菜单
Stop repeated failed writing, coding, manuscript, rebuttal, or restructuring revisions when the same issue has gone through 3+ unsatisfactory edits, vague feedback such as still wrong/weird/unclear/weak/越改越乱, version contamination, or possible gap/claim/evidence/venue-fit drift.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Use when AI-assisted thesis edits risk claim drift, scope creep, loss of author control, or repeated revisions that fail to converge; provides spine cards, edit contracts, drift audits, revision escalation, and human gates.
Build and audit the manuscript argument system: intent, gap-contribution alignment, hierarchical claims, evidence balance, limitations, and reviewer attack surfaces. Use when a paper, thesis chapter, review article, or rebuttal needs explicit intent registers, contribution chains, claim hierarchy, argument maps, evidence-fit checks, or reviewer-risk matrices before drafting, revision, peer review, or self-review.
Review another author's manuscript, paper, thesis chapter, proposal, or preprint as an external reviewer. Use when asked to evaluate novelty, significance, gap-contribution fit, claim-evidence adequacy, methods, evaluation, overclaim risks, structure, writing, required revisions, or recommendation without rewriting the manuscript or using private author context.
Review the user's own manuscript, paper, thesis chapter, rebuttal, or release packet with clean-room anti-contamination controls. Use when asked for self-review, internal review, pre-submission review, readiness check, reviewer simulation on own work, or claim-evidence self-audit where prior chat memory, unstated assumptions, model background knowledge, or unlisted local notes must not be treated as evidence.
Reframe academic manuscript drafts that read like engineering reports, module inventories, internal validation packets, or system notes into paper-form scientific arguments with clear gap, contribution chain, results narrative, figure/table roles, AI-assisted component boundaries, and submission-readiness blockers.
Use when validating, repairing, or mapping human-evaluation handoff packages, filled annotation CSVs, rebuttal annotation UIs, or reviewer annotation returns across package versions. Applies to checking row alignment, detecting cross-snapshot contamination, converting old labels to current schemas, generating refill/import CSVs, and deciding whether filled annotations are safe for formal aggregation.
| name | revision-escalation |
| description | Stop repeated failed writing, coding, manuscript, rebuttal, or restructuring revisions when the same issue has gone through 3+ unsatisfactory edits, vague feedback such as still wrong/weird/unclear/weak/越改越乱, version contamination, or possible gap/claim/evidence/venue-fit drift. |
| allowed-tools | Read, Glob, Grep, Edit, Write, Bash |
Prevent repeated local patches from making a draft or code path more inconsistent. Use this when the issue may no longer be wording or implementation detail, but specification, structure, evidence, or version-control drift.
This skill activates on: revision escalation, 3-strike, three strikes, stop and diagnose, still wrong, still weird, unclear, weak, 越改越乱, 还是不对, 还是怪, 不够清楚, 逻辑还是混乱, /revision-escalation.
If the same issue remains unresolved after 3 revision attempts, treat it as a specification or structure problem before treating it as another local editing task.
Do not make a fourth patch immediately.
Before editing again, classify the problem:
| Category | Meaning | Next action |
|---|---|---|
| Underspecified request | Target, constraint, audience, venue, or expected output is missing. | Ask for a concrete target before editing. |
| Ambiguous feedback | Feedback is evaluative but not operational: "weird", "weak", "unclear", "not good enough". | Ask what should change. |
| Local execution problem | The goal is clear, but the previous patch implemented it incorrectly. | Make one small targeted patch. |
| Structural mismatch | The issue affects the research question, gap, contribution, evidence chain, section structure, module boundary, or venue framing. | Propose a restructure plan before editing. |
| Evidence gap | The desired claim is unsupported by available data, experiments, citations, or files. | Downgrade the claim or request evidence. |
| Version contamination | Repeated patches have mixed old assumptions with new requirements, causing inconsistency, duplication, or bloat. | Recommend a new version, branch, or consolidated brief. |
When triggered, respond in this structure before any patch:
I should pause before making another patch.
This issue has already gone through several revision rounds and may not be a local wording or implementation problem anymore.
Current diagnosis:
* Category:
* Why:
* What is missing or conflicting:
* Recommended next action:
Options:
A. Clarify the concrete target and continue a local patch.
B. Consolidate all current requirements into a single brief, then retry.
C. Create a new version or branch and restructure the section/module.
D. Reframe the paper/project from the research question, gap, and evidence chain.
Classify manuscript work before editing:
For full reframing, do not directly rewrite the manuscript. First produce a reframing brief with target venue, research question, gap, core claim, available evidence, claims that must not be made, and proposed new structure.
Stop and diagnose when thinking:
Never continue accumulating edits on a structurally inconsistent manuscript, rebuttal, or code path.