AgentSkillsCN

rubric-writer

基于评分标准撰写同行评审报告（`output/REVIEW.md`），综合提炼出关键论断与证据缺口（包括新颖性、稳健性、清晰度与影响力）。 **触发条件**：评分标准评审、审稿人报告、同行评审撰写、审稿报告、REVIEW.md。 **适用场景**：在同行评审流程的最后阶段（C3），已产出 `output/CLAIMS.md` 与 `output/MISSING_EVIDENCE.md`（以及可选的新颖性矩阵）时。 **跳过条件**：若上游产出尚未就绪（论断或证据缺口缺失），或您无意输出完整的审稿报告。 **网络需求**：无。 **约束规则**：提供切实可行的改进建议（可操作的反馈），并全面覆盖新颖性、稳健性、清晰度与影响力；避免泛泛而谈。

SKILL.md
--- frontmatter
name: rubric-writer
description: |
  Write a rubric-based peer review report (`output/REVIEW.md`) using extracted claims and evidence gaps (novelty/soundness/clarity/impact).
  **Trigger**: rubric review, referee report, peer review write-up, 审稿报告, REVIEW.md.
  **Use when**: peer-review pipeline 的最后阶段（C3），已有 `output/CLAIMS.md` + `output/MISSING_EVIDENCE.md`（以及可选 novelty matrix）。
  **Skip if**: 上游产物未就绪（claims/evidence gaps 缺失）或你不打算输出完整审稿报告。
  **Network**: none.
  **Guardrail**: 给可执行建议（actionable feedback），并覆盖 novelty/soundness/clarity/impact；避免泛泛而谈。

Rubric Writer (referee report)

Goal: write a complete review that is grounded in extracted claims and evidence gaps.

Role cards (use explicitly)

Referee (fair but sharp)

Mission: evaluate novelty/soundness/clarity/impact with evidence-backed, actionable feedback.

Do:

  • •Tie critiques to extracted claims and evidence gaps (not impressions).
  • •Separate major vs minor issues; propose minimal fixes.
  • •Keep tone calm and professional.

Avoid:

  • •Turning the review into a rewrite of the paper.
  • •Generic comments ("needs more experiments") without specifying which and why.

Reproducibility Auditor

Mission: identify missing details that block replication and fair comparison.

Do:

  • •Ask for protocol details, baselines, ablations, and threat models where missing.
  • •Flag underspecified quantitative claims (metric/constraint not stated).

Avoid:

  • •Assuming details that are not present in the claims/evidence.

Role prompt: Referee Report Writer

text
You are writing a referee report.

Your job is to be useful to authors and reviewers:
- summarize contributions (bounded)
- evaluate novelty/soundness/clarity/impact
- list actionable major concerns (problem -> why it matters -> minimal fix)
- list minor comments

Constraints:
- ground critique in output/CLAIMS.md and output/MISSING_EVIDENCE.md
- avoid vague requests; specify the missing baseline/metric/protocol detail

Style:
- professional, concise, specific

Inputs

Required:

  • •output/CLAIMS.md
  • •output/MISSING_EVIDENCE.md

Optional:

  • •output/NOVELTY_MATRIX.md
  • •DECISIONS.md (if you have reviewer constraints/format)

Outputs

  • •output/REVIEW.md

Workflow

  1. •

    If DECISIONS.md exists, follow any required reviewer format/constraints.

  2. •

    One-paragraph summary (bounded)

    • •Summarize the paper’s goal + main contributions using output/CLAIMS.md.
  3. •

    Rubric sections

    • •Novelty: reference output/NOVELTY_MATRIX.md (if present) and/or the related work discussion.
    • •Soundness: reference the concrete gaps from output/MISSING_EVIDENCE.md.
    • •Clarity: identify the top issues that block understanding/reproduction.
    • •Impact: discuss likely relevance if the issues were fixed.
  4. •

    Actionable feedback

    • •Major concerns: each with “problem → why it matters → minimal fix”.
    • •Minor comments: clarity, presentation, missing details.
  5. •

    Final recommendation

    • •Choose a decision label and justify it primarily via soundness + evidence quality.

Mini examples (actionable feedback)

Major concern template (good):

  • •Problem: The main performance claim is underspecified (task/metric/budget not stated).
  • •Why it matters: Without a fixed protocol, comparisons to baselines are not interpretable.
  • •Minimal fix: Add a table that lists task, metric definition, budget/tool access assumptions, and seeds; rerun the main comparison under that protocol.

Generic (bad):

  • •The paper needs more experiments.

Definition of Done

  • • output/REVIEW.md covers novelty/soundness/clarity/impact.
  • • Major concerns are actionable (each has a minimal fix).
  • • Critiques are traceable to output/CLAIMS.md / output/MISSING_EVIDENCE.md (not free-floating).

Troubleshooting

Issue: review turns into a rewrite of the paper

Fix:

  • •Cut; keep to critique + actionable fixes and avoid adding new content.

Issue: review is generic (“needs more experiments”)

Fix:

  • •Replace with concrete gaps from output/MISSING_EVIDENCE.md (which baseline, which dataset, which ablation).