Skills that make AI coding assistants more useful — neurodivergent support, LoRA training guidance, and more. Built for Claude Code, works with any LLM agent that reads markdown.
| Skill | What It Does | Standalone? |
|---|---|---|
| nd-core | Communication style — bullets over walls of text, explicit over ambiguous, structured over freeform | Yes (recommended base) |
| nd-frustration-support | Notices overwhelm signals and checks in gently — no therapy, no patronizing | Yes |
| nd-focus-support | Catches ADHD drift and nudges back on topic without judgment | Yes |
| nd-plural-support | Warm check-in for plural systems with simple session logging | Yes |
| Skill | What It Does | Standalone? |
|---|---|---|
| sounding-board | Active listening and talk-it-out space — reflects, asks light questions, reminds you of your own coping tools, knows when to refer out | Yes |
| Skill | What It Does | Standalone? |
|---|---|---|
| lora-trainer-guide | Preset selection + parameter reference for LoRA training — covers SD1.5, SDXL, SD3, Flux1 with 38 real-world presets | Yes |
Copy the skills you want to ~/.claude/skills/:
# All skills
cp -r skills/* ~/.claude/skills/
# Just the ones you want
cp -r skills/nd-core ~/.claude/skills/
cp -r skills/nd-plural-support ~/.claude/skills/Any agent that loads context from files can use these. Point it at the SKILL.md files in whatever way your agent supports.
See SKILL_TEMPLATE.md for a fill-in-the-blank template.
- Modular — install only what you need, remove what you don't
- Casual, not clinical — pair-programmer vibes, not therapist energy
- Signal-based, not diagnostic — respond to what you see, don't label why
- Lean — ship minimal, iterate later
See SKILL_IDEAS.md for ideas in the pipeline (cloud GPU deployment, model architecture guide, and more).
PRs welcome. If you've got a useful pattern for AI assistants, add it.