A tool that lets AI agents reuse hard-won project know-how
self-learning-skills is an open-source tool that helps reuse work methods they discovered during difficult tasks. It works with Claude Code, Cursor, Codex, and other agents that read AGENTS.md or standing instruction files. The main idea is to save repeatable steps such as deploy commands, access paths, credential locations, or live verification checks after the agent has figured them out through trial and error.
Claude Code saves this as a new SKILL.md file, Cursor saves it as a learned rule file, and Codex, Zed, Aider, and similar tools can save it in AGENTS.md or . It can be installed with npx skills add kulaxyz/self-learning-skills, with options for global installation or a specific agent. Multi-step repeatable workflows become a new skill or rule, one-line facts become lightweight memory, and true one-off lessons are skipped.
It is designed not to store passwords, tokens, connection strings, or API keys, and instead records only where those secrets should be found.
Key points
- can reuse project workflows discovered through trial and error.
- Claude Code stores the lesson as SKILL.md, Cursor stores it as a rule, and Codex-style agents can use AGENTS.md or memory.
- Installation is available through npx skills add kulaxyz/self-learning-skills.
- Repeatable multi-step workflows are saved, small facts go to memory, and one-off lessons are skipped.
- Secrets such as passwords, tokens, and API keys are not supposed to be written into shared files.