manuscript-reframe
Reframe report-like academic drafts into paper-form scientific arguments while preserving or explicitly renegotiating author intent; requires an approved old-versus-proposed spine, evidence and argument baselines, analysis-role control, and post-edit drift review.
yha9806 · 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 yha9806/academic-writing-toolkit --skill manuscript-reframe技能说明
name: manuscript-reframe description: Reframe report-like academic drafts into paper-form scientific arguments while preserving or explicitly renegotiating author intent; requires an approved old-versus-proposed spine, evidence and argument baselines, analysis-role control, and post-edit drift review. allowed-tools: Read, Glob, Grep, Edit, Write, Bash
/manuscript-reframe - Scientific Manuscript Reframing
Purpose
Turn a technically valid but report-like manuscript into a scientific paper argument. Use this when a draft contains many modules, raw tables, internal artifact language, scattered figures, or exploratory AI/report-generation material but the scientific gap and contribution are not yet crisp.
Trigger Words
This skill activates on: manuscript reframe, paper form, too much like a report, gap and contribution, scientific framing, journal framing, submission clean, methods paper, /manuscript-reframe.
Core Rule
Write around the scientific problem, not the implementation inventory.
Reframing may reorganise the manuscript argument. It may not silently redefine the real-world problem, intended use, research object, core question, primary experiment, contribution order, headline claim, or evidence boundary. Treat changes to any of those items as an author decision, not as polishing.
Before editing, force the draft into one sentence:
This paper solves [specific reliability / evidence / workflow gap] between [upstream technical output] and [downstream scientific or operator-facing use].
If the sentence cannot be written, do not rewrite sections yet. First produce a gap-and-contribution diagnosis.
Workflow
0. Lock Author Intent And The Reframe Contract
Read the canonical /thesis-control intent and manuscript contract. For a
lightweight project, read 00_AUTHOR_INTENT.md,
01_EVIDENCE_AND_CLAIMS.md, and 02_REVISION_LOG.md. Record:
- real-world problem and intended use
- current method or software task
- research object and core scientific question
- primary experiment and supporting analyses
- headline claim and evidence boundary
- concepts that must remain visible in the title or abstract
- evidence baseline and argument baseline
Then show an old-versus-proposed comparison for every item above, plus the
title/abstract thesis, contribution order, and proposed structure. Mark each
row preserve, narrow, promote, demote, remove, or change. Do not
edit the manuscript until the author explicitly approves the proposed reframe.
If no approved intent exists, stop and establish it through /thesis-control.
Do not infer author intent from whichever AI-edited version happens to be
latest.
Keep application purpose and evidence conclusion separate. Lack of hardware, clinical, causal, deployment, or transfer validation narrows the conclusion; it does not by itself make the motivating application irrelevant.
1. Locate The Current Draft And Evidence Anchors
Inspect manuscript sources, figures, tables, supplementary files, result summaries, reference files, and revision reports. Identify which artifacts are canonical and which are exploratory.
Create a short working note with:
- manuscript source path
- intended venue or article type
- canonical result artifacts
- exploratory-only artifacts
- forbidden or overclaim risks
- unresolved submission metadata
2. Diagnose The Paper Spine
Check whether the Introduction answers:
- What do existing approaches output?
- Why is that output insufficient for downstream use?
- What gap does this paper evaluate or close?
- What is the proposed interface, workflow, or method?
- What evidence shows the gap and the method effect?
Prefer gap forms such as:
- Perception outputs are not automatically evidence objects.
- Segmentation metrics do not guarantee downstream task reliability.
- Uncertain perception should be routed rather than silently converted into support.
- Reports and interfaces need explicit claim boundaries.
- Machine-readable records must not imply execution permission unless validated for that role.
3. Rebuild Contributions As A Chain
Avoid module lists. Use a small contribution chain:
- Problem formulation: define the reliability, evidence, or workflow gap.
- Interface definition: convert raw outputs into structured evidence objects.
- Bottleneck measurement: quantify where upstream metrics and downstream claims diverge.
- Conservative routing: expose reliability-coverage trade-offs and review capture.
- Bounded output: produce traceable reports or records without unsupported claims.
Every contribution should answer: what gap does this close, and what evidence supports it?
Before changing contribution prominence, apply the analysis-admission record
from 01_EVIDENCE_AND_CLAIMS.md or the canonical argument-governance packet.
Classify each experiment or analysis as primary, explanatory,
robustness, development_record, or out_of_scope. A completed or fully
auditable analysis does not automatically belong in the main text. Record and
obtain author approval for every promotion or demotion.
4. Rewrite Results As A Narrative Transition
Do not present Results as raw artifact counts. Use paragraphs that explain the transition:
- perception metric: what the frozen model or upstream system achieved
- direct conversion: what fails when outputs are treated as downstream evidence
- gate/routing effect: how selective support changes agreement and coverage
- failure taxonomy: which bottleneck dominates
- bounded output audit: whether reports or records stayed within evidence boundaries
Keep exact values in tables. In prose, translate audit-zero metrics into natural language when clearer:
The audit detected no unsupported statements, contradictions, or prohibited action claims.
Do not write prose that sounds like a metric dump:
unsupported rate 0.000, contradiction rate 0.000, prohibited rate 0.000