research-planning

name	research-planning
description	Generate a fine-grained research execution plan with task tracking dashboard. Use when user says "制定计划", "research plan", "make plan", or after idea validation is complete.
argument-hint	["idea01.md reference or project description"]
allowed-tools	Bash(*), Read, Write, Edit, Grep, Glob, WebSearch, WebFetch, Agent

Skill-3: Research Planning

Generate execution plan for: $ARGUMENTS

Overview

This skill reads idea01.md and produces a two-stage execution plan:

1. Stage 1 (High-Level Plan): Based on idea01.md alone — defines the overall research phases, task structure, and dashboard. Ontology design (skill-4) is included as a task within the plan.
2. Stage 2 (Plan Refinement): After skill-4 completes and produces ontology03.md + data_requirements03.md, the plan is automatically refined with detailed data collection specs, KG construction tasks, and concrete field-level requirements.

This two-stage approach allows planning to start immediately after idea validation, without waiting for ontology design.

Key features:

• Detailed task decomposition — every step explicitly defined
• SubAgent assignment — special tasks assigned to specific subagents
• Dependency graph — explicit blocked-by relationships
• Dashboard — live status panel at the top of plan04.md, auto-updated by downstream skills/subagents
• Project CLAUDE.md generation — after plan output, regenerate CLAUDE.md to reflect the actual project scope

Outputs:

• output/plans/plan04.md — full plan with dashboard
• Updated root CLAUDE.md — customized to the actual research project

Workflow

Phase 1: Read Inputs

Stage 1 (initial plan) requires only:

1. output/ideas/idea01.md — research question, innovation, methodology overview, evaluation plan

Stage 2 (plan refinement, triggered after skill-4 completes) also reads: 2. output/ontology/ontology03.md — entities, relationships, constraints, Graph RAG alignment 3. output/ontology/data_requirements03.md — data sources, formats, volumes, field specs

From idea01.md, extract:

• Overall research scope and phases
• What types of data will be needed (high-level)
• What KG approach is planned
• What Graph RAG strategy to build
• What evaluation to run
• What paper sections to write

Phase 2: Task Decomposition

Break the entire research project into tasks organized by phases. Each task must be atomic enough that one skill or subagent can complete it in a focused session.

Task Status System

Symbol	Status	Meaning
[ ]	Pending	Not started, prerequisites may not be met
[~]	In Progress	Currently being executed
[✓]	Completed	Done, output verified
[!]	Blocked	Cannot proceed, dependency unmet or issue found

Task Schema

Every task MUST include:

### Task X.Y: [Task Name]
- **Status**: [ ]
- **Assigned to**: [skill-N / SubAgent-N / Manual]
- **Input**: [specific files or prior task outputs]
- **Output**: [specific file path]
- **Depends on**: [Task IDs that must be ✓ before this can start]
- **Description**: [what exactly needs to be done]
- **Acceptance criteria**: [how to verify completion]

Required Task Phases

Decompose into these phases (adjust task count based on actual project):

Phase O: Ontology Design (assigned to skill-4)

This phase is part of the plan. After it completes, the plan auto-refines Phase A tasks with concrete data specs from data_requirements03.md.

Task	Name	Assigned	Depends On
O.1	Read references & survey existing ontologies	skill-4	—
O.2	Generate competency questions & user approval	skill-4	O.1
O.3	Entity, relationship, taxonomy design	skill-4	O.2
O.4	Constraint rules & Graph RAG alignment check	skill-4	O.3
O.5	Data requirements document	skill-4	O.3
O.6	Output ontology03.md + data_requirements03.md	skill-4	O.4, O.5
O.7	Plan refinement — update Phase A tasks with data_requirements03.md specs	skill-3	O.6

Phase A: Data Collection & Preprocessing (assigned to SubAgent1)

Phase A tasks are initially high-level (based on idea01.md). After Phase O completes (task O.7), tasks A.8-A.9 are refined with concrete field specs from data_requirements03.md.

Task	Name	Assigned	Depends On
A.1	News data source identification & search	SubAgent1	O.6
A.2	News article crawling & raw text extraction	SubAgent1	A.1
A.3	Academic paper search (Semantic Scholar, arXiv)	SubAgent1	O.6
A.4	Paper metadata & abstract collection	SubAgent1	A.3
A.5	Paper full-text / key section extraction	SubAgent1	A.4
A.6	Open data download & format conversion	SubAgent1	O.6
A.7	Data cleaning & deduplication	SubAgent1	A.2, A.5, A.6
A.8	Data formatting per data_requirements03.md specs	SubAgent1	A.7, O.7
A.9	Data quality validation & report	SubAgent1	A.8

Phase B: Knowledge Extraction & KG Construction (assigned to SubAgent2)

Task	Name	Assigned	Depends On
B.1	LLM extraction prompt design per ontology03.md	SubAgent2	A.9, O.6
B.2	Entity extraction from news/report data	SubAgent2	B.1
B.3	Entity extraction from paper data	SubAgent2	B.1
B.4	Relationship extraction & chain validation	SubAgent2	B.2, B.3
B.5	Taxonomy node initialization	SubAgent2	—
B.6	Graph database setup (Neo4j)	SubAgent2	—
B.7	Entity & relationship import to graph	SubAgent2	B.4, B.5, B.6
B.8	Constraint validation & orphan cleanup	SubAgent2	B.7
B.9	KG statistics & quality report	SubAgent2	B.8

Phase C: Graph RAG Implementation (assigned to SubAgent2)

Task	Name	Assigned	Depends On
C.1	Graph retrieval strategy implementation	SubAgent2	B.8
C.2	Vector store construction (for hybrid retrieval)	SubAgent2	B.8
C.3	Query-to-graph mapping logic	SubAgent2	C.1
C.4	LLM generation pipeline with graph context	SubAgent2	C.1, C.2
C.5	Graph RAG integration testing	SubAgent2	C.3, C.4

Phase D: Chatbot / Q&A System (assigned to SubAgent2)

Task	Name	Assigned	Depends On
D.1	Chatbot interface design	SubAgent2	C.5
D.2	Query handling & response formatting	SubAgent2	C.5
D.3	Chatbot integration with Graph RAG backend	SubAgent2	D.1, D.2
D.4	End-to-end chatbot testing	SubAgent2	D.3

Phase E: Evaluation & Validation (assigned to SubAgent3)

Task	Name	Assigned	Depends On
E.1	Evaluation metric selection & implementation	SubAgent3	C.5
E.2	Test dataset / question set preparation	SubAgent3	C.5
E.3	Knowledge extraction accuracy evaluation (ROUGE/METEOR)	SubAgent3	E.1, E.2
E.4	Graph RAG retrieval quality evaluation	SubAgent3	E.1, E.2
E.5	Baseline comparison experiments (e.g., different LLMs)	SubAgent3	E.3, E.4
E.6	Results table & figure generation	SubAgent3	E.5
E.7	Evaluation report writing	SubAgent3	E.6

Phase F: Methodology, Results & Reproducibility Documentation (assigned to SubAgent2)

Task	Name	Assigned	Depends On
F.1	Methodology framework writing (must mirror actual code logic)	SubAgent2	C.5, D.4
F.2	System architecture diagram	SubAgent2	F.1
F.3	Result report writing (KG stats, RAG performance, case studies)	SubAgent2	B.9, C.5, D.4
F.4	Reproducibility manual (manual_evaluation.md)	SubAgent2	F.1, F.3, E.7

Phase G: Literature Review (assigned to skill-5)

Task	Name	Assigned	Depends On
G.1	Review framework finalization	skill-5	—
G.2	Paper collection (~20 papers, last 5 years)	skill-5	G.1
G.3	Literature review writing with citations	skill-5	G.2

Phase H: Paper Writing (assigned to skill-6, skill-7)

Task	Name	Assigned	Depends On
H.1	Introduction writing	skill-6	G.3, F.1
H.2	Literature review integration	skill-6	G.3
H.3	Methodology section integration	skill-6	F.1, F.2
H.4	Results section integration	skill-6	F.3, E.7
H.5	Discussion writing	skill-6	H.3, H.4
H.6	Conclusion writing	skill-6	H.5
H.7	Full paper assembly & formatting	skill-6	H.1-H.6
H.8	Figure generation	skill-7	H.7, all figure_prompts
H.9	Export to .docx	skill-8	H.7, H.8

Note: This is a template. Actual tasks MUST be customized based on the specific idea and data requirements. Tasks may be added later (e.g., paper figure/diagram creation) — update this SKILL.md when that happens.

Phase 3: Dependency Graph

Generate a Mermaid dependency graph showing all tasks and their relationships:

graph LR
    O.1 --> O.2 --> O.3 --> O.4 --> O.6
    O.3 --> O.5 --> O.6 --> O.7
    O.6 --> A.1
    O.6 --> A.3
    O.6 --> A.6
    A.1 --> A.2 --> A.7
    A.3 --> A.4 --> A.5 --> A.7
    A.6 --> A.7 --> A.8
    O.7 --> A.8
    A.8 --> A.9
    A.9 --> B.1
    O.6 --> B.1
    B.1 --> B.2 --> B.4
    B.1 --> B.3 --> B.4
    B.4 --> B.7
    B.5 --> B.7
    B.6 --> B.7
    B.7 --> B.8 --> B.9
    B.8 --> C.1 --> C.3 --> C.5
    B.8 --> C.2 --> C.4 --> C.5
    C.5 --> D.1 --> D.3 --> D.4
    C.5 --> D.2 --> D.3
    C.5 --> E.1 --> E.3 --> E.5 --> E.6 --> E.7
    C.5 --> E.2 --> E.3
    E.2 --> E.4 --> E.5
    C.5 --> F.1 --> F.2
    D.4 --> F.1
    B.9 --> F.3
    D.4 --> F.3
    F.1 --> F.4
    F.3 --> F.4
    E.7 --> F.4
    F.1 --> H.1
    G.1 --> G.2 --> G.3
    G.3 --> H.1
    G.3 --> H.2
    F.1 --> H.3
    F.2 --> H.3
    F.3 --> H.4
    E.7 --> H.4
    H.3 --> H.5
    H.4 --> H.5
    H.5 --> H.6
    H.1 --> H.7
    H.2 --> H.7
    H.3 --> H.7
    H.4 --> H.7
    H.5 --> H.7
    H.6 --> H.7
    H.7 --> H.8 --> H.9

Phase 4: Build Dashboard

The dashboard sits at the very top of plan04.md and shows real-time progress.

# ═══════════════════════════════════════════════════
#  PROJECT DASHBOARD — Auto-updated by skills/agents
# ═══════════════════════════════════════════════════

**Project**: [Title from idea01.md]
**Last Updated**: [date + time]
**Overall Progress**: [X/N tasks completed] ([XX%])

## Phase Summary

| Phase | Name | Tasks | Done | Status |
|-------|------|-------|------|--------|
| O | Ontology Design | 7 | 0 | [ ] |
| A | Data Collection & Preprocessing | 9 | 0 | [ ] |
| B | KG Construction | 9 | 0 | [ ] |
| C | Graph RAG Implementation | 5 | 0 | [ ] |
| D | Chatbot / Q&A System | 4 | 0 | [ ] |
| E | Evaluation & Validation | 7 | 0 | [ ] |
| F | Methodology, Results & Reproducibility | 4 | 0 | [ ] |
| G | Literature Review | 3 | 0 | [ ] |
| H | Paper Writing & Export | 9 | 0 | [ ] |

## Blocked Tasks
| Task | Blocked By | Issue |
|------|-----------|-------|
| (none) | | |

## Recently Completed
| Task | Completed At | Output |
|------|-------------|--------|
| (none) | | |

# ═══════════════════════════════════════════════════

Phase 5: Present to User & Refine

Present the full plan:

• Dashboard
• All tasks with assignments, dependencies, acceptance criteria
• Dependency graph
• Identify parallelizable work (e.g., Phase G can run alongside A-E)

Wait for user approval. Adjust tasks as requested.

Phase 6: Output — Write plan04.md

After approval, write output/plans/plan04.md with:

1. Dashboard (top)
2. Project metadata
3. All task definitions (full schema per task)
4. Dependency graph (Mermaid)
5. SubAgent assignment summary
6. Auto-update instructions

Phase 7: Generate Project-Specific CLAUDE.md

After plan04.md is written, regenerate the root CLAUDE.md from the initial template to create a project-customized version.

The current CLAUDE.md is a generic framework template. Phase 7 replaces it with a version that contains the actual project details while preserving all framework rules.

Based on idea01.md + plan04.md, the new CLAUDE.md MUST include ALL of the following sections:

Section 1: Project Overview (customized)

• Replace generic text with actual research title, research question, target domain
• Keep Graph RAG focus statement
• Add target journal (from idea01.md)

Section 2: Project Structure (keep as-is)

• File tree stays the same — it's structural, not project-specific

Section 3: Core Rules (keep all 14 rules + add project-specific ones)

• All existing rules (1-14) preserved
• Add any project-specific rules derived from the idea (e.g., "All entities must follow the [domain] ontology defined in ontology03.md")

Section 4: Plan Auto-Update Protocol (keep as-is)

• Unchanged — this is a universal protocol

Section 5: Skill Execution Order (customized)

• Replace generic flow with actual execution order from plan04.md
• Show which phases are parallelizable for this specific project

Section 6: All 8 Skills — Full Description (MUST include all)

For each skill (1-8), write:

• Skill name and slash command
• What it does (1-2 sentences, project-specific)
• Input — actual file names from this project
• Output — actual file paths
• Key tools/references used
• Project-specific notes (e.g., for skill-4: "Ontology covers [domain] with [N] entity types")

Section 7: All 3 SubAgents — Full Description (MUST include all)

For each SubAgent, write:

• Agent name and @-mention command
• What it does (project-specific)
• Input/Output — actual paths
• Assigned plan phases
• Key configuration needed (Neo4j, API keys, etc.)

Section 8: External Tools & Services (customized)

• List ALL external tools this project needs with configuration status
• Codex MCP, Zotero MCP, Neo4j, Gemini API, LLM providers, etc.

Section 9: Output File Naming Convention (keep + extend)

• Keep existing naming table
• Add any project-specific output files

Section 10: Project-Specific Context (NEW)

• Domain description: What this project's domain is about (from idea01.md)
• Research question: The specific research question being addressed
• Innovation points: 2-3 key innovations (from idea01.md)
• Target audience: Who will use the chatbot / read the paper
• Data sources summary: What data will be collected (high-level from idea01.md)

This section gives Claude full context about the project in every future conversation.

Section 11: Important Notes (keep + extend)

Present the complete CLAUDE.md to user for approval before writing.

Why this matters: CLAUDE.md is read by Claude at the start of every conversation. If it only contains generic framework info, Claude won't understand the specific project. The customized version ensures Claude always has the research context.

Auto-Update Protocol

This is the protocol that ALL downstream skills and subagents must follow after completing a task:

AUTO-UPDATE PROTOCOL FOR plan04.md
===================================
After completing any task listed in plan04.md:

1. READ plan04.md
2. FIND the completed task by ID (e.g., "Task A.1")
3. UPDATE the task:
   - Change status: [ ] or [~] → [✓]
   - Add completion line: "**Completed**: [date] | Output: [file path]"
4. UPDATE the dashboard:
   - Increment "Done" count for the task's phase
   - Recalculate overall progress percentage
   - If all tasks in a phase are ✓, mark phase status as [✓]
5. CHECK downstream tasks:
   - For each task that depends on the completed task:
     - If ALL its dependencies are now [✓]: change from [!] to [ ] (unblocked)
6. UPDATE "Recently Completed" table in dashboard
7. WRITE the modified plan04.md

This protocol must be embedded in the CLAUDE.md generated in Phase 7.

Key Rules

1. Every task must be atomic: One clear action, one clear output, one clear assignee.
2. Dependencies are mandatory: No task without explicit Depends on field.
3. SubAgent assignment is explicit: Tasks for SubAgent1/2/3 must be clearly marked.
4. Dashboard is live: Updated after every task completion, not manually maintained.
5. User approval twice: After plan presentation (Phase 5) and after CLAUDE.md regeneration (Phase 7).
6. plan04.md is the single source of truth: All progress tracking lives here.
7. Parallelism encouraged: Identify tasks that can run concurrently (e.g., literature review alongside data collection).
8. Customization required: The task template is a starting point — actual tasks MUST reflect the specific project.
9. Extensible: New tasks (e.g., paper figures) can be added later. Update this SKILL.md accordingly.

Composing with Other Skills

/idea-checking            → idea01.md
/research-planning        → plan04.md (high-level) + CLAUDE.md  ← YOU ARE HERE
/ontology-design          → executes Phase O tasks → ontology03.md + data_requirements03.md
                            → triggers plan refinement (task O.7)
SubAgent1                 → executes Phase A tasks (input: data_requirements03.md)
SubAgent2                 → executes Phase B, C, D, F tasks (input: ontology03.md + data)
SubAgent3                 → executes Phase E tasks (input: KG/RAG outputs)
/review-writing           → executes Phase G tasks (can run in parallel with A-E)
/paper-writing            → executes Phase H.1-H.7 tasks
/figure-generation        → executes Phase H.8 task
/paper-doc                → executes Phase H.9 task