)
Stage 2: Validate each candidate
Validate the seo_title, passing the register and the paired h1 so alignment + number
preservation are checked:
python -m scripts.validate.title_validator "{seo_title}" \
--primary "{primary_keyword}" --register "{register}" --h1 "{h1}" --json
Filter to those with all_passed: true (no hard issues). The warnings[] array is advisory —
prefer candidates with the fewest warnings but do not reject on warnings alone. If fewer than 3
pass, regenerate. Common hard failures now: seo_title >65 chars, a number in seo_title absent
from h1, a register-banned term. (Title Case and bracketed/parenthesized years are advisory
warnings, NOT hard fails — though sentence case + year-in-h1 remain the house preference.)
Stage 3: LLM CTR scoring
For surviving candidates, ask Claude Opus:
- Score 0-100 on: clarity, urgency, specificity, hook strength, primary keyword placement
- Rank top-3
Stage 3.5: SERP-clone rejection (anti-homogenization)
After CTR scoring, compare each surviving cand
阅读完整来源 ↗