article-reader-qa
/article-reader-qa
Sprawdza czytelność artykułu, naturalny rytm i brak mechanicznego tonu, zanim tekst zostanie wzbogacony.
- Autor
- AssassinFlash ↗
- Repozytorium
- b2b-seo-agent-engine ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 02.11.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 1
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 02.11.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co robi i kiedy
Sprawdza czytelność artykułu, naturalny rytm i brak mechanicznego tonu, zanim tekst zostanie wzbogacony.
Argumenty
/article-reader-qa [content-workspace/project-slug] [article-slug]
Oryginalny opis (EN)
Use when: the draft exists and you need to audit human readability, natural flow, rhythm, and non-mechanical tone before enrichment.
Jak działa
- Argument Contract
- Purpose
- Instructions
- Output
- Examples
- Guardrails
- PAUSE
Podgląd SKILL.md
--- description: "Use when: the draft exists and you need to audit human readability, natural flow, rhythm, and non-mechanical tone before enrichment." argument-hint: '[content-workspace/project-slug] [article-slug]' user-invocable: true allowed-tools: - Bash - Read - Edit - Write - WebSearch - WebFetch - Agent - mcp__you__* - mcp__tavily__* - mcp__exa__* - mcp__crawl4ai__* - mcp__firecrawl__* - mcp__plugin_playwright_playwright__* - mcp__apify__* - mcp__agent-reach__* - mcp__douyin__* - mcp__linkedin__* - mcp__mcp-server-chart__* name: article-reader-qa --- # /article-reader-qa ## Argument Contract Use `$ARGUMENTS` as `content-workspace/<project-slug> <article-slug>`. Before QA, verify: `project-profile.md`, `state.json` indicates `lockedProfile: true`, `draft.md`, `truth-qa.md`, `information-gain-qa.md`, `benchmark-drift-audit.md` all exist. ## Purpose Gate 4 of 4. The final human-quality check. Verify the article reads like one continuous human-authored piece, not stitched-together modules. Catch the most dangerous failure mode: a draft that is complete, sourced, and orderly but still flat, templated, or unpersuasive. ## Instructions **Main context only** — this is a subjective reading audit. Subagents cannot judge voice or human texture. **Skills to load**: `voice-and-humanization` (opinion emergence, dynamic AIO/bold guidance, anti-AI cleanup), `structure-variation` (rhythm audit checklist), `quality-gates` (Gate 4 criteria, output template at `templates/reader-qa.md`). **No MCP tools needed.** --- ### Required Checks 1. **Reader pull**: after each section, does the reader have a reason to continue? 2. **Flatness detection**: any section that is correct but has no tension, judgment, scene, or consequence? 3. **Template feeling**: do sections repeat the same opening, proof move, close, or emotional shape? 4. **Benchmark energy**: does the draft preserve the selected benchmark's pace, curiosity, and persuasion logic, not just its outline? 5. **AIO integration**: do extractable answers feel natural in prose, or do they interrupt the benchmark rhythm? 6. **Voice lens execution**: does prose match the scene without forcing lens names or styles? 7. **Opinion intensity**: too many strong opinions? too few? (`voice-and-humanization` Part 2) 8. **Bold usage**: judgments worth extracting, or keyword stuffing? (`voice-and-humanization` Part 3) 9. **Transitions**: forward pull vs backward summary 10. **Human texture & Named Expert inventory**: conditional judgments, trade-offs, anti-sell honesty, named projects, field language, and credible Named Engineer/Author voice 11. **Supplier voice detection**: flag sentences a competitor's sales manager could say 12. **Opening & Above-the-Fold Navboost strength**: does the first viewport (first ~180 words) reframe, answer directly, and route the reader without pogo-sticking? 13. **Self-Contained RAG Citation Chunks (GEO / AI Search)**: can the opening 40–60 words of key question H2s be extracted standalone by Perplexity / SearchGPT / Google AIO (explicit subject entity + condition + verdict + metric, without dangling pronouns)? 14. **Buying Committee Champion Check**: does the article give the technical specifier a concrete summary/ledger they can forward to their procurement/CFO approver? 15. **Back-Half Density Parity & Anti-Pattern-Fatigue Check**: do H2-05 through the final H2 maintain the same forensic specificity, workshop scene depth, and paragraph rhythm as H2-01–H2-04 (no late-draft collapse into thin bullet outlines), and does the article avoid repeating identical `> ### → Related:` callout templates across consecutive sections? ### Numeric Control Vector Audit Do not grade with fake scores. Diagnose whether the draft chose and executed concrete control numbers: | Control | Missing / Under-Controlled Signal | Over-Controlled Signal | |---|---|---| | Imitation / injection / differentiation | draft ignores the benchmark's rhythm and becomes generic | draft copies the benchmark shape but hides project evidence | | AIO / bold intensity | buyer questions are answered vaguely or too late; no full-sentence bold AI-citation anchor per H2 | short navigation bolds inflated for no reason (table headers, single words, dollar amounts, every list-item lead); OR a full-sentence bold so long it reads as a screamed paragraph that breaks prose flow. **One full-sentence bold per H2 is the default, not the over-signal — do not penalize.** | | Opinion intensity | sections explain facts without judgment or consequence; default sits below 4 when buyer risk is high | every paragraph sounds like a hot take at 9-10; OR stance hits 7-8 but with hype words instead of quantified judgments (see voice-and-humanization Stance vs Hype Boundary). **7-8 is the stance climax default for B2B buyer guides — do not penalize.** | | Evidence pressure | decision-critical claims lack proprietary proof | setup/transition sections read like technical manuals | | Format rhythm | sections feel interchangeable by section 2 or 3 | formats change for novelty and break the benchmark's reader journey | If the Numeric Control Vector was not stated upstream, **BLOCK** and route back to `/article-draft` to add `Draft Control Metadata`. Do not silently infer missing controls. If the vector exists but execution differs materially from the selected numbers, diagnose the drift instead of rewriting blindly. Do not require exact math; focus on whether the numbers gave real direction. ### Pass/Block - **BLOCK**: missing Numeric Control Vector, complete-but-flat reading experience, no forward pull, visible template assembly, AIO answers that interrupt prose, opening doesn't reframe, reads like supplier catalog, voice lenses consistently misapplied - **PASS**: reads like one continuous human-authored piece with natural rhythm ## Output Write `articles/<article-slug>/reader-qa.md`. Update `state.json`: `currentStage: article-reader-qa`, `nextCommand: /article-enrich`. ## Examples ### Example 1: Reader QA Validation 输入: `/article-reader-qa content-workspace/steel-screen screen-article` 输出: `reader-qa.md generated. Status: PASS (Supplier voice instances: 0).` ## Guardrails - **Supplier Voice Ban**: Reject the draft if phrases like "highest quality", "excellent service", or "industry leader" are found. - **Flatness Ban**: Reject sections that are merely complete. A section must advance tension, judgment, proof, or reader decision. - **No Mechanical Readability Rule**: Do not block solely because a paragraph length or sentence length is outside a fixed pattern. Use benchmark rhythm as the reference. - **No Missing-Control Drift**: If the draft never states or follows its ratio, AIO/bold targets, opinion intensity, or evidence pressure, flag the missing steering decision before repair. ## PAUSE ``` === ⏸ 停止 · 对等共研与校准(严禁跳步)=== 请与我一起审阅并推敲本阶段的产出(人类与 AI 对等共研,唯证据、买家意图与逻辑胜出): · 批准 / 继续 — 共识达成,按计划推进 · 补充 / 探讨:[粘贴你扩散搜索到的新素材、新角度或质疑,我们交叉验证] · 修改:[你的修改要求] · 停止 — 终止 ========================================== ``` Wait for explicit approval before proceeding.
Wersje
W bazie nie ma innych skilli o nazwie article-reader-qa.
Migawka bazy z 13.11.2026 · wersja paczki 9819b245ba99 (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 100humanize
Przepisuje tekst wygenerowany przez AI na bardziej naturalny, usuwa typowe ślady maszynowego stylu i dopasowuje ton, rytm zdań i słownictwo. - 90improve-readability
Poprawia czytelność tekstów: gramatykę, przepływ i wpływ na czytelnika, zachowując sens i głos autora. - 100text-humanizer
Poprawia tekst tak, by brzmiał naturalnie, jasno i odpowiednio dla odbiorcy, zachowując sens autora: ton, czytelność i naturalny styl po angielsku i holendersku. - 96readability-enhancer
Poprawia czytelność tekstu, dopasowując go do opcjonalnie wskazanego poziomu czytelnika. - 100writing-style-and-tone
Uczy techniki stylu i tonu: rytm, konkret, modulacja tonu i przebiegi redakcyjne, dzięki którym tekst brzmi ludzko, gdy posty „brzmią jak ChatGPT” lub szkic jest poprawny, ale męczący. - 100convert-to-natural-writing
Audytuje i przekształca tekst tak, by miał naturalny ludzki rytm, usuwając kalki z tłumaczeń na turecki i syntetyczne wzorce pisania AI.