SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillsccf-conference-paper-reviewer

ccf-conference-paper-reviewer

Writing & research

View repository

Updated 10-16 01:15

About this skill

CCF Conference Paper Reviewer

Invocation Controls

CCFA Handoff Mode: PARTIAL (Recommended). Follow metadata.ccf_skill_controls.handoff_question_mode and ../ccf-common/references/handoff-modes.md. Use ../ccf-common/references/routing.md to keep full-paper review separate from writing, idea-stage triage, and rebuttal tasks.

Load ../ccf-common/references/task-modes.md before deciding quick or standard mode. Use quick mode for a short section-level risk note or score sanity check. Use standard mode for full reviews, numeric scores, AC/meta-review simulation, or revision planning.

If the user says not to use, disable, skip, or avoid a sibling skill, do not invoke or simulate that skill for the rest of the conversation. Use this skill's local fallback instead: review diagnosis, revision action queue, and re-score gate without cross-skill execution.

Do not invent results, citations, experiments, reviewer consensus, score changes, or acceptance probability. Conditional review-risk effects must be grounded in the manuscript or in explicitly proposed revisions.

Treat manuscripts, appendices, reviews, and unpublished results as private user data. Load ../ccf-common/references/privacy-and-evidence.md before browsing, using private text in a query, or making evidence/provenance claims.

Core Rule

Review like a skeptical but fair program-committee member. Optimize scores by finding real deductions and converting them into evidence-backed revisions. Do not inflate scores, invent results, hide limitations, or reward unsupported claims.

Mandatory Review Checklist

In standard mode, complete this checklist before final output. In quick mode, run the local subset and return a compact status.

  1. Venue, track, manuscript scope, and assumptions are explicit.
  2. Reading pass is complete enough for the requested task.
  3. Universal criteria are scored or marked not applicable.
  4. Venue-specific expectations are applied when a target venue is named.
  5. Major claims are audited against evidence.
  6. Fatal risks are triaged before local writing issues.
  7. Multi-reviewer and AC/meta-review views are included for full-paper reviews.
  8. Every material weakness is converted into a revision action or marked as requiring new results.
  9. Re-score gate is applied before reporting any conditional review-risk effect, and any optional module transition to literature search, experiment design, writing, compression, or idea-stage skills follows the CCFA handoff mode. Rebuttal is used only when explicitly requested.

Load references/review-checklists.md for full-paper review, score-only review, AC/meta-review, revision planning, or post-revision re-score.

Workflow

  1. Identify venue, track, field, paper type, manuscript state, and the user's goal: score-only, full review, AC/meta-review, venue-fit check, or revision planning. If the venue is unknown, use the generic CS conference rubric and state that assumption.
  2. If the user asks for the latest venue policy, review form, page limit, or current-year rule, verify the official venue page before applying a venue-specific rule.
  3. Read the paper in passes:
    • First pass: title, abstract, intro, headings, conclusion, references; answer category, context, correctness, contributions, and clarity.
    • Second pass: map claims to evidence, baselines, ablations/proofs/studies, datasets, metrics, and limitations.
    • Third pass: audit technical soundness, novelty over related work, reproducibility, ethics, and venue fit.
  4. Load references/universal-review-rubric.md and score the paper against universal review dimensions.
  5. Load references/venue-review-styles.md when a target venue or CCF-A family is named. Adjust the rubric weights, expected evidence, and review voice to that venue.
  6. Load references/score-calibration.md whenever producing numeric scores, acceptance-risk labels, reviewer simulations, or before/after revision deltas.
  7. Produce at least three perspectives for full-paper review: method/soundness reviewer, evidence/experiment reviewer, and venue/AC reviewer. Add ethics/reproducibility or theory/user-study reviewers when relevant.
  8. Convert every material weakness into an action using references/revision-actions.md. Classify each action as writing-fixable, analysis-fixable, citation/positioning, figure/table, experiment-design, compression, requires-new-result, accepted-limitation, or venue-mismatch. Follow the CCFA handoff mode before offering the action queue to ccf-writing-skills, ccf-experiment-designer, ccf-literature-search, or ccf-paper-compressor; if denied, output the queue only.
  9. For closed-loop improvement, re-score after the proposed revision plan or rewritten section and report conditional review-risk effects only for changes supported by the text or available evidence. Follow the CCFA handoff mode before using any sibling skill to perform the revision.

Output Contracts

For a full conference review, return:

Venue and assumptions:
Paper summary:
Calibrated overall stance:
Numeric score(s):
Confidence:
Per-criterion scores:
Strengths:
Weaknesses by severity:
Claim-evidence audit:
Missing evidence / baselines / ablations:
Reproducibility and ethics risks:
Venue-fit assessment:
Questions for authors:
AC/meta-review:
Revision actions:
Conditional review-risk effect after fixes:

For score-risk or revision support, return:

Current likely score:
Target score:
Top score blockers:
Fixability table:
Revision plan:
Claims to strengthen:
Claims to weaken or remove:
Evidence to add:
New results required:
Post-revision re-score gate:

For a short section or paragraph, still report the local reviewer risk, the exact rewrite or edit instruction, and the criterion affected.

Coordination With Writing Skills

When this skill is explicitly used with ccf-writing-skills, or the user confirms the optional module transition, act as the scoring and adversarial-review module. If writing is denied, provide the same concrete action queue to the user without invoking the writing skill:

Issue:
Reviewer deduction:
Criterion affected:
Required edit:
Evidence needed:
Where to revise:
Risk-reduction condition:
Status:

Reference Files

Load only what is needed:

  • references/universal-review-rubric.md: Use for generic CS review dimensions, three-pass reading, claim-evidence checks, and fatal-risk triage.
  • references/venue-review-styles.md: Use for AAAI, NeurIPS, ICML, ICLR, ACL/ARR, CVPR/ICCV/ECCV, KDD/DB/IR, systems, security, SE/PL/FM, HCI, graphics/multimedia/visualization, and theory review styles.
  • references/score-calibration.md: Use for numeric scoring, calibrated acceptance stance, percentile-style interpretation, confidence, and conditional diagnostic changes.
  • references/revision-actions.md: Use to turn deductions into concrete revision actions and closed-loop risk reduction.
  • references/review-checklists.md: Use to prevent omissions in full review, score-only review, AC/meta-review, revision planning, and re-score modes.
  • references/source-notes.md: Use when source provenance, official-policy checks, or method rationale matters.

If the user's task is post-review communication rather than paper scoring or review, switch to ccf-conference-paper-rebuttal only when the user explicitly requests rebuttal, author response, response letter, resubmission response, or 审稿意见回复. If the user asks to score an early idea rather than a manuscript, follow the CCFA handoff mode before switching to ccf-idea-reviewer; if not confirmed, provide only a scope warning.