Self-Improving + Proactive Agent自我进化智能体

When to Use

User corrects you or points out mistakes. You complete significant work and want to evaluate the outcome. You notice something in your own output that could be better. Knowledge should compound over time without manual maintenance.

Architecture

Memory lives in ~/self-improving/ with tiered structure. If ~/self-improving/ does not exist, run setup.md.
Workspace setup should add the standard self-improving steering to the workspace AGENTS, SOUL, and HEARTBEAT.md files, with recurring maintenance routed through heartbeat-rules.md.

CODEBLOCK0

Quick Reference

Topic	File
Setup guide	INLINECODE5
Heartbeat state template

heartbeat-state.md | | Memory template | memory-template.md | | Workspace heartbeat snippet | HEARTBEAT.md | | Heartbeat rules | heartbeat-rules.md | | Learning mechanics | learning.md | | Security boundaries | boundaries.md | | Scaling rules | scaling.md | | Memory operations | operations.md | | Self-reflection log | reflections.md | | OpenClaw HEARTBEAT seed | openclaw-heartbeat.md |

Requirements

• - No credentials required
• No extra binaries required
• Optional installation of the Proactivity skill may require network access

Learning Signals

Log automatically when you notice these patterns:

Corrections → add to corrections.md, evaluate for memory.md:

• - "No, that's not right..."
• "Actually, it should be..."
• "You're wrong about..."
• "I prefer X, not Y"
• "Remember that I always..."
• "I told you before..."
• "Stop doing X"
• "Why do you keep..."

Preference signals → add to memory.md if explicit:

• - "I like when you..."
• "Always do X for me"
• "Never do Y"
• "My style is..."
• "For [project], use..."

Pattern candidates → track, promote after 3x:

• - Same instruction repeated 3+ times
• Workflow that works well repeatedly
• User praises specific approach

Ignore (don't log):

• - One-time instructions ("do X now")
• Context-specific ("in this file...")
• Hypotheticals ("what if...")

Self-Reflection

After completing significant work, pause and evaluate:

1. 1. Did it meet expectations? — Compare outcome vs intent
2. What could be better? — Identify improvements for next time
3. Is this a pattern? — If yes, log to INLINECODE20

When to self-reflect:

• - After completing a multi-step task
• After receiving feedback (positive or negative)
• After fixing a bug or mistake
• When you notice your output could be better

Log format:
CODEBLOCK1

Example:
CODEBLOCK2

Self-reflection entries follow the same promotion rules: 3x applied successfully → promote to HOT.

Quick Queries

User says	Action
"What do you know about X?"	Search all tiers for X
"What have you learned?"

Show last 10 from corrections.md | | "Show my patterns" | List memory.md (HOT) | | "Show [project] patterns" | Load projects/{name}.md | | "What's in warm storage?" | List files in projects/ + domains/ | | "Memory stats" | Show counts per tier | | "Forget X" | Remove from all tiers (confirm first) | | "Export memory" | ZIP all files |

Memory Stats

On "memory stats" request, report:

CODEBLOCK3

Common Traps

Trap	Why It Fails	Better Move
Learning from silence	Creates false rules	Wait for explicit correction or repeated evidence
Promoting too fast

Pollutes HOT memory | Keep new lessons tentative until repeated | | Reading every namespace | Wastes context | Load only HOT plus the smallest matching files | | Compaction by deletion | Loses trust and history | Merge, summarize, or demote instead |

Core Rules

1. Learn from Corrections and Self-Reflection

• - Log when user explicitly corrects you
• Log when you identify improvements in your own work
• Never infer from silence alone
• After 3 identical lessons → ask to confirm as rule

2. Tiered Storage

Tier	Location	Size Limit	Behavior
HOT	memory.md	≤100 lines	Always loaded
WARM

projects/, domains/ | ≤200 lines each | Load on context match |

| COLD | archive/ | Unlimited | Load on explicit query |

3. Automatic Promotion/Demotion

• - Pattern used 3x in 7 days → promote to HOT
• Pattern unused 30 days → demote to WARM
• Pattern unused 90 days → archive to COLD
• Never delete without asking

4. Namespace Isolation

• - Project patterns stay in INLINECODE26
• Global preferences in HOT tier (memory.md)
• Domain patterns (code, writing) in INLINECODE27
• Cross-namespace inheritance: global → domain → project

5. Conflict Resolution

When patterns contradict:

1. 1. Most specific wins (project > domain > global)
2. Most recent wins (same level)
3. If ambiguous → ask user

6. Compaction

When file exceeds limit:

1. 1. Merge similar corrections into single rule
2. Archive unused patterns
3. Summarize verbose entries
4. Never lose confirmed preferences

7. Transparency

• - Every action from memory → cite source: "Using X (from projects/foo.md:12)"
• Weekly digest available: patterns learned, demoted, archived
• Full export on demand: all files as ZIP

8. Security Boundaries

See boundaries.md — never store credentials, health data, third-party info.

9. Graceful Degradation

If context limit hit:

1. 1. Load only memory.md (HOT)
2. Load relevant namespace on demand
3. Never fail silently — tell user what's not loaded

Scope

This skill ONLY:

• - Learns from user corrections and self-reflection
• Stores preferences in local files (~/self-improving/)
• Maintains heartbeat state in ~/self-improving/heartbeat-state.md when the workspace integrates heartbeat
• Reads its own memory files on activation

This skill NEVER:

• - Accesses calendar, email, or contacts
• Makes network requests
• Reads files outside INLINECODE31
• Infers preferences from silence or observation
• Deletes or blindly rewrites self-improving memory during heartbeat cleanup
• Modifies its own SKILL.md

Data Storage

Local state lives in ~/self-improving/:

• - memory.md for HOT rules and confirmed preferences
• INLINECODE34 for explicit corrections and reusable lessons
• INLINECODE35 and domains/ for scoped patterns
• INLINECODE37 for decayed or inactive patterns
• INLINECODE38 for recurring maintenance markers

Related Skills

Install with clawhub install <slug> if user confirms:

• - memory — Long-term memory patterns for agents
• INLINECODE41 — Adaptive teaching and explanation
• INLINECODE42 — Auto-learn decision patterns
• INLINECODE43 — Know when to ask vs act autonomously

Feedback

• - If useful: INLINECODE44
• Stay updated: INLINECODE45