Polish academic prose, de-AI-ify text, control claim strength, or rewrite method sections for CS/AI/ML papers. Executes Prose Quality Gate, Claim Strength Audit, and de-AI pass. Use when: removing AI writing patterns from paper text, adjusting claim strength to match evidence level, rewriting method sections with proper narrative flow, improving academic writing quality, checking for overclaiming. Triggers on: 润色, polish, improve writing, 去AI, de-AI, claim strength, 改写, rewrite method, prose quality, 降级表述, remove AI patterns, academic writing polish, 学术润色, 去AI化, 降级结论, improve prose.
Installation
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Read manifest.yaml. It declares always_load files, axes, and references.on_demand.
Read every file listed under always_load. These are the skill's binding rules — not reference material.
Apply the loaded material as constraints:
stance.md defines non-negotiable rules, AI Traffic Light, claim strength, evidence-aware boundary, and scope.
red-lines.md defines absolute prohibitions. Do not negotiate these.
output-contract.md defines deliverables and completion criteria per mode.
anti-patterns.md defines known failure modes and their correct alternatives.
Detect the mode using the manifest's mode axis. If ambiguous, ask one concise question.
Echo the selected mode to the user before executing.
Reach for references/ only when the manifest's references.on_demand condition is satisfied.
Structural debt is not language debt. If section_contract_debt = open, only apply local safe edits and return the diagnosis.
Modes
Mode
Use when
prose-quality-gate
General prose quality check + rewrite
method-prose-rewrite
Method section narrative: 问题→设计→机制→收益/边界
de-ai-pass
Remove AI writing patterns only
claim-strength-audit
Audit and adjust claim strength
Execution
Called internally by academic-paper-writer orchestrator at Step 9.6. The main agent reads this file and references/ to execute polishing in-process (no subagent dispatch). Polishing must not modify project source code, configuration, or data files. Max 2 rewrite rounds.
