スキル一覧に戻る
jiamingkong

planner

by jiamingkong

A minimalist python implementation of CLI coding agents, using interleaved thinking (think with tools) and dynamic skill loading. The whole core mechanism implemented in two Python files.

4🍴 2📅 2026年1月20日
GitHubで見るManusで実行

SKILL.md


name: planner description: File-based planning system. allowed-tools:

  • init_planning
  • resume_last_run
  • read_plan
  • update_plan_status
  • mark_step_complete
  • add_finding
  • erase_plans

Planner Skill

This skill implements a persistent planning pattern using markdown files.

Files Managed

  • task_plan.md: Tracks phases, progress, and high-level goal.
  • findings.md: Records research, technical decisions, and resources.
  • progress.md: Detailed session log of actions and thoughts.

Tools

init_planning

Initialize the planning files in the current working directory. For coding task, you do not have to provide rigorous testing.

  • task_description: What needs to be done.
  • phases: List of high-level phases.
  • phase_steps: REQUIRED. List of lists of strings. Each inner list provides the detailed checklist steps for the corresponding phase.
    • Example: [["Research the official API documentation for authentication methods", "Read the developer guides on rate limiting strategies"], ["Set up the initial project structure using Python Poetry and git", "Implement the core logic for the connection manager class"]]

resume_last_run

Reads existing planning files to allow the agent to resume an interrupted task. Returns the content of task_plan.md and the recent log from progress.md.

read_plan

Read the current status from the planning files.

update_plan_status

Update the status of a phase in task_plan.md and log a note in progress.md.

  • phase_name: Name of the phase.
  • status: New status (pending, in_progress, completed).
  • notes: Optional note to append to progress log.

mark_step_complete

Mark a specific step in task_plan.md as complete ([x]).

  • phase_name: The name of the phase containing the step.
  • step_keyword: A unique substring to identify the step line.

add_finding

Record a finding, decision, or resource in findings.md.

  • category: Target section ("Requirements", "Research", "Technical Decisions", "Resources").
  • content: The text to add. For technical decisions, you can use markdown table row format | Decision | Rationale |.

erase_plans

Quickly erase existing plans in artifacts/cache by deleting the three planning files.

  • This is useful when you want to start fresh without any leftover plan files.
  • No parameters needed.

Best Practices for Agents

  1. Start Strong: Always begin complex tasks with init_planning. Break down phases into granular, actionable steps in phase_steps rather than broad goals.
  2. Update Frequently: As you complete sub-tasks, use mark_step_complete IMMEDIATELY. This acts as a "save point" and keeps your reasoning grounded.
  3. Log Discoveries: Don't rely on conversation history for long-term facts. Use add_finding to record important URLs, file paths, or architectural decisions.
  4. Transition Phases: When a phase is done, explicitly use update_plan_status to mark it as completed and the next one as in_progress.
  5. Resume Smart: If you see existing plan files, prefer resume_last_run over starting from scratch.

スコア

総合スコア

60/100

リポジトリの品質指標に基づく評価

SKILL.md

SKILL.mdファイルが含まれている

+20
LICENSE

ライセンスが設定されている

0/10
説明文

100文字以上の説明がある

+10
人気

GitHub Stars 100以上

0/15
最近の活動

3ヶ月以内に更新がある

0/10
フォーク

10回以上フォークされている

0/5
Issue管理

オープンIssueが50未満

+5
言語

プログラミング言語が設定されている

+5
タグ

1つ以上のタグが設定されている

0/5

レビュー

💬

レビュー機能は近日公開予定です