[Skills Index] ATOM — Top 10 Claude Code Skills Roadmap #33
Description
ATOM Skills Roadmap
Top 10 Claude Code skills to accelerate AMD's LLM inference engine development. Each skill encodes team knowledge into a reusable, shareable automation.
Skills Tracker
| # | Skill | Priority | Issue | Status |
|---|---|---|---|---|
| 1 | quant-validate |
P0 | #23 | 🔲 Not Started |
| 2 | model-onboard |
P1 | #24 | 🔲 Not Started |
| 3 | benchmark-report |
P1 | #25 | 🔲 Not Started |
| 4 | profile-decode |
P1 | #26 | 🔲 Not Started |
| 5 | docker-clean-build |
P1 | #27 | 🔲 Not Started |
| 6 | config-tune |
P2 | #28 | 🔲 Not Started |
| 7 | upstream-sync |
P2 | #29 | 🔲 Not Started |
| 8 | ckfree-validate |
P1 | #30 | 🔲 Not Started |
| 9 | serve-model |
P1 | #31 | 🔲 Not Started |
| 10 | accuracy-compare |
P2 | #32 | 🔲 Not Started |
How to Contribute
- Pick a skill from the table above
- Assign yourself to the corresponding issue
- Build the skill following the spec in the issue
- Skills live in
.claude/skills/asSKILL.mdfiles - Submit a PR with the skill implementation and update this tracker
Context
ATOM is AMD's lightweight LLM inference engine built on AITER kernels. These skills cover accuracy validation, model onboarding, Docker workflows, performance benchmarking, and serving automation. Target hardware: MI300X (gfx942) and MI355X (gfx950).