ai-from-first-principles

Imported from raiseCatError/ai-from-first-principles (AGENTS.md). Install upstream with npx skills add raiseCatError/ai-from-first-principles. Copyright stays with the author.

Agent instructions for this educational repository

This is a learner-owned education repo, not a production model codebase. The priority is that the learner writes and understands implementations.

• Read README.md, progress.json, and the active weeks/NN.md before advising.
• Do not complete all exercises, generate working solutions into TODOs, or bypass acceptance gates unless the learner explicitly requests it.
• Give concise explanations, derivatives, tests, small hints and code review first.
• Respect .gitignore and do not commit secret keys, big datasets, trained weights or private user information.
• Do not run expensive compute, create external accounts, publish or overwrite repos, or delete experimental data without explicit authorization.
• On macOS, heavy work must pass the external-drive check. No fallback to internal disk without consent.
• When users request actual software changes, ask only essential questions; avoid destructive operations and preserve learner code.
• Never fabricate test success, model accuracy, learning completion, paper sources or training runs.
• Mark tracker progress done only on explicit user request with an evidence artifact.

Skills: skills/ai-learning-mentor/SKILL.md, skills/ml-debugger/SKILL.md, skills/research-reproduction/SKILL.md.