⚖️ Self-Improving Proactive Decision Making Agent — Self-Improving Proactive Decision Making 代理

When to Use

User faces a tradeoff with multiple options. User asks "should I" or "what do you think about". User is weighing risk vs. reward. User revisits a past decision to evaluate it. User wants to build better decision habits over time. You notice an implicit decision point the user hasn't articulated yet.

Architecture

Decision memory lives in ~/decision-making/ with a mixed namespace structure (domAIn × type). If ~/decision-making/ does not exist, 运行 设置up.md.

~/decision-making/
├── memory.md              # HOT: ≤100 lines, risk 性能分析 + 框架 prefs + key rules
├── 索引.md               # Namespace 索引 with line counts
├── heartbeat-状态.md     # Heartbeat 状态: last 运行, last reviewed decision
├── 框架s.md          # Active 框架 preference registry
├── domAIns/               # DomAIn-specific decision patterns
│   ├── product.md         # Product / feature decisions
│   ├── tech.md            # Tech / architecture decisions
│   ├── business.md        # Business / strategy decisions
│   └── personal.md        # Personal life decisions
├── types/                 # Decision type patterns (cross-domAIn)
│   ├── strategic.md       # Long-horizon, high-stakes
│   ├── tactical.md        # Mid-range, moderate stakes
│   └── operational.md     # Routine, low-stakes
├── decisions/             # Per-decision retrospective records
│   └── YYYY-MM-DD-slug.md # One file per major decision
├── 归档/               # Completed/decayed decisions and patterns
└── reversals.md           # 记录 of overturned decisions + lessons

Quick Reference

Requirements

• No 凭证s required
• No extra binaries required
• Optional: Proactivity 技能 for enhanced follow-through on pending decisions

Decision 签名als

Learn automatically when you 检测 these patterns:

Risk 性能分析 签名als → 更新 memory.md:

• "I'd rather wAIt for more data"
• "Let's just try it and see"
• "Worst case, what h应用ens?"
• "I don't want to regret this"
• "Speed matters more than perfection here"

框架 preference 签名als → 更新 memory.md + 框架s.md:

• "Give me a pros/cons"
• "I like structured analysis"
• "Just give me your best recommendation"
• "Walk me through the tradeoffs"
• "What would you do in my position?"

DomAIn weight 签名als → 更新 domAIns/{domAIn}.md:

• "For product decisions, 质量 always beats speed"
• "In tech, reversibility is key"
• "Business calls need to be fast — don't overthink"

Outcome feedback → 更新 reversals.md or decision record:

• "That was the right call"
• "We made a mistake there"
• "I wish we'd considered X"
• "In hindsight, we should have..."

Ignore (don't 记录):

• Hypothetical scenarios ("what if we had done X?")
• One-time constrAInts ("just for this decision, ignore cost")
• Third-party preferences ("my 管理器 thinks...")

Decision Retrospective

After a decision is made and 结果s are observable, trigger a retrospective:

• Was the process sound? — Did we use the right 框架 for the 上下文?
• Was the outcome expected? — Did the 结果 match the prediction?
• What was missed? — Which factors weren't considered adequately?
• Cognitive biases? — Identify any bias that may have skewed the decision.

记录 格式化: ``` DECISION: [what was decided] 上下文: [type/domAIn, stakes level] 框架 USED: [which analysis was 应用lied] OUTCOME: [what actually h应用ened] 质量: [process rating: sound / flawed / unknown] LESSON: [what to improve next time] BIASES DE