Building a Claude Skills Library Across Projects

One developer’s growing collection of AI memory.

The Ecosystem

After several months with Claude Code, I have skills scattered across projects:

Project	Skills	Purpose
step_quest	2	Changelog style, WIF setup
C4IROcean	7	Ocean data, STAC, Dask, notebooks
form-copilot	1	Form navigation commands

Each project teaches Claude something different. Together, they form a library.

Patterns That Emerged

1. Name After Keywords

changelog.md     → triggers on "changelog"
wif.md           → triggers on "WIF", "keyless CI"
dask-odp.md      → triggers on "dask", "distributed"

2. Have Claude Write Them

Don’t write skills from scratch. Do the work once with Claude, then:

“Save what we just did as a Claude skill for next time.”

Claude writes the skill. You review. Future sessions inherit it.

3. Project vs Global

~/.claude/skills/           # Follows you everywhere
project/.claude/skills/     # Project-specific

Global: coding style, git conventions, personal preferences Project: domain knowledge, API patterns, team standards

4. Link Skills to Docs

## Further Reading
See the [WIF skill](link) for copy-paste setup.

Blog posts explain why. Skills provide how.

The Compound Effect

New project? Copy relevant skills:

cp -r ../other-project/.claude/skills/useful-skill.md .claude/skills/

Skills accumulate. Context grows. AI gets smarter per-project.

What I’d Do Differently

• Start with skills earlier (not after repeating myself 10 times)
• Keep them shorter (under 100 lines)
• Add examples, not just explanations

---

Your AI has memory. Use it.

Step Quest skills · Ocean Data skills