Scholar Rebuttal Pro

Scholar Rebuttal Pro

Enhanced academic paper review response workflow combining Agy/CLI collaborative analysis with multi-perspective discussion. Produces structured, evidence-based rebuttal documents optimized for conference-specific requirements.

Pre-load (before execution)

1. Codebase docs: If .workflow/codebase/ARCHITECTURE.md exists, read for project context
2. Specs: maestro load --type spec --category coding — load coding conventions
3. Wiki knowledge: maestro search "academic writing research paper" --json — top 5 entries as prior context
4. All optional — proceed without if unavailable

Architecture Overview

┌─────────────────────────────────────────────────────────────────┐
│  Scholar Rebuttal Pro Orchestrator (SKILL.md)                    │
│  → Pure coordinator: Execute phases, parse outputs, pass context  │
│  → Run lifecycle: create/resume → phases → check → complete       │
└───────────────────────┬─────────────────────────────────────────┘
                        │
    ┌─────────┐ ┌─────────┐ ┌─────────┐ ┌─────────┐ ┌─────────┐
    │ Phase 1 │ │ Phase 2 │ │ Phase 3 │ │ Phase 4 │ │ Phase 5 │
    │ Review  │ │ Multi-  │ │Strategy │ │Rebuttal │ │ Quality │
    │ Parsing │ │Perspect │ │Formula  │ │ Writing │ │Validat  │
    └─────────┘ └─────────┘ └─────────┘ └─────────┘ └─────────┘
      reviewA    discussion   strategy    rebuttal    quality
      nalysis    Consensus    Matrix      Draft       Score

Key Design Principles

1. CLI-Assisted Analysis: Leverage Agy CLI for semantic analysis, evidence gathering, and quality validation
2. Multi-Perspective Discussion: Simulate author/reviewer/expert viewpoints to develop robust strategies
3. Evidence-Based Responses: Link every response to paper content or experimental evidence
4. Conference-Agnostic Templates: Support extensible template system for different venues
5. Progressive Disclosure: Load phase documents on-demand to manage context window

Interactive Preference Collection

Collect workflow preferences via AskUserQuestion before dispatching to phases:

const prefResponse = AskUserQuestion({
  questions: [
    {
      question: "是否跳过所有确认步骤（自动模式）？",
      header: "Auto Mode",
      multiSelect: false,
      options: [
        { label: "Interactive (Recommended)", description: "交互模式，每阶段后确认" },
        { label: "Auto", description: "跳过所有确认，自动执行" }
      ]
    },
    {
      question: "论文内容来源？（用于策略制定时查找支撑证据）",
      header: "Paper Source",
      multiSelect: false,
      options: [
        { label: "Provide Path", description: "指定论文 PDF/LaTeX 路径" },
        { label: "Current Directory", description: "自动搜索当前目录" },
        { label: "Review Only", description: "仅基于审稿意见回复" }
      ]
    },
    {
      question: "目标会议类型？（影响模板和策略选择）",
      header: "Conference",
      multiSelect: false,
      options: [
        { label: "ML Conferences", description: "NeurIPS/ICML/ICLR" },
        { label: "CV Conferences", description: "CVPR/ECCV/ICCV" },
        { label: "NLP Conferences", description: "ACL/EMNLP" },
        { label: "Generic", description: "通用模板" }
      ]
    }
  ]
})

// Derive workflowPreferences from user selection
workflowPreferences = {
  autoYes: prefResponse["Auto Mode"] === "Auto",
  paperSource: prefResponse["Paper Source"],
  conferenceType: prefResponse["Conference"]
}

workflowPreferences is passed to phase execution as context variable. Phases reference as workflowPreferences.autoYes, workflowPreferences.paperSource, etc.

Auto Mode Defaults

When workflowPreferences.autoYes === true:

• Skip confirmation after each phase
• Use recommended strategies from multi-perspective discussion
• Apply default conference template (Generic)
• Auto-proceed to quality validation

Execution Flow

⚠️ COMPACT DIRECTIVE: Context compression MUST check TodoWrite phase status. The phase currently marked in_progress is the active execution phase — preserve its FULL content. Only compress phases marked completed or pending.

Run Setup (see run-mode.md):
   └─ Birth packet injected run_id/run_dir? → use them, skip create.
      Else self-start: maestro run create scholar-rebuttal-pro --session <YYYYMMDD-scholar-rebuttal-pro-{topic}> --intent "..."
      (Optional --resume <run_id> → maestro run brief <run_id> to continue an existing Run.)
   └─ output_base = {run_dir}/outputs

Input Parsing:
   └─ Convert user input to structured format (reviewCommentsPath + paperPath + conferenceType)

Phase 1: Review Parsing & Classification
   └─ Ref: phases/01-review-parsing.md
      ├─ Tasks attached: Parse reviewer comments structure → Classify comments using Agy CLI → Extract sentiment and key concerns → Generate review-analysis.json
      └─ Output: reviewAnalysis, commentCategories, ${output_base}/review-analysis.json, ${output_base}/comment-classification.md

Phase 2: Multi-Perspective Discussion
   └─ Ref: phases/02-multi-perspective-discussion.md
      ├─ Tasks attached: Author perspective: effective response strategies → Reviewer perspective: persuasive arguments → Expert perspective: technical accuracy and academic norms → Synthesize consensus strategies
      └─ Output: discussionConsensus, strategicRecommendations, ${output_base}/discussion-log.md, ${output_base}/consensus-strategies.json

Phase 3: Strategy Formulation
   └─ Ref: phases/03-strategy-formulation.md
      ├─ Tasks attached: Map comments to response strategies → Search paper content for evidence using CLI → Identify gaps requiring new experiments → Generate strategy matrix
      └─ Output: strategyMatrix, evidenceMap, ${output_base}/strategy-matrix.md, ${output_base}/evidence-references.json

Phase 4: Rebuttal Writing
   └─ Ref: phases/04-rebuttal-writing.md
      ├─ Tasks attached: Apply conference-specific template → Write point-by-point responses → Integrate evidence and citations → Optimize professional tone
      └─ Output: rebuttalDraft, rebuttal.md, ${output_base}/rebuttal-draft-v1.md

Phase 5: Quality Validation
   └─ Ref: phases/05-quality-validation.md
      ├─ Tasks attached: Check completeness (all comments addressed) → Assess professionalism and tone → Evaluate persuasiveness and evidence strength → Generate improvement recommendations
      └─ Output: qualityScore, improvements, ${output_base}/quality-report.md, ${output_base}/improvement-suggestions.json

Run Closure (see run-mode.md):
   └─ maestro run check {run_id} → repair any reported gate → maestro session done {run_id}
      (Report success only after session done.)

Return:
   └─ Summary with recommended next steps

Phase Reference Documents (read on-demand when phase executes):

Compact Rules:

1. TodoWrite in_progress → 保留完整内容，禁止压缩
2. TodoWrite completed → 可压缩为摘要
3. 🔄 sentinel fallback → 带此标记的 phase 包含 compact sentinel；若 compact 后仅存 sentinel 而无完整 Step 协议，必须立即 Read() 恢复

Core Rules

1. Start Immediately: First action is TodoWrite initialization, second action is Phase 1 execution
2. No Preliminary Analysis: Do not read files or gather context before Phase 1
3. Parse Every Output: Extract required data from each phase for next phase
4. Auto-Continue: Check TodoList status to execute next pending phase automatically
5. Track Progress: Update TodoWrite dynamically with task attachment/collapse pattern
6. Progressive Phase Loading: Read phase docs ONLY when that phase is about to execute
7. DO NOT STOP: Continuous multi-phase workflow until all phases complete
8. CLI Integration: Use maestro delegate --to agy --mode analysis for semantic analysis tasks
9. Evidence Linking: Every response strategy must link to paper content or experimental evidence

Input Processing

User provides review comments in one of these formats:

1. File path: reviews.txt, reviewer-comments.md, reviews.pdf
2. Inline text: Paste reviewer comments directly
3. Structured JSON: Pre-parsed review structure

Optional flag: --resume <run_id> to continue an existing Run.

Run Resolution

The Run is the single source of truth (see run-mode.md). Resolve run_dir, then derive output_base:

// If the birth packet injected run_id/run_dir, use them (do NOT create).
// Else if --resume <run_id>: maestro run brief <run_id> → run_dir.
// Else self-start: maestro run create scholar-rebuttal-pro --session <slug> --intent "..."
//   (slug: YYYYMMDD-scholar-rebuttal-pro-{topic}, ASCII-only, ≤64 chars)

const output_base = `${run_dir}/outputs`;  // all phase outputs land here
const cleanArgs = $ARGUMENTS.replace(/--resume\s+\S+/, '').trim();

Structured Input

Convert to structured format:

const structuredInput = {
  reviewCommentsPath: <path or inline text>,
  paperPath: workflowPreferences.paperSource === "Provide Path" ? <user-provided> : <auto-discovered>,
  conferenceType: workflowPreferences.conferenceType,
  autoMode: workflowPreferences.autoYes,
  output_base: output_base  // {run_dir}/outputs — all phase outputs use this base path
}

Data Flow

User Input (review comments + paper path + conference type [+ --resume <run_id>])
    |
[Run Resolution]  (see run-mode.md)
    | run_dir = birth packet | --resume brief | self-start create
    | output_base = {run_dir}/outputs
    | mkdir -p ${output_base}
    |
[Convert to Structured Format]
    |
Phase 1: Review Parsing & Classification
    | Input: reviewCommentsPath + conferenceType
    | Output: reviewAnalysis + commentCategories
    | Files: ${output_base}/review-analysis.json, ${output_base}/comment-classification.md
    |
Phase 2: Multi-Perspective Discussion
    | Input: reviewAnalysis + commentCategories
    | Output: discussionConsensus + strategicRecommendations
    | Files: ${output_base}/discussion-log.md, ${output_base}/consensus-strategies.json
    |
Phase 3: Strategy Formulation
    | Input: discussionConsensus + strategicRecommendations + paperPath
    | Output: strategyMatrix + evidenceMap
    | Files: ${output_base}/strategy-matrix.md, ${output_base}/evidence-references.json
    |
Phase 4: Rebuttal Writing
    | Input: strategyMatrix + evidenceMap + conferenceType
    | Output: rebuttalDraft
    | Files: ${output_base}/rebuttal-draft-v1.md
    |
Phase 5: Quality Validation
    | Input: rebuttalDraft
    | Output: qualityScore + improvements
    | Files: ${output_base}/quality-report.md, ${output_base}/improvement-suggestions.json
    |
[Run Closure]  (see run-mode.md)
    | maestro run check {run_id} → repair gates → maestro session done {run_id}
    |
Return summary to user

TodoWrite Pattern

Core Concept: Dynamic task attachment and collapse for real-time visibility.

Key Principles

1. Task Attachment (when phase executed):

   • Sub-tasks are attached to orchestrator's TodoWrite
   • Phase 1, 2, 3, 4, 5: Multiple sub-tasks attached

2. Task Collapse (after sub-tasks complete):

   • Applies to Phase 1, 2, 3, 4, 5: Remove sub-tasks, collapse to summary
   • Maintains clean orchestrator-level view

3. Continuous Execution: After completion, automatically proceed to next phase

Phase 1 (Tasks Attached):

[
  {"content": "Phase 1: Review Parsing & Classification", "status": "in_progress"},
  {"content": "  → Parse reviewer comments structure", "status": "in_progress"},
  {"content": "  → Classify comments using Agy CLI", "status": "pending"},
  {"content": "  → Extract sentiment and key concerns", "status": "pending"},
  {"content": "  → Generate review-analysis.json", "status": "pending"},
  {"content": "Phase 2: Multi-Perspective Discussion", "status": "pending"},
  {"content": "Phase 3: Strategy Formulation", "status": "pending"},
  {"content": "Phase 4: Rebuttal Writing", "status": "pending"},
  {"content": "Phase 5: Quality Validation", "status": "pending"},
  {"content": "Run Closure: check + complete", "status": "pending"}
]

Phase 1 (Collapsed):

[
  {"content": "Phase 1: Review Parsing & Classification", "status": "completed"},
  {"content": "Phase 2: Multi-Perspective Discussion", "status": "pending"},
  {"content": "Phase 3: Strategy Formulation", "status": "pending"},
  {"content": "Phase 4: Rebuttal Writing", "status": "pending"},
  {"content": "Phase 5: Quality Validation", "status": "pending"},
  {"content": "Run Closure: check + complete", "status": "pending"}
]

Post-Phase Updates

After each phase completes:

1. Phase 1 → Phase 2: Pass reviewAnalysis and commentCategories to discussion phase
2. Phase 2 → Phase 3: Pass discussionConsensus and strategicRecommendations to strategy formulation
3. Phase 3 → Phase 4: Pass strategyMatrix and evidenceMap to rebuttal writing
4. Phase 4 → Phase 5: Pass rebuttalDraft to quality validation
5. Phase 5 → Return: Present quality report and improvement suggestions to user

Error Handling

• Parsing Failure: If output parsing fails, retry once, then report error
• Validation Failure: Report which file/data is missing
• Command Failure: Keep phase in_progress, report error, do not proceed
• CLI Failure: If Agy CLI fails, fall back to direct analysis or report error
• Paper Not Found: If paper path invalid, proceed with review-only mode

Coordinator Checklist

Before Phase 1:

• TodoWrite initialized with all 5 phases
• User preferences collected (autoMode, paperSource, conferenceType)
• Review comments path validated
• Paper path validated (if provided)

Between Phases:

• Previous phase marked completed
• Current phase marked in_progress
• Output variables extracted and passed to next phase
• Sub-tasks collapsed to summary

After Phase 5:

• All phases marked completed
• Quality report generated
• Improvement suggestions presented
• Final rebuttal.md file written

Run Closure:

• maestro run check {run_id} clean (repair any reported gate)
• maestro session done {run_id} succeeded before reporting success

Run Closure

Runtime-owned protocol files (session.json, run.json, artifacts.json) MUST NOT be edited directly, and no second manifest/index is maintained. Artifact registration and handoff are derived by the runtime from {run_dir}/outputs/. See run-mode.md.

After Phase 5 completes:

1. maestro run check {run_id} — repair any blocking artifact or exit gate it reports.
2. Optionally write {run_dir}/report.md (verdict + summary of the rebuttal and quality score).
3. maestro session done {run_id}. Report success only once the Run is completed.

Related Commands

Prerequisites:

• /research-init - Initialize research project structure
• /zotero-review - Import and review literature

Follow-ups:

• /commit - Commit rebuttal document to version control
• /presentation - Prepare conference presentation after acceptance
• /poster - Generate academic poster

CLI Integration Details

This skill uses maestro delegate for enhanced analysis:

Phase 1 - Review Parsing:

maestro delegate "PURPOSE: Parse and classify reviewer comments by type (Major/Minor/Typo/Misunderstanding)
TASK: • Extract comment structure • Classify by severity • Identify sentiment
MODE: analysis
CONTEXT: @<review-file>
EXPECTED: JSON with classification results" --to agy --mode analysis

Phase 2 - Multi-Perspective Discussion: Uses team-ultra-analyze skill or custom discussion agent to simulate multiple perspectives.

Phase 3 - Strategy Formulation:

maestro delegate "PURPOSE: Search paper content for evidence supporting response strategies
TASK: • Locate relevant sections • Extract supporting data • Identify evidence gaps
MODE: analysis
CONTEXT: @<paper-file>
EXPECTED: Evidence map with file:line references" --to agy --mode analysis

Phase 5 - Quality Validation:

maestro delegate "PURPOSE: Validate rebuttal quality (completeness, professionalism, persuasiveness)
TASK: • Check all comments addressed • Assess tone • Evaluate evidence strength
MODE: analysis
CONTEXT: @<rebuttal-file>
EXPECTED: Quality report with improvement suggestions" --to agy --mode analysis

Conference Template System

Templates are loaded from (first match wins):

1. Custom: templates/{templateId}-template.md under the skill directory (user-provided)
2. Custom generic: templates/discussion.md under the skill directory (user-provided)
3. Built-in: Generic rebuttal template hardcoded in Phase 4 (always available, no files needed)

Template selection based on workflowPreferences.conferenceType:

• ML Conferences: NeurIPS/ICML/ICLR strategies (novelty, theory, experiments)
• CV Conferences: CVPR/ECCV/ICCV strategies (visual results, one-page limit)
• NLP Conferences: ACL/EMNLP strategies (linguistic appropriateness, ethics)
• Generic: Universal template for all venues