dotfiles

Get yourself to actually OPEN a dense paper you have been avoiding, using one "bet before you read" prompt. Claude reads the paper silently, then asks you to predict its most interesting real finding BEFORE you read, withholds the answer, and points you to exactly where in the paper it is settled. The withheld answer is the reason to open the paper. Use when the user is procrastinating on / avoiding a paper, says "I can't get myself to read this", "make me want to read this", "get me to start", "I keep putting this paper off". Do NOT use to summarize (that kills the urge to read), to quiz someone who has already read it (use quiz-paper), or to critique it (use socratic).

Turn reading an arxiv paper into a Predict-then-Reveal game where curiosity, not points, is the reward. Claude reads the paper silently, then asks you to predict answers before revealing them, one question at a time, with rising difficulty. Use when the user wants to ENJOY reading a paper, be quizzed on it, learn it actively, or "make reading this fun" instead of getting a summary dumped at them. Triggers on "quiz me on this paper", "make me read this", "test me on arXiv 1234.5678", "gamify this paper", "help me actually learn this paper". Do NOT use for a passive summary (use skim:story/skim:deep) or an adversarial critique (use socratic).

Use this skill whenever the user wants to improve the WRITING itself — prose quality, clarity, storytelling, cutting clutter, making dense technical text readable — for ML/AI research writing: paper drafts, abstracts, intros, related work, rebuttals, blog posts, README, grant text, or any passage they want to read better. Triggers: 'make this clearer', 'tighten this', 'edit my writing', 'is this well written', 'help me write the intro/abstract', 'cut the clutter', 'this reads badly', 'improve the prose', 'rewrite this paragraph', 'review my writing', 'draft my paper', 'write the paper/section from my results/code', 'write from scratch', 'draft the related work from these papers', 'turn my notes into a draft'. This governs the CRAFT of the words (clarity, story, value-to-reader, sharpness) — NOT file formatting (docx/latex), citations (phd-skills), or slides (research-deck). Grounded in Zinsser (On Writing Well), Schimel (Writing Science / OCAR), McEnerney (communicating value), Lanham (paramedic method), Will

Use this skill whenever the user wants to create or improve a slide deck / presentation for MACHINE-LEARNING or AI research — paper talks, conference presentations (NeurIPS/ICML/ICLR/ACL/CVPR style), seminar/reading-group slides, thesis or proposal defenses, lab meetings, poster talks, or any deck where the audience evaluates a method and its results. Triggers: 'make slides for my paper', 'conference talk', 'research presentation', 'ML deck', 'thesis defense', 'reading group slides', 'present these results', or a .pptx filename in a research context. This skill governs CONTENT, ARGUMENT STRUCTURE, and NARRATIVE (the what/why/how/results arc). For the technical work of writing the .pptx file, it drives the installed `pptx` skill. Adapted from Gabberflast/academic-pptx-skill (MIT), ML-tuned.

Socratic critical-thinking partner for ML research. Interrogates a claim/hypothesis/method/result — surfaces hidden assumptions, demands evidence, plays devil's advocate, steelmans the opposite, checks against real literature. Use when the user wants to pressure-test an idea, asks "poke holes in this", "am I missing something", "is this novel/sound", "critique my hypothesis/method/result", or "argue against this".

Read staged files and write a concise commit message for user approval.

Explain code, concepts, or architecture with analogies, diagrams, and a conversational tone. Use when the user asks "how does this work?", "what does this do?", or "explain this".

Test the last implemented feature, then ask if it should be added as a pytest test. Use after implementing any feature or change.