What problem does it solve? AI agents repeat the same mistakes across sessions because failures, user corrections, and discovered best practices are never recorded. This Skill creates a closed-loop feedback system that captures errors, learnings, and feature requests into structured Markdown logs, then promotes proven patterns into long-term memory. ## Core Features & Use Cases - Structured Logging: Records errors (ERR), learnings (LRN), and feature requests (FEAT) with unique IDs, priorities, statuses, and metadata into LEARNINGS.md, ERRORS.md, and FEATURE_REQUESTS.md files. - Scoped Storage: Writes to a skill-level data directory by default, with optional project-level (.learnings/) and public promotion areas for cross-project rules. - Promotion Workflow: The promote command copies mature entries to a public area, marks the source as promoted, and prevents duplicates; stable rules can then be distilled into a memory system. - Use Case: After a shell command fails with a non-obvious cause, run the auto_log.sh error command to record the failure and fix; weeks later, search the logs before a similar task to avoid repeating the diagnosis. ## Quick Start Ask the agent to log the current error or correction using the self-improving-agent skill, for example by running scripts/auto_log.sh with the error or learning subcommand.
