What problem does it solve? AI agents repeatedly make the same mistakes because insights from failures, user corrections, and discovered best practices are never persisted. This Skill gives the agent a structured memory system that captures experience during work and promotes recurring lessons into permanent behavioral rules. ## Core Features & Use Cases - Seven trigger conditions: Automatically evaluates whether to log entries when commands fail, users correct the agent, capabilities are missing, APIs error, knowledge is outdated, better practices are found, or tasks complete. - Structured three-file system: Writes categorized entries with unique IDs into LEARNINGS.md, ERRORS.md, and FEATURE_REQUESTS.md under a .learnings/ directory, plus a machine-readable JSONL changelog. - Experience evolution with guardrails: Promotes patterns recurring three or more times into AGENTS.md, TOOLS.md, or SOUL.md, extracts reusable lessons into standalone skills, and applies ADL anti-drift and VFM value-first scoring to prevent low-value changes. - Use Case: After an agent hits a rate limit on the Semantic Scholar API, it logs the error, records the 3-second interval workaround as a learning, and on the third recurrence promotes the rule into TOOLS.md so all future sessions respect the limit. ## Quick Start Ask the agent to create a .learnings directory in the workspace and start recording errors, corrections, and learnings according to this skill's trigger rules.
