Self improving

Self-reflection + Self-criticism + Self-learning + Self-organizing memory. Agent evaluates its own work, catches mistakes, and improves permanently. Use when...

This skill is internally consistent and does what it says: it will create ~/self-improving/, log corrections/preferences, and update workspace steering files to make the agent behave more proactive. Before installing: (1) review and back up your workspace (AGENTS.md, SOUL.md, HEARTBEAT.md) because the skill suggests non‑destructive edits there; (2) understand that the agent will persistently read/write files in your home directory and load memory.md on sessions — avoid putting secrets or sensitive personal data into these files; (3) the only network action is the optional 'clawhub install proactivity' which will run only if you explicitly agree — decline it if you don't want additional network installs; (4) test in a disposable workspace or VM if you want to verify behavior first; and (5) if you ever ask the agent to 'forget everything', verify the exported backup before accepting deletion. Overall the footprint is proportionate to the stated purpose, but treat the created local memory as data you own and audit.

Type: OpenClaw Skill Name: self-improving-100 Version: 1.0.0 The 'Self-Improving + Proactive Agent' skill bundle is designed to create a persistent, tiered memory system (HOT/WARM/COLD) for an AI agent to learn from user corrections and self-reflection. It includes a robust security framework in 'boundaries.md' that explicitly forbids the storage of sensitive information such as credentials, PII, or financial data. While the 'setup.md' file instructs the agent to modify its core steering files (SOUL.md, AGENTS.md, HEARTBEAT.md) to ensure persistence of the learning logic, these actions are transparently documented and aligned with the stated purpose of the skill. The bundle demonstrates good security practices by requiring user consent before installing external dependencies (the 'proactivity' skill) and lacks any indicators of data exfiltration or unauthorized execution.