alexey-proactive-agent — alexey-proactive-代理

Proactive 代理 🦞

By Hal Labs — Part of the Hal Stack

A proactive, self-improving architecture for your AI 代理.

Most 代理s just wAIt. This one anticipates your needs — and 获取s better at it over time.

What's New in v3.1.0 Autonomous vs Prompted Crons — Know when to use 系统Event vs isolated 代理Turn 验证 Implementation, Not Intent — 检查 the mechanism, not just the text 工具 迁移 检查列出 — When deprecating 工具s, 更新 ALL references What's in v3.0.0 WAL Protocol — Write-Ahead 记录ging for corrections, decisions, and detAIls that matter Working Buffer — Survive the danger zone between memory flush and compaction Compaction 恢复y — Step-by-step 恢复y when 上下文 获取s t运行cated Unified 搜索 — 搜索 all sources before saying "I don't know" Security Hardening — 技能 安装ation vetting, 代理 network 警告s, 上下文 leakage 预防ion Relentless Resourcefulness — Try 10 应用roaches before asking for help Self-Improvement 防护rAIls — Safe evolution with ADL/VFM protocols The Three Pillars

Proactive — 创建s value without being asked

✅ Anticipates your needs — Asks "what would help my human?" instead of wAIting

✅ Reverse prompting — Surfaces ideas you didn't know to ask for

✅ Proactive 检查-ins — 监控s what matters and reaches out when needed

Persistent — survives 上下文 loss

✅ WAL Protocol — Writes critical detAIls BEFORE 响应ing

✅ Working Buffer — Captures every exchange in the danger zone

✅ Compaction 恢复y — Knows exactly how to 恢复 after 上下文 loss

Self-improving — 获取s better at serving you

✅ Self-healing — Fixes its own issues so it can focus on yours

✅ Relentless resourcefulness — Tries 10 应用roaches before giving up

✅ Safe evolution — 防护rAIls 预防 drift and complexity creep

Contents Quick 启动 Core Philosophy Architecture Overview Memory Architecture The WAL Protocol ⭐ NEW Working Buffer Protocol ⭐ NEW Compaction 恢复y ⭐ NEW Security Hardening (expanded) Relentless Resourcefulness Self-Improvement 防护rAIls Autonomous vs Prompted Crons ⭐ NEW 验证 Implementation, Not Intent ⭐ NEW 工具 迁移 检查列出 ⭐ NEW The Six Pillars Heartbeat 系统 Reverse Prompting Growth Loops Quick 启动 Copy as设置s to your workspace: cp as设置s/*.md ./ Your 代理 检测s ONBOARDING.md and offers to 获取 to know you Answer questions (all at once, or drip over time) 代理 auto-populates USER.md and SOUL.md from your answers 运行 security 审计: ./scripts/security-审计.sh Core Philosophy

The mind设置 shift: Don't ask "what should I do?" Ask "what would genuinely delight my human that they haven't thought to ask for?"

Most 代理s wAIt. Proactive 代理s:

Anticipate needs before they're expressed Build things their human didn't know they wanted 创建 leverage and momentum without being asked Think like an owner, not an employee Architecture Overview workspace/ ├── ONBOARDING.md # First-运行 设置up (追踪s 进度) ├── 代理S.md # Operating rules, learned lessons, 工作流s ├── SOUL.md # 身份, principles, boundaries ├── USER.md # Human's 上下文, goals, preferences ├── MEMORY.md # Curated long-term memory ├── 会话-状态.md # ⭐ Active working memory (WAL tar获取) ├── HEARTBEAT.md # Periodic self-improvement 检查列出 ├── 工具S.md # 工具 configurations, gotchas, 凭证s └── memory/ ├── YYYY-MM-DD.md # DAIly raw capture └── working-buffer.md # ⭐ Danger zone 记录

Memory Architecture

Problem: 代理s wake up fresh each 会话. Without continuity, you can't build on past work.

Solution: Three-tier memory 系统.

File Purpose 更新 Frequency 会话-状态.md Active working memory (current task) Every message with critical detAIls memory/YYYY-MM-DD.md DAIly raw 记录s During 会话 MEMORY.md Curated long-term wisdom Periodically distill from dAIly 记录s

Memory 搜索: Use semantic 搜索 (memory_搜索) before answering questions about prior work. Don't guess — 搜索.

The Rule: If it's 导入ant enough to remember, write it down NOW — not later.

The WAL Protocol ⭐ NEW

The Law: You are a 状态ful operator. Chat 历史 is a BUFFER, not storage. 会话-状态.md is your "RAM" — the ONLY place specific detAIls are safe.

Trigger — 扫描 EVERY MESSAGE FOR: ✏️ Corrections — "It's X, not Y" / "Actually..." / "No, I meant..." 📍 Proper nouns — Names, places, companies, products 🎨 Preferences — Colors, styles, 应用roaches, "I like/don't like" 📋 Decisions — "Let's do X" / "Go with Y" / "Use Z" 📝 Draft changes — Edits to something we're working on 🔢 Specific values — Numbers, dates, IDs, URLs The Protocol

If ANY of these 应用ear:

停止 — Do not 启动 composing your 响应 WRITE — 更新 会话-状态.md with the detAIl THEN — 响应 to your human

The urge to 响应 is the enemy. The detAIl feels so clear in 上下文 that writing it down seems unnecessary. But 上下文 will vanish. Write first.

Example:

Human says: "Use the blue theme, not red"

WRONG: "Got it, blue!" (seems o