content-qa

Content QA skill

The adversarial quality-assurance pack — the QA moat. These prompts drive the multi-model QA stage: topic-delivery and internal-consistency gates, the publication-readiness critic, the aggregate-rewrite pass, writer self-review for contradictions, the self-consistency sampling rail, an LLM quality rubric, and the two vision-QA prompts. The architect routes on the description above; UnifiedPromptManager resolves each template by key (Langfuse override still wins over the bodies below).

Default prompts — basic but functional; production-quality prompt packs ship as a premium add-on.

qa.content_review

Review this content for quality. Return JSON with keys: score (1-10), issues (list), suggestions (list).
Content: {content}

qa.self_critique

Self-critique this content. Return JSON with keys: strengths (list), weaknesses (list), improvements (list).
Content: {content}

qa.topic_delivery

You are a strict editor checking whether an article
delivers on its topic. A reader clicking this article expects what the topic
promises. Did the writer deliver?

REQUESTED TOPIC: {topic}

ARTICLE OPENING (first ~1000 words):
{opening}

Check these specific failure modes:

  1. Numeric promises. If the topic says "10 X" or "11 Y" or "5 Z", does the
     body actually list that many? Partial lists (two items then a pivot to
     generalities) FAIL.
  2. Named entities. If the topic names a specific product, person, or
     technology ("Llama 4", "Claude", "indie hackers making $1M+"), does the
     body actually discuss that specific thing? An article titled "Llama 4"
     that only discusses Llama 3.1 FAILS.
  3. Format promise. If the topic implies a guide, tutorial, list, or review,
     does the body deliver that format? A "guide" that's actually an opinion
     piece FAILS.
  4. Angle/thesis. Is the article's thesis actually about the topic, or did
     the writer pivot to a tangential point they preferred?

Respond with ONLY valid JSON:
{{"delivers": true/false, "score": NUMBER 0-100, "reason": "concise — name the specific gap when one exists"}}

Scoring guidance: delivers=true and score 85-100 if the body is a faithful
execution of the topic. delivers=false and score 0-40 if the body is a
bait-and-switch or numeric underdelivery or misnamed version. delivers=true
and score 60-80 if the body is mostly on-topic but weaker than the topic
implies.