What problem does it solve? AI agent memory files (AGENTS.md / CLAUDE.md) drift out of date, grow bloated, or become disorganized, causing agents to ignore rules or repeat questions. This Skill keeps project memory accurate, concise (60-line cap), and tool-agnostic by extracting facts from project sources. ## Core Features & Use Cases - Smart detection and creation: Scans package manifests, README, git history, and code structure to create AGENTS.md on first run, or detects whether an existing file needs updating or restructuring. - Update mode (--update): Extracts tech stack, ADR summaries, domain terms, active tasks, and run commands; for keel projects it also syncs numbering state into .claude/rules/devdocs-state.md with health-baseline size checks. - Restructure mode (--restructure): Reorganizes an existing AGENTS.md against best-practice rules (line limits, tool-agnosticism, anti-pattern detection) while preserving valid user rules. - Use Case: After finishing a development sprint, run the update mode so AGENTS.md reflects the current tech stack, active tasks, and workflow routing section without manual editing. ## Quick Start Ask the agent to run /agent-memory to detect the current state of the project's AGENTS.md and create or update it from project sources.