Self-Improvement Log

Problem Addressed

Long-running coding agents can repeat the same mistakes: failed commands, user corrections, changed API behavior, and outdated knowledge. Without durable notes, the next session starts from scratch and the user has to re-explain context. This skill turns errors, corrections, feature requests, and best practices into searchable project logs instead of ephemeral chat history.

How It Works and Limits

• Initialisation: create .learnings/ at the workspace root with LEARNINGS.md, ERRORS.md, and FEATURE_REQUESTS.md, without overwriting existing files.
• Logging: append structured entries when a command fails, the user corrects the agent, knowledge gaps appear, or feature requests are raised, using IDs such as LRN-YYYYMMDD-XXX, ERR-..., and FEAT-..., plus priority, scope, and status.
• Reuse: search existing logs before creating duplicates, add See Also links, raise priority for recurring issues, and distill stable rules into CLAUDE.md, AGENTS.md, or .github/copilot-instructions.md when they are broadly applicable.

It is not a general article rewriter and does not automatically fix bugs. Its value depends on immediate, specific logging and promoting durable rules from temporary logs into long-lived project memory. Avoid recording secrets, tokens, full source files, or sensitive command output unless the user explicitly asks.