🦋 Skills¶

Skills are reusable instruction packs — the difference between “generic coding bot” and “knows how we ship”. ✨

Mental model 🧠¶

bundled skills  +  project overrides  +  learned skills
        └──────────► matched into TaskPacks 📦

Index file: .slmcode/SKILLS.md (auto-maintained — don’t micro-manage unless you must).

SKILL.md shape 🧬¶

---
name: atomic-coding
description: Prefer tiny diffs, clear acceptance checks
triggers: refactor, cleanup, helper
agents: worker, deep, corrector
user-invocable: true
---

# Atomic coding

- One concern per task
- Touch only listed files unless discovery proves otherwise
- Leave tests greener than you found them

Day-to-day commands 🛠️¶

slmcode skills                 # list
slmcode skills show atomic-coding
slmcode skills new my-skill --agents worker
slmcode skills edit my-skill   # project override

Pin or reference 📌¶

slmcode run --skill atomic-coding "Refactor helpers"
slmcode run "Fix login @skill:multipass-quality"

Studio → Skills panel has pin chips. Config:

pinned_skills:
  - atomic-coding
  - multipass-quality

Bundled starters (taste) 🍿¶

Exact set evolves — slmcode skills is truth. Docs can lie; the CLI rarely does.

Teaching the flywheel 🦋¶

After good runs, check:

slmcode docs show SKILLS.md
ls .slmcode/skills/learned/

Promote hard-won lessons into a project skill when they stop being optional.

Related 🔗¶

• 🧩 Agents — who consumes skills
• 🧠 Concepts — flywheel
• ⚙️ Config — pinned_skills, skills_dirs

☀️ Made with ♥ by UnicoLab