Self-Improving with Reflection

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

This skill is internally coherent and appears to do what it advertises: keep a local, tiered memory in ./self-improving/ and use it to learn from corrections and reflections. Before enabling it in a real workspace: 1) Review and sandbox the ./self-improving/ directory after initial setup to ensure nothing sensitive was logged; 2) Review diffs when the skill suggests edits to AGENTS.md, SOUL.md, or other workspace files; 3) Confirm the 'Never Store' boundaries are enforced in practice (inspect memory.md/corrections.md) and periodically run the provided 'forget everything'/export flows; 4) Avoid enabling on shared/team or production hosts unless you trust the storage location and audit changes; 5) If you prefer tighter control, run in Passive mode or require explicit confirmation before promoting patterns. These steps will reduce risk from accidental persistence or inappropriate storage of sensitive data.

Type: OpenClaw Skill Name: self-improving-with-reflection Version: 1.2.11 The 'Self-Improving with Reflection' skill is a well-documented framework for enabling AI agents to learn from user feedback and self-evaluation using a local, tiered file-based memory system. It includes explicit security boundaries in 'boundaries.md' that prohibit the storage of sensitive data like credentials, PII, or financial information. The skill's operations are confined to the './self-improving/' directory and standard workspace configuration files (e.g., AGENTS.md, SOUL.md), and it lacks any indicators of data exfiltration, malicious execution, or unauthorized network access.