Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
A self-improving skill for AI coding agents (Claude Code, Cursor, AGENTS.md): recognize a hard-won golden path in a session and harvest it into a reusable skill/rule for next time.
| Date | Stars |
|---|---|
| 2026-07-31 | 929 |
| 2026-08-03 | 936 |
| 2026-08-06 | 936 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
# self-learning-skills **A self-improving skill for AI coding agents.** Works with Claude Code, Cursor, and any agent that reads an `AGENTS.md` / standing-instructions file. Every session you do hard debugging or rediscover the same thing — *how do I reach the prod DB? where do the creds live? what's the deploy command? how do I verify this live?* — and that hard-won knowledge evaporates when the session ends. The next session starts from zero and re-learns it. **self-learning** fixes that. It teaches your agent to recognize the moment it has just earned a reusable **golden path** and persist it where the tool will auto-load it next time — so the next session starts already knowing the route instead of rediscovering it. It's a *meta-skill*: it doesn't do the work, it captures **how** the work got done — including the **failures**, since skipping a known dead-end next session is often worth more than the win itself. ## The loop (same everywhere) 1. **Recognize the moment** — a task that only worked after several tries, a non-obvious command, a project fact you didn't know up front, an operational workflow likely to recur, or you simply saying *"remember this"*. 2. **Capture it, no prompt needed** — it acts on the cue immediately, picks the scope/name itself, and tells you afterward. The *procedure* is captured (not a one-off answer), plus a "what didn't work" note. 3. **Reuse** — next session the entry loads automatically, by skill/rule description or because the instructions file is always read. What differs per tool is only *where* knowledge is persisted and *how* it's auto-loaded: | Tool | Persists golden paths to | Auto-loads via | |---|---|---| | Claude Code, Codex, Agent Skills clients | a new `skills/<name>/SKILL.md` | skill description matching | | Cursor | a new `.cursor/rules/learned/<name>.mdc` | rule description / globs | | Zed, Aider, Gemini CLI, … | `AGENTS.md` (or project notes/memory) | always-read instructions | ## Install ### `npx` — recommended (works with 70+ agents) Uses the community [`skills`](https://github.com/vercel-labs/skills) CLI, which installs into whatever agents it detects — Claude Code, Cursor, Codex, Cline, OpenCode, and more: ```bash 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: ```bash 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 <details> <summary>Copy the files into place yourself</summary> ```bash git clone https://github.com/kulaxyz/self-learning-skills # Claude Code — global (or into a project's .claude/skills/ to share via git) cp -R self-learning-skills/skills/self-learning ~/.claude/skills/ # Cursor — auto-loads .cursor/rules/ (harvested rules land in .cursor/rules/learned/) mkdir -p .cursor/rules cp self-learning-skills/.cursor/rules/self-learning.mdc .cursor/rules/ # Any AGENTS.md agent (Codex, Zed, Aider, Gemini CLI, …) curl https://raw.githubusercontent.com/kulaxyz/self-learning-skills/main/AGENTS.md >> AGENTS.md ``` </details> ## Triage: skill, memory, or skip? It won't bloat your config with one-liners. Each lesson is routed: | Lesson | Where it goes | |---|---| | A multi-step, reusable **procedure/workflow** | a new skill / rule | | A single **fact or one-line correction** | lightweight notes/memory (e.g. a `MEMORY.md`) | | A genuine **one-off** | skipped | ## 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** hol
Excerpt of 5,851 characters
Read on GitHub
5
Would you bet a product on this? Bounded 0–100 and slow moving.
matched fp:9a6e69edbf2fa933, llm:description: 'A self-improving skill for AI coding agents ... recognize a hard-won golden path in a session and harvest it into a reusable skill/rule for next time.'
matched fp:9a6e69edbf2fa933, llm:description: 'A self-improving skill for AI coding agents ... recognize a hard-won golden path in a session and harvest it into a reusable skill/rule for next time.'
matched fp:9a6e69edbf2fa933, llm:description: 'A self-improving skill for AI coding agents ... recognize a hard-won golden path in a session and harvest it into a reusable skill/rule for next time.'