Memory-backed coding agent
Suggested by runchr · yesterdayPulled 1 time
Draft — a suggested stack, not yet verified by a member.
Keep what an agent learned about a repo across sessions.
What they did
This stack gives Codex a memory plugin and the GitHub MCP server. The agent records the decisions it made about the repo, recalls them in the next session, and keeps the issue it is working on current. Suggested, not verified.
How it went
An agent that stops re-learning the same repository every session.
Worth knowing
Memory can carry a wrong decision forward, so review what it stores.
Wiring
Install Codex, add the memory plugin, and connect the GitHub MCP server. Have the agent write a short decision note after each task and read the notes at the start of the next one.
Try it yourself
Recall what we decided about the auth module last week, continue that work, and update the issue with what changed.
Tools in this stack
Run this stack
Install the stack into this project. It writes AGENTS.md and the manifest, merges the MCP servers into .mcp.json, and — with --yes — runs the tool installs.
npx runchr install runchr/memory-backed-coding-agentcurl -s https://api.runchr.com/api/recipes/runchr/memory-backed-coding-agent/manifestSystem prompt
You are a coding agent with memory. You recall earlier decisions before you act, and you write down the decisions you make.
The manifest is also at api.runchr.com/api/recipes/runchr/memory-backed-coding-agent/manifest.