How to make AI coding agents automatically apply lessons from previous tasks
This plugin reduces the friction of having to look up what was learned in previous sessions every time you fix code. It provides a cycle where the AI agent organizes requirements, plans, writes code, and reviews it, then documents the knowledge gained during that process. Because the agent reads these documents when the next task begins, its judgment improves as the project progresses.
The six-step loop for accumulating knowledge
The core is a six-step loop: 'brainstorm-plan-execute-simplify-review-learn'. You use /ce-brainstorm to flesh out ideas and /ce-plan to develop them into an actionable plan. Finally, /ce-compound records what was learned from the current task in the docs/solutions/ folder.
Context-specific skills and autonomous execution
In addition to the default loop, skills tailored to specific situations are available.
- Support for various hosts It works on 14 AI agent hosts, including Claude Code, Cursor, Codex, and GitHub Copilot.
Installation and initial setup
The installation command varies depending on the AI coding tool you are using. After installation, running /ce-setup in your project generates configuration files and reports on necessary tool features. If a previous version is already installed, you must refresh the marketplace cache and then update to apply the latest version. Be careful, as running only /plugin update may leave the previous version in place.
Considerations before adoption
This plugin is released under the MIT license and is primarily maintained by Kieran Klaassen and Trevin Chow. It reflects a specific perspective on AI engineering philosophy, so not all contributions are accepted. Some hosts, such as OpenCode, Pi, and oh-my-pi(omp), use native package loading, and the Bun CLI is for repository development rather than general installation. If you want to apply organization-wide conventions or security policies to all repositories, you can review the experimental Compound Packs feature, which enforces pre-defined rule folders during the planning and review stages. Before adopting it, it is advisable to check whether your workflow fits this cycle and whether your AI agent host is on the official support list.
