Agent Automation & Development Skills

Applies the framework from Stuart Russell & Peter Norvig's "Artificial Intelligence: A Modern Approach" (AIMA) — the rational-agent model (PEAS, environment properties), search (uninformed, informed/heuristic A, adversarial minimax/alpha-beta), constraint satisfaction, knowledge representation and logical inference, reasoning under uncertainty (Bayesian networks, Markov models), and Markov Decision Processes / reinforcement learning — to problems that involve designing an agent, choosing a search or planning algorithm, modeling uncertainty, or picking the right AI technique for a task. Use whenever the user wants an "AI: a modern approach" or "AIMA" lens, mentions "rational agent," "PEAS," "heuristic search," "A search," "minimax," "alpha-beta pruning," "constraint satisfaction," "Bayesian network," "Markov decision process," or is deciding which class of algorithm/approach fits an AI or search/optimization problem — even if they don't name the book.

Applies Al Sweigart's "Automate the Boring Stuff with Python" framework — spotting repetitive/rule-based manual tasks worth scripting, picking the right tool (regex, file/folder ops via os/shutil/pathlib, CSV/Excel/PDF/Word/JSON handling, web scraping with requests/BeautifulSoup or Selenium, scheduling, email/SMS, GUI automation with pyautogui as a last resort), and safe-automation habits (dry runs, testing on samples, try/except around flaky steps) — to any task about automating a repetitive computer task with Python. Use whenever the user wants to "automate the boring stuff," mentions renaming/organizing/moving files in bulk, scraping a website, reading/writing Excel/CSV/PDF/Word programmatically, scheduling a script, sending automated emails, or automating clicks/keystrokes — even if they don't name the book.

Applies Kernighan & Ritchie's "The C Programming Language" (2nd Edition, ANSI C) framework — pointer/array duality and pointer arithmetic, reading declarations by the right-left rule, manual memory management and ownership, undefined behavior, the preprocessor and separate compilation, storage classes/linkage, and C's "trust the programmer" minimalism — to writing, reviewing, or debugging C code. Use whenever the user wants a "K&R approach," mentions "pointer arithmetic," "declaration reflects use," "malloc/free," "undefined behavior," "storage class," "header guards," or is writing/debugging C, tracking down a segfault, buffer overflow, or memory leak, or asking how a gnarly C declaration should be read — even if they don't name the book.

Applies Ben Horowitz's "The Hard Thing About Hard Things" framework (The Struggle, wartime vs. peacetime CEO, no-easy-answers decision making, the Law of Crappy People, hiring for strength vs. lack of weakness, taking care of people/products/profits in order, managing your own psychology, firing/demoting fast but respectfully, good vs. bad process, organizational debt) to hard leadership, CEO, hiring, firing, layoff, org-design, or crisis-decision problems. Use whenever the user wants a "hard thing about hard things approach," mentions "The Struggle," "wartime CEO," "peacetime CEO," "Law of Crappy People," "organizational debt," or is wrestling with a decision that has no good options, a layoff, a firing, a demotion, or how to keep functioning while a company/team is under real stress — even if they don't name the book.

Applies Tom M. Mitchell's "Machine Learning" (1997) framework — the well-posed learning problem (Task/Performance/Experience), hypothesis space and inductive bias, overfitting and the bias-variance tradeoff, proper train/test/cross-validation evaluation with statistical significance, the Bayesian learning perspective (MAP, ML, naive Bayes, Occam's razor via MDL), PAC learning and sample complexity, and matching algorithm family (decision trees, neural networks, instance-based, rule learning, reinforcement learning) to problem structure — to problems about designing, evaluating, or choosing a machine learning approach. Use whenever the user wants a "machine learning fundamentals approach," mentions "inductive bias," "overfitting," "bias-variance," "hypothesis space," "PAC learning," "MAP vs. maximum likelihood," "cross-validation," or is deciding which ML algorithm fits a problem, how to evaluate a model properly, or why a model generalizes poorly — even if they don't name the book.