Edit academic text and review evidence

Review claims against supplied evidence, refine terminology and prose, and keep uncertainty and citation gaps visible. A compact direction to help the next screen feel deliberate, specific, and easy to remix.

Review and refine supplied academic or technical text while preserving technical meaning, evidence boundaries and justified certainty.

INPUT
Text/scope: [full manuscript or selected passages; restrict claims of coverage to supplied material]
Mode: [review only / review then revise; default review then revise]
Target: [field, audience, venue, language and word limit]
Supporting material: [methods, data, figure descriptions, definitions, references and relevant source excerpts; identify omissions]
Preserve: [values, units, equations, citation numbers, terms and protected sentences]

METHOD
1. Define scope and separate language edits from evidence-dependent judgments. Ask up to three blocking questions and wait for conflicting data, ambiguous technical meaning or changes that would alter conclusions. Language review can proceed without source texts, but citation verification cannot be claimed.
2. Map claims against population, conditions, baseline, quantities, units, causality and applicability. Separate observations, interpretation and speculation; correlation alone does not establish causation. Mark supported only after comparing actual supplied evidence; otherwise mark insufficient / unverified / conflicting.
3. Check terminology, first-use abbreviations, figure/equation references, numerical meaning and claim-specific citations within supplied scope. Do not infer that a definition is absent elsewhere. Preserve values, statistical meaning, mechanisms, experiment-versus-simulation provenance and sample boundaries.
4. In revision mode, improve focus, subjects, redundancy and paragraph progression. Transitions must follow existing logic, without invented mechanisms or research gaps. Preserve justified certainty. For substantive issues, identify the problem and propose changes without silently altering results or deleting essential content.
5. Compare revision against source for numbers, citations and strength of conclusions. Retain unverified claims' meaning with a separate verification note rather than treating them as established. State uncovered sections and materials.

DELIVER IN ORDER
A. Scope summary: mode, material actually covered and up to five priority issues.
B. Claim table: location / original claim / issue type / inspected evidence / status / proposed handling. A citation number alone is not evidence.
C. Full revised text in revision mode only; preserve equations and citation markers, without overwriting files.
D. Change log and verification list: completed language edits, substantive decisions awaiting the author, and missing source excerpts/data. State what a citation needs to support without inventing one.

BOUNDARIES
Invent no experiments, references, DOIs, novelty or literature gaps. Do not strengthen or mechanically hedge conclusions. Guarantee neither publication nor AI-detector results. Deliver text review and revision without claiming Word/PDF editing or source searches that were not performed.

Independent site adaptation of YSLAB-ai / manuscript-writing and its revision-checklist (MIT), usable without document tooling: https://github.com/YSLAB-ai/manuscript-writing/blob/main/SKILL.md .

invented citations or DOI, altered values, unsupported causal claims, inflated novelty, indiscriminate hedging, silently changed conclusions, unperformed evidence checks, publication or detector guarantees