← ClaudeAtlas

experimento-kagglelisted

Run /skill:experimento-kaggle <descripción> in Pi to propose a block diagram + notebook for a code/model experiment and run it on Kaggle after user approval.
amalvarezme/chec-local-uiti-vano-interpreter · ★ 0 · AI & Automation · score 63
Install: claude install-skill amalvarezme/chec-local-uiti-vano-interpreter
# Pi Kaggle Experiment Skill Use this skill for: ```text /skill:experimento-kaggle <descripción del experimento> ``` This is a thin Pi adapter over the canonical skill at `.claude/skills/experimento-kaggle/SKILL.md`. That file owns the full Activation Contract, Hard Rules, Execution Steps, and Output Contract — including the mandatory human-approval gate before any Kaggle push and the smoke-before-full verification rule. Read and follow it directly; do not duplicate its logic here. ## Rules - Follow the canonical skill's Hard Rules exactly, in particular: never push to Kaggle without explicit user approval of the diagram and notebook, never touch or store the user's Kaggle credentials, and never reimplement a model that already exists in `src/`. The no-reimplementation rule, its precondition guard, and the Experiment Families dispatch table it feeds live only in the canonical skill — do not copy their text or baseline digits here. - Uploading any local file to Kaggle as a private dataset needs its own second, explicit approval — separate from the notebook-approval gate. See the canonical skill's Dataset Transport section for which datasets each family needs; do not duplicate slugs or file lists here. - Kaggle package install, credential presence check, the interactive Auth & Config flow (token placement instructions, resolving the real account username from the CLI's own config, the auth liveness check, and the full-mode accelerator prompt) live only