prose-writer_skill

Overview

This skill turns approved outlines and verified evidence packs into a section-by-section survey draft or a one-page snapshot, producing output/DRAFT.md or output/SNAPSHOT.md. It enforces evidence-first constraints: only citation keys from citations/ref.bib are allowed, and writing is blocked until approvals and complete evidence are present. The result reads like a research paper: claim → cited evidence → synthesis, with explicit evaluation anchors and limitations per subsection.

How this skill works

Before any prose is produced, the skill checks DECISIONS.md for an Approve C2 flag and validates each H3 brief and its evidence pack for completeness and absence of scaffold markers. For each subsection it runs a planner pass (thesis, contrasts, evaluation anchor, limitation, citation placement) and then generates paragraph-level prose that embeds citations inside the sentences that carry claims. If required evidence is missing or would require inventing facts, the skill halts and requests upstream fixes.

When to use it

• You have an approved structure (Approve C2 present) and completed evidence packs for every H3.
• You want a draft that enforces per-subsection argument moves (thesis, contrasts, evaluation anchor, synthesis, limitation).
• You require strict citation verifiability using only keys in citations/ref.bib.
• You need a snapshot summary of available evidence and outstanding gaps.
• You must avoid pipeline voice or invented claims in the final prose.

Best practices

• Run the gate check early: confirm DECISIONS.md and evidence pack completeness before drafting.
• Ensure each subsection brief fills scope_rule, rq, axes, clusters, and paragraph_plan with no TODOs.
• Provide concrete evidence snippets and at least three concrete comparisons per evidence pack to enable synthesis.
• Place evaluation anchors (task/metric/constraint) in the same sentence as quantitative claims.
• Keep transitions minimal and content-bearing between adjacent sections to preserve coherence.

Example use cases

• Generate a full survey draft once H3 briefs and evidence drafts are complete and Approve C2 is set.
• Produce a one-page snapshot summarizing which H3s have sufficient evidence and which need more work.
• Enforce citation-only writing when collaborating across a team to prevent invented references.
• Preflight subsections to identify gaps: thesis, contrasts, evaluation anchor, and limitation before committing to prose.