From claude-eureka
Debugging guidance for research code — triggers on: error, exception, failed, crash, NaN, OOM, CUDA, traceback, bug, broken
How this skill is triggered — by the user, by Claude, or both
Slash command
/claude-eureka:research-debuggingThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
When encountering errors in research code, follow this order strictly.
When encountering errors in research code, follow this order strictly.
Check in this sequence — most bugs are data bugs:
git diff HEAD~3 --stat
git log --oneline -5
If it worked before, the bug is in the diff.
NaN values:
torch.autograd.set_detect_anomaly(True) temporarilyOOM (Out of Memory):
.detach() or with torch.no_grad())torch.cuda.memory_summary()CUDA errors:
.device for all tensors involved.shape at each stepnvidia-smi and torch.cuda.is_available()npx claudepluginhub gaaaavin/claude-eureka --plugin claude-eurekaDiagnoses ML/AI failures like OOM, NaN, divergence, crashes, bad throughput, wrong outputs, and dependency conflicts using grounded framework docs and citations.
Provides systematic debugging workflow including error classification, binary search, and log tracing for investigating code errors, exceptions, and failures.
Debugs Python errors, exceptions, and unexpected behavior by analyzing tracebacks, reproducing issues, isolating with pdb/icecream/print debugging, identifying root causes, and providing fixes.