civaapple-alt/pi-cordis--packages-plugins-todo-tracker ↗★ 0

@pi-cordis/plugin-todo-tracker

Session task management and dynamic prompt injection plugin for Pi-Cordis 适合需管理多步任务清单的用户;状态按会话保存且更新具事务性。

パッケージ
@pi-cordis/plugin-todo-tracker
互換性
未検証
Cordis ピア範囲
^4.0.1
バージョン
0.1.0
最終更新
2026/08/21

インストール

検証済み bundle がないか、互換性チェックに失敗しています。先にリポジトリの説明を読んでください。 README 全文を読む ↗

ドキュメント

README 全文を読む ↗

@pi-cordis/plugin-todo-tracker

English | 中文

Native Cordis task tracking and adaptive prompt injection plugin. It provides todo_write and todo_read tools supporting a 4-state task machine (pending, in_progress, completed, cancelled), topological sort cycle validation (dependsOn), and adaptive prompt compression that collapses finished tasks to save token budget.

Todo state is keyed by Pi session and survives default/ptc Profile remounts within the current process. Updates are transactional: the plugin validates the complete candidate dependency graph before committing any field, so a rejected update cannot partially change a task.

Tools

todo_write

Accepts:

  • action ("add" | "update" | "clear", required): Action to perform.
  • id (string, optional): Task ID for update operations.
  • title (string, optional): Task title or summary.
  • status ("pending" | "in_progress" | "completed" | "cancelled", optional): Task status.
  • category (string, optional): Grouping category.
  • dependsOn (string[], optional): List of task IDs this task depends on (automatically validated for self-dependency and circular dependency loops).

todo_read

Returns:

  • total (number): Total tasks in list.
  • active (number): Active pending/in_progress count.
  • completed (number): Completed task count.
  • todos (array): Full list of task items.

Adaptive Prompt Injection

When injectToPrompt: true, active non-completed tasks and their blocker dependencies (blocked by: ...) are listed in the prompt, while completed/cancelled tasks are summarized into compact counts (✓ N completed hidden) to preserve conversation tokens.

Model Experience

  • Focus and Progress: Keeps the model oriented on remaining work and dependencies across long, multi-turn sessions.
  • Dependency Guard: Pre-validates acyclic dependency graphs to prevent task deadlocks.
  • Stable Profile Switching: Capability presentation can change without resetting the current session's task state.
  • Token Efficiency: Finished tasks do not consume repetitive prompt tokens.