Analysis updated 2026-05-18
Give a coding agent persistent memory across sessions using plain markdown files in git.
Track decisions and their reasoning so the next session does not repeat past mistakes.
Log tricky problems the agent hit so future sessions do not fall into the same trap.
Get a safety reminder when too many commits pile up without an updated state summary.
| kaiwutech-tw/flightwake | agi-ruby/ai-gpt_image2-seedance_2.0-video-skills | arindam200/wisprgemma | |
|---|---|---|---|
| Stars | 10 | 10 | 10 |
| Language | JavaScript | JavaScript | JavaScript |
| Setup difficulty | easy | easy | moderate |
| Complexity | 2/5 | 1/5 | 3/5 |
| Audience | developer | vibe coder | general |
Figures from each repo's GitHub metadata at analysis time.
Requires Node.js 18 or later at install time.
Flightwake is a lightweight framework that helps AI coding agents like Claude Code keep track of what they did across work sessions. Since an AI agent's memory resets whenever a session ends, it can be hard for a new session to pick up where the last one left off. This tool solves that by having the agent write short records as it works: a current state summary, a running log of decisions that closed off other options and why, and a log of tricky problems it ran into, all stored as plain markdown files inside your git repository. It installs by running a single command in your project, which sets up a folder of templates, adds a handful of skills the agent can use, and updates your agent's instruction file, whether that is for Claude Code, Codex, or Gemini, with a table describing when it should create each kind of record. There are no ongoing dependencies once it is installed, aside from Node.js being available to run the setup and hooks. Day to day, you start a session by telling the model to read its notes, and it summarizes where things stand and where to continue. As the agent works, it appends short entries whenever it makes a hard decision or hits an unexpected problem, and when you say you are wrapping up, it writes a fuller record and updates its state summary. If too many changes pile up without the state file being updated, a safety check will pause the session once to remind you. The project explains it is meant to add memory and discipline to an already capable AI model, not to make a weaker model smarter. It grew out of watching a real multi day coding session where useful notes were improvised on the spot, and it turns that improvised habit into a repeatable, installable convention. The README notes there are translations available in Traditional Chinese, Simplified Chinese, and Japanese, and the project is distributed through npm.
A lightweight markdown based framework that gives AI coding agents like Claude Code persistent memory across sessions by recording state, decisions, and traps directly in git.
Mainly JavaScript. The stack also includes JavaScript, Node.js, Markdown.
The README excerpt does not state a license, so terms of use are unknown.
Setup difficulty is rated easy, with roughly 5min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.