Self Improving Agent 1.0.0

Captures learnings, errors, and corrections to enable continuous improvement. Use when: (1) A command or operation fails unexpectedly, (2) User corrects Claude ('No, that's wrong...', 'Actually...'), (3) User requests a capability that doesn't exist, (4) An external API or tool fails, (5) Claude realizes its knowledge is outdated or incorrect, (6) A better approach is discovered for a recurring task. Also review learnings before major tasks.

This skill is coherent and appears to do what it says: log learnings/errors to .learnings/*. Before installing, consider these operational safeguards: (1) Decide whether the agent should write directly to the repo or only prepare entries for human review — require manual approval before committing/promoting entries. (2) Add .learnings/* to .gitignore or otherwise ensure sensitive logs aren't accidentally committed to VCS. (3) Add an explicit redaction step to the skill (or your agent workflow) so environment details, stack traces, or pasted inputs are scrubbed for secrets (API keys, passwords, tokens, PII) before being saved. (4) Limit who or what can invoke this skill if you don't want autonomous edits. (5) If you prefer centralized, auditable storage for learnings, adapt the workflow to send sanitized entries to a secure logging store rather than raw files. These mitigations will preserve the skill's usefulness while reducing accidental leakage or unwanted repo modification.

Type: OpenClaw Skill Name: self-improving-agent-1-0-0 Version: 1.0.0 The skill is designed for agent self-improvement by logging learnings and errors. While its stated purpose is benign, it instructs the agent to modify core project memory files like `CLAUDE.md` and `AGENTS.md` (as described in SKILL.md under 'Promoting to Project Memory'). This capability, although intended for adding 'rules' and 'facts', allows the agent to alter its own future instructions and knowledge base. This presents a significant risk for persistent prompt injection or self-modification if the agent were to process a malicious 'learning' or be compromised, classifying it as suspicious due to a risky capability without explicit malicious intent within the provided instructions.