XNews

claude-reflect: plugin to fix agent memory

claude-reflect: plugin to fix agent memory

claude-reflect is a Claude Code plugin that captures corrections, approvals, and 'remember:' statements during sessions. Running /reflect presents them for review before writing to CLAUDE.md, ensuring future sessions remember the corrections. It supports multiple agent frameworks and languages.

Key insights

Plugin captures corrections and writes them to persistent memory

The plugin hooks into sessions to capture corrections, approvals, and 'remember:' statements. Running /reflect presents a table of captured items for the user to accept, modify, or skip. Accepted items are written to CLAUDE.md, which is loaded in future sessions. It supports project-level, subdirectory, and AGENTS.md files, working with Codex and Cursor as well.

v2 adds skill suggestion from conversation history

Version 2 includes /reflect-skills, which reviews the past two weeks of session logs to identify frequently repeated requests (e.g., 'check my efficiency today') and suggests creating a command. It generates a draft for the user to approve. Corrections in Chinese are also recognized via AI semantic filtering.

Source posts