Kulaxyz/self-learning-skills

Install

npx — recommended (works with 70+ agents)

Uses the community skills CLI, which installs into whatever agents it detects — Claude Code, Cursor, Codex, Cline, OpenCode, and more:

npx skills add kulaxyz/self-learning-skills                 # this project (auto-detects agents)
npx skills add kulaxyz/self-learning-skills -g              # global — all your projects
npx skills add kulaxyz/self-learning-skills -a claude-code  # a specific agent

Try it once without installing:

npx skills use kulaxyz/self-learning-skills --skill self-learning | claude

Claude Code plugin

/plugin marketplace add kulaxyz/self-learning-skills
/plugin install self-learning@self-learning-skills

Manual

Triage: skill, memory, or skip?

It won't bloat your config with one-liners. Each lesson is routed:

Promotion rule (don't enshrine guesses)

Triage decides granularity; the promotion rule decides confidence. A skill is authoritative — the next session trusts it without re-deriving it — so a session is only promoted to a skill when all three hold:

1. A passing check — the path was actually verified (a test passed, a clean exit, a green build, a reproduced repro). "Seemed to work" doesn't count.
2. A named failure pattern — the failure it avoids or diagnoses, named.
3. At least one ruled-out dead-end — a concrete approach tried and eliminated.

Miss any one and it stays a tentative memory note (or is skipped) rather than a skill. This keeps confident-but-unverified guesses out of the skill set. (Promotion rule suggested by community feedback.)

Safety

Harvested skills/rules get committed and shared, so this is built to never write secret values — no passwords, tokens, connection strings, or API keys. It records only where to find a secret (env var name, a client/selector function, an MCP tool, a secret manager). Reproducing a secret into a shared file leaks it.

Repo layout

self-learning-skills/
├── AGENTS.md                          # generic, cross-tool version of the loop
├── skills.sh.json                     # registry manifest for `npx skills` / skills.sh
├── .claude-plugin/
│   └── marketplace.json               # Claude Code plugin manifest
├── skills/
│   └── self-learning/                 # Agent Skills standard (Claude Code + clients)
│       ├── SKILL.md                   # recognize-the-moment + harvest procedure
│       ├── references/
│       │   └── skill-authoring.md     # condensed spec the writer loads to author a good skill
│       └── assets/
│           └── SKILL.template.md      # fill-in template for harvested skills
└── .cursor/
    └── rules/
        ├── self-learning.mdc          # Cursor adapter (always-applied rule)
        └── learned/                   # harvested Cursor rules land here

Built on the open Agent Skills standard.

License

MIT © kulaxyz