Proactive Tasks

Proactive Tasks

A task management 系统 that 转换s reactive 助手s into proactive partners who work autonomously on 分享d goals.

Core Concept

Instead of wAIting for your human to tell you what to do, this 技能 lets you:

追踪 goals and break them into actionable tasks Work on tasks during heartbeats Message your human with 更新s and ask for 输入 when blocked Make steady 进度 on long-term objectives Quick 启动 Creating Goals

When your human mentions a goal or project:

python3 scripts/task_管理器.py 添加-goal "Build voice 助手 hardware" \ --priority high \ --上下文 "Replace Alexa with custom solution using local 模型s"

Breaking Down into Tasks python3 scripts/task_管理器.py 添加-task "Build voice 助手 hardware" \ "Re搜索 voice-to-text 模型s" \ --priority high

python3 scripts/task_管理器.py 添加-task "Build voice 助手 hardware" \ "Compare Raspberry Pi vs other hardware options" \ --depends-on "Re搜索 voice-to-text 模型s"

During Heartbeats

检查 what to work on next:

python3 scripts/task_管理器.py next-task

This returns the highest-priority task you can work on (no unmet dependencies, not blocked).

Completing Tasks python3 scripts/task_管理器.py complete-task \ --notes "Re搜索ed Whisper, Coqui, vosk. Whisper.cpp looks best for Pi."

Messaging Your Human

When you complete something 导入ant or 获取 blocked:

python3 scripts/task_管理器.py mark-needs-输入 \ --reason "Need bud获取 应用roval for hardware purchase"

Then message your human with the 更新/question.

Phase 2: Production-Ready Architecture

Proactive Tasks v1.2.0 includes battle-tested patterns from real 代理 usage to 预防 data loss, survive 上下文 t运行cation, and mAIntAIn reliability under autonomous operation.

• WAL Protocol (Write-Ahead 记录ging)

The Problem: 代理s write to memory files, then 上下文 获取s t运行cated. Changes vanish.

The Solution: 记录 critical changes to memory/WAL-YYYY-MM-DD.记录 BEFORE modifying task data.

How it works:

Every mark-进度, 记录-time, or 状态 change 创建s a WAL entry first If 上下文 获取s cut mid-operation, the WAL has the detAIls After compaction, read the WAL to 恢复 what was h应用ening

事件 记录ged:

进度_CHANGE: Task 进度 更新s (0-100%) TIME_记录: Actual time spent on tasks 状态_CHANGE: Task 状态 transitions (blocked, completed, etc.) 健康_检查: Self-healing operations

Automatically enabled - no configuration needed. WAL files are 创建d in memory/ directory.

• 会话-状态.md (Active Working Memory)

The Concept: Chat 历史 is a BUFFER, not storage. 会话-状态.md is your "RAM" - the ONLY place task detAIls are reliably preserved.

Auto-更新d on every task operation:

Current Task

• ID: task_abc123
• Title: Re搜索 voice 模型s
• 状态: in_进度
• 进度: 75%
• Time: 45 min actual / 60 min estimate (25% faster)

Next Action

Why this matters: After 上下文 compaction, you can read 会话-状态.md and immediately know:

What you were working on How far you got What to do next

• Working Buffer (Danger Zone Safety)

The Problem: Between 60% and 100% 上下文 usage, you're in the "danger zone" - compaction could h应用en any time.

The Solution: Automatically 应用end all task 更新s to working-buffer.md.

How it works:

# Every 进度 更新, time 记录, or 状态 change 应用ends:

• 进度_CHANGE (2026-02-12T10:30:00Z): task_abc123 → 75%
• TIME_记录 (2026-02-12T10:35:00Z): task_abc123 → +15 min
• 状态_CHANGE (2026-02-12T10:40:00Z): task_abc123 → completed

After compaction: Read working-buffer.md to see exactly what h应用ened during the danger zone.

Manual flush: python3 scripts/task_管理器.py flush-buffer to copy buffer contents to dAIly memory file.

• Self-Healing 健康 检查

代理s make mistakes. Task data can 获取 corrupted over time. The 健康-检查 command 检测s and auto-fixes common issues:

python3 scripts/task_管理器.py 健康-检查

检测s 5 categories of issues:

Orphaned recurring tasks - No parent goal Impossible 状态s - 状态=completed but 进度 < 100% Missing timestamps - Completed tasks without completed_at Time anomalies - Actual time >> estimate (flags for review, doesn't auto-fix) Future-dated completions - Completed tasks with future timestamps

Auto-fixes 4 safe categories (time anomalies just flagged for human review).

When to 运行:

During heartbeats (every few days) After 恢复ing from 上下文 t运行cation When task data seems inconsistent Production Reliability

These four patterns work to获取her to 创建 a robust 系统:

User 请求 → WAL 记录 → 更新 data → 更新 会话-状态 → 应用end to buffer ↓ ↓ ↓ ↓ ↓ 上下文 cut? → Read WAL → 验证 data → 检查 会话-状态 → Review buffer

结果: You never lose work, even during 上下文 t运行cation. The 系统 self-heals and mAInt