claude-reflect
A self-learning system for Claude Code that captures corrections and workflow patterns and turns them into CLAUDE.md memory and reusable skills.
At a Glance
claude-reflect plugin is released under the MIT License, an OSI-approved license, and is free to use, copy, modify and distribute.
Engagement
Available On
Listed Nov 2026
About claude-reflect
claude-reflect is an open-source, MIT-licensed system for Claude Code by Bayram Annakov. It captures the corrections and preferences you give Claude, then syncs them to CLAUDE.md and AGENTS.md as permanent memory. It also analyzes session history to propose reusable commands. The latest release listed is v3.3.1, and it ships in two forms: a Python plugin and a newer early-access mod.
What It Is
claude-reflect adds a learning loop to Claude Code. When you correct Claude (for example, "no, use gpt-5.1 not gpt-5"), the correction is detected and queued. You then review it and write it into your memory files, so Claude follows it in later sessions. A second capability, /reflect-skills, looks across past sessions for repeating requests and drafts skill files in .claude/commands/.
How the plugin works
The Python plugin runs in two stages. Hooks capture corrections automatically: at session start, on every prompt, before compaction, and after a git commit. Regex patterns catch corrections, positive feedback, and explicit remember: markers. You then run /reflect to review the queue. A semantic AI filter handles corrections in other languages and removes false positives, and each learning gets a confidence score. For each item you can apply, edit, or skip it.
Approved learnings can be written to:
- the global
~/.claude/CLAUDE.md - project and subdirectory CLAUDE.md files
- skill files in
.claude/commands/ - AGENTS.md, which works with Codex, Cursor, Aider, Jules, Zed and Factory
- documents referenced from those files
Other commands scan all past sessions for missed learnings (--scan-history), preview changes (--dry-run), and consolidate similar entries (--dedupe).
Skill discovery and improvement
/reflect-skills uses semantic reasoning to group differently phrased requests that share an intent. It reports candidates with evidence counts and asks where each skill should live, in a project or globally. It then generates draft command files. Corrections made while using a skill, such as /deploy, can be routed back into that skill's file.
reflect-mod (early access)
The mod runs inside a Claude Code session and needs Claude Code 2.1.286 or newer with mods enabled. Each short prompt goes to a one-shot model check (haiku by default) that decides whether it contains a reusable rule. Detected rules appear in a band above the prompt with Save, Global/This repo instead, Edit and Skip options. Nothing is written without a press. The README says to install either the mod or the plugin, not both, and that the plugin remains the choice for history scanning, dedupe and skill discovery.
Setup path
Install through the Claude Code plugin marketplace, then restart Claude Code. The plugin requires the Claude Code CLI and Python 3.6+, and supports macOS, Linux and Windows.
Community Discussions
Be the first to start a conversation about claude-reflect
Share your experience with claude-reflect, ask questions, or help others learn from your insights.
Pricing
Open Source (MIT)
claude-reflect plugin is released under the MIT License, an OSI-approved license, and is free to use, copy, modify and distribute.
- MIT License
- Captures corrections and syncs them to CLAUDE.md and AGENTS.md
- Skill discovery via /reflect-skills
- Multi-language correction detection
- Supports macOS, Linux and Windows
Capabilities
Key Features
- Captures corrections, positive feedback and explicit remember: markers via hooks
- Syncs approved learnings to CLAUDE.md and AGENTS.md
- Human review with apply, edit or skip
- Multi-language correction detection via semantic AI validation
- Confidence scoring of captured learnings
- Historical scan of past sessions
- Semantic deduplication of CLAUDE.md entries
- Skill discovery from repeating session patterns with /reflect-skills
- Routes corrections back into skill files
- reflect-mod: in-session review band with save, edit and skip
