SuperAPI AI Directory

creative-plan

Writing & research

View repository

Updated 10-28 01:26

About this skill

Creative Project Planning Skill

Overview

This Skill provides a general-purpose hierarchical planning workflow for creative projects, supporting a variety of creative domains (novels, screenplays, games, videos, etc.). It dynamically loads planning guidelines according to the domain configuration and sets up human review gates at key decision points.

Core Principles

  • Structure First: Establish a solid macro structure first, then refine layer by layer
  • Human in the Loop: Request user review after each level is completed
  • Coherence: Ensure logical coherence between levels
  • Traceability: All planning decisions can be traced back to the configuration file
  • Domain Awareness: Use the corresponding planning strategy for different domains

Prerequisite Steps (Domain Loading)

  1. Use Read to read .creative/config.yaml to obtain the project configuration
    • If the configuration does not exist, prompt the user to run /creative-init
    • Extract the domain value from the configuration (for example: novel, screenplay, game)
  1. Calculate the PLUGIN_ROOT path

    • Obtain the current path from the Base directory in the Skill header
    • PLUGIN_ROOT = Base directory with the /skills/creative-plan part removed
    • For example:/Users/user/.claude/skill-cache/novel/skills/creative-plan → /Users/user/.claude/skill-cache/novel
  2. Set the DOMAIN_DIR path

    • DOMAIN_DIR = $PLUGIN_ROOT/shared/domains/{domain}/
    • For example: domain=novel → /Users/user/.claude/skill-cache/novel/shared/domains/novel/
  3. Use Read to read the domain configuration

    • Read $DOMAIN_DIR/domain.yaml to obtain the layers definition and path mapping
    • Read $DOMAIN_DIR/plan-guide.md to obtain the domain-specific planning guidelines
  4. Use Read to check .creative/status.json to understand the current progress

Workflow

Based on the layers defined in domain.yaml, execute the planning for each level in sequence:

Level Iteration Logic

For each layer defined in domain.yaml (L1, L2, L3...):

  1. Check prerequisites

    • If the level depends on a preceding level (for example, L2 depends on L1), check whether the preceding level has been completed
    • Read the current progress from status.json
  2. Read the domain guidelines

    • Read the specific planning rules for that level from plan-guide.md
    • Including input requirements, output format, and quality standards
  3. Collect input

    • If the level is the starting level (L1), use AskUserQuestion to collect the user's creative ideas
    • If it is a subsequent level, use Read to read the output of the preceding level as input
    • Based on the requirements in the guide, decide the specific questions to ask the user
  4. Execute planning

    • Generate the content for that level according to the rules in the guide
    • Ensure compliance with the output path format defined in domain.yaml
    • Apply domain-specific structural patterns (for example, the five-act structure of a novel)
  5. Batch generation strategy(applicable to fine-grained levels such as L2, L3)

    • Read the batch size recommendation from the guide (for example, groups of 10 units each)
    • Use Read to read the preceding content to ensure coherence
    • Generate one batch of units at a time (chapters, scenes, paragraphs, etc.)
  6. Write output

    • Determine the output path according to path_patterns in domain.yaml
    • Use Write to write the generated content
    • Follow the document format defined in the guide
  7. Quality check

    • Execute the quality check checklist for this level defined in the guide
    • Ensure the output meets standards such as word count requirements and structural completeness
  8. Human review gate

    • Use AskUserQuestion to request user review
    • According to the review question template defined in the guide
    • Wait for the user's reply: confirm / modify / replan
  9. Update status

    • Use Edit to update .creative/status.json
    • Record the completed level, batch, and last update time

Coherence assurance

When generating each unit (chapter, scene, etc.):

  1. Use Read to read the content of adjacent units (the 2-3 units before and after)
  2. Check according to the coherence rules in the guide:
    • Whether the content transitions smoothly
    • Whether elements (characters, settings, etc.) are used consistently
    • The planting and payoff of foreshadowing and clues

Knowledge base synchronization

If knowledge base management rules are defined in the guide:

  1. Identify new elements (new characters, new settings, etc.) during the generation process
  2. Write to the corresponding file according to the knowledge_base path in domain.yaml
  3. Use the knowledge base file format defined in the guide

Special command handling

Specified unit handling

When the user specifies a specific unit (for example /creative-plan unit-005):

  1. Parse the level and unit number specified by the user
  2. Use Read to read the preceding content of the level to which the unit belongs
  3. Use Read to read adjacent units to ensure coherence
  4. Only generate or update the specified unit

Replan a specific part

When the user requests to replan a certain part (for example /creative-plan section-3-revise):

  1. Use Read to read the current structure
  2. Use AskUserQuestion to ask about modification requirements
  3. Only modify the structure of the specified part
  4. Check the impact on other parts
  5. Mark affected units as needing regeneration

Batch generation mode

When the user requests batch generation (for example /creative-plan batch-generate 10):

  1. Read the current progress from status.json
  2. Generate the next N units (N is specified by the user or uses the guide's default value)
  3. Update the status after each unit is generated
  4. Request review collectively after the batch is completed

Quality check framework

Perform general checks after each level is completed:

Structural completeness

  • All required fields are filled in
  • References between levels are correct
  • Output paths conform to the specification

Content coherence

  • Transitions from preceding units are smooth
  • Elements are used consistently
  • No logical contradictions

Format specification

  • Conforms to the format defined in the guide
  • Document metadata is complete
  • Path naming is correct

Domain-specific checks

  • Execute the domain-specific check checklist defined in the guide

Status management

Use .creative/status.json to track planning progress:

{
  "project_name": "專案名稱",
  "domain": "領域類型",
  "status": {
    "layer1": {
      "completed": true,
      "last_updated": "ISO8601 時間戳",
      "approved": true
    },
    "layer2": {
      "total_units": 50,
      "completed_units": 20,
      "last_unit": 20,
      "last_updated": "ISO8601 時間戳",
      "approved_batches": [1, 2]
    }
  },
  "next_action": "下一步建議"
}

Error Handling

Configuration File Does Not Exist

If .creative/config.yaml does not exist, inform the user that they need to run /creative-init to initialize the project first.

Domain Configuration Missing

If domain.yaml or plan-guide.md does not exist, report an error and prompt to check the Skill installation.

Previous Level Not Completed

If the user skips the previous level and directly requests a subsequent level, prompt that the previous level must be completed first.

Incorrect Order

If the user requests generating unit N but unit N-1 has not been completed, ask whether to:

  • Complete the previous unit first
  • Skip generation (may affect coherence)

Best Practices

  1. Progressive Generation: Do not generate all units at once; give the user opportunities to review and adjust
  2. Maintain Consistency: Frequently reference the macro structure and previous units
  3. Document Completeness: Ensure every output file contains the necessary metadata
  4. User Communication: Use clear questions at key decision points
  5. Version Logging: Record generation time and configuration version in the files
  6. Domain Expertise: Strictly follow the domain-specific rules in the guide

Output Path Mapping

Dynamically determined based on the path_patterns in domain.yaml:

  • Configuration:.creative/config.yaml, .creative/status.json
  • Planning output: based on the path_pattern of each level
  • Knowledge base: based on the knowledge_base configuration
  • Status file:.creative/status.json

Example Interaction Flow

用戶: /creative-plan

Claude:
[讀取 .creative/config.yaml]
[載入 domain.yaml 和 plan-guide.md]
我將開始創意專案規劃。領域:小說,目前狀態:未開始

首先進行 L1 宏觀結構規劃。根據小說領域指引,我需要了解...
[顯示 guide 中定義的問題]

用戶: [回答創意想法]

Claude:
[根據 guide 生成 L1 結構]
[寫入對應路徑]
L1 宏觀結構已生成完畢,請審核...
[顯示 guide 中定義的審核問題]

用戶: 確認

Claude:
[開始 L2 規劃]
開始生成第 1-10 個單位...
[根據 guide 逐個生成]
第 1-10 單位生成完畢,請審核...

[繼續流程...]

Final Reminder: This Skill is a creative assistance tool; AI-generated plans are a starting point, not the endpoint. Encourage users to provide revision feedback at each review point so the plan better matches their personal creative vision. For detailed domain-specific rules, please refer to shared/domains/{domain}/plan-guide.md。