Home-made skills for OpenCode.
| Skill | What it does |
|---|---|
skills/nim-development |
Writing, building, and debugging Nim — decision guidance plus a research discipline for finding current, version-correct Nim info (thin training-data footprint). Carries a persistent trap log (nim4friends) with an owner-verified inbox. |
skills/doc-sync |
Sync project docs to reality for a cold handoff. Inventory-first, six mandatory steps. |
skills/nftt |
Drive a consuming project through the nFTT fine-tuning loop — init, config/data declaration, pipeline, train, export, eval, reading results. Topic-split reference under references/topics/. |
Clone once, symlink the skills your loader discovers:
git clone git@github.com:DAHLS/agent-skills.git ~/.agents/agent-skills
ln -s ../agent-skills/skills/nim-development ~/.agents/skills/nim-development
ln -s ../agent-skills/skills/doc-sync ~/.agents/skills/doc-sync
ln -s ../agent-skills/skills/nftt ~/.agents/skills/nftt(Adjust paths to wherever your tool expects skills; this is the layout used
on the author's machines.) Pulling inside ~/.agents/agent-skills updates
every installed skill at once.
Every SKILL.md may carry a local house field risk:; spec-compliant
loaders ignore unknown fields. Each skill may ship an agents/ folder with
tool-specific interface hints (openai.yaml).
Run the official checker from agentskills/agentskills:
skills-ref validate ~/.agents/agent-skills/skills/<name>GPL-3.0. See LICENSE.