Copy this prompt into the AI assistant you're using.
Follow https://skillsmp.com/skill-install/prompt.md to install Agent Skills from https://github.com/markusstrasser/skills.Loading the repository README…
Showing 40 of 55 collected skills.
Use when: writing, refactoring, reviewing or fact-checking flashcards — the operator's Spaced cards (~/cards, formerly Anki), suggestions in his Suggestions view, generated cards, card ctx/answers/precision, card diagrams. Rules digest (no cloze, one answer,…
Use when: drafting email/text/outreach/LinkedIn/scheduling AS Markus. Registers + banned vocab. NOT research memos (/research) or AI-prose cleanup only (/de-slop).
Use when: YouTube/podcast transcript via yt-dlp for cite or corpus. Transcript-only — NOT video analysis.
Use when: monitoring curated X accounts, cashtags, finance tweet pulls. Paid API ($/post). NOT general search (/research).
Use when: improving a GPT Image prompt before generation (current: GPT Image 2.6 Sunburst). NOT scene injection (/person-into-scene).
Use when: /llmx-guide, writing llmx calls, debugging llmx failures, choosing model/provider, or reading/messaging a running Codex Desktop thread. Python/Bash gotchas — flags, transports, subscription vs API.
Use when: photoreal Markus in scene template, 'put me in this photo', dating/lifestyle/cinematic shots, likeness fix. NOT generic image gen (/gpt-image-2-6).
Use when: stuck, about to wrap up, or the operator asks what you haven't thought of. One thinking prompt or hole-poke from his usual moves (operator-deck): a divergent angle or an obvious miss, never process or agent orchestration. NOT a checklist to run…
Use when: why/root-cause, confounder, regression, ACH, DAG adjustment, weakest-link, official/administrative statistic. Lenses: null, causal, dag, hypotheses, fraud, audit, pipeline, stop. NOT ideation (/brainstorm).
Use when: Modal deploy/run/debug, GPU/resource config, `from modal import`. v1 API gotchas. NOT generic cloud (/bash only).
One-shot source-grounded research: /research, 'find papers about', 'what's known about', literature reviews, and science/benchmark memos. Match depth to the question. Not implementation; recurring research and compilation use /research-ops.
Choose a model and effort by principle: the cheapest frontier model that passes the task's verifier, cross-lab review for consequential judgment, facts from tools. Pareto-frontier roster (Opus 5.5, GPT-6.1 Sol, GPT-6 Astra/Luna, Gemini 4 Argon, Fable 5.1,…
Use when: /critique, 'review plan', 'what's wrong', fact-check plans/findings/closeout. Modes: model, verify, close. NOT code diffs (/diff-review) — critique owns design/plan layer.
Use when: /diff-review, review our diff/commit range/changed files, plan/phase closeout diff layer. GPT-6 Astra (codex) scout; parent validates against code. NOT plan/findings review (/critique); NOT Anthropic's built-in /code-review.
Use when: /eval, new benchmark/grader/judge, model bakeoff, auditing eval design. Pre-reg decision rule + power preflight. Modes: design, review. NOT spot-checks (/verify-before) or code diffs (/diff-review).
Use when: /brainstorm, divergent design space, denial/domain/constraint perturbation. Optional multi-model volume. NOT convergent review (/critique) or decisions (/decide).
Use when: /doe, designing a NEW experiment/probe/prereg from a hypothesis+budget — generates arms/shams/controls/bands/power/stopping-rule as a fill-in doc. Complements /eval (grades/scaffolds benchmarks) and dispatch_lint (mechanically enforces a prereg…
Use when: /ambition, 'make this more ambitious', or before a substantive task whose brief is the operator's first guess at a means. Elicits from Markus the goal beyond the goal, the raw source, the outside judge and the fields to borrow, then rewrites the…
Use when: 'de-slop', AI-prose cleanup, pre-publish voice check. Hunts vocabulary/structure tells — NOT full drafting (/writing-style).
Analyze agent sessions and improve shared workflows or codebases through /observe modes. Excludes ordinary edits, single bugs, diff reviews and session-end digests.
Use when: /rsi close, session-end digest nudge, goal achieved + verify one claim. fm.py evidence or one [obs]. NOT deep retro (/observe retro).
Use when: parallel adversarial bug-hunt fan-out or pre-commit audit; scouts write files, orchestrator model triages. NOT a diff-review scout (/diff-review) or plan critique (/critique).
Execute approved plan with hygiene — phased, verify-before-claim, granular commits, worktree subagents, inline tooling. /critique close at end. After /decide or plan file. NOT exploration (/decide).
Use when: reviewing animation/motion/transition code, 'does this feel right', easing/duration choices, hover/press states, motion polish pass. Motion craft only — NOT general code review (/diff-review) or interaction architecture (/interface-thinking).
Headless/scripted browser automation from Bash (Vercel Labs CLI, Rust/CDP). Use for navigating pages, filling forms, clicking, screenshots, scraping JS-rendered pages, testing web apps, Electron apps, or any programmatic web interaction — especially in…
Web scraping and data download toolkit — curl_cffi, Scrapfly, Firecrawl, Browserbase, claude-in-chrome, Exa, agent-browser. Covers which tool for which situation, API keys, fallback chains, structured extraction, authenticated session approaches, and what…
Architecture/strategy decision arc — ground → diverge → cross-model critique → capture. For hard-to-reverse decisions (schema, breaking refactor, tech choice). NOT trivial (/critique) or ideation (/brainstorm).
Use when: building or restyling an app, interface, page, 3D scene, animation, diagram, poster or artwork and choosing a visual direction — pull exemplars from Markus's own saved media (Pinterest, Twitter, Instagram) before designing. NOT data-form choice…
Use when: headless `cursor-agent` CLI from shell/CI/scripts, install/auth, or exact Cursor Grok 4.7 read-only repo review. Composer is retired (2026-10-07). NOT in-editor Cursor agent or llmx/claude-cli.
Use when: dry-run before expensive work, 'did it work/crash', live status probe, pre-register prediction before results. Modes: probe, status, preregister. NOT eval design (/eval) or adversarial review (/critique).
Use when: downloading/staging external data (Census, IPUMS, NCES, intel download scripts). Probe→stage→register. Then /dataset-register. NOT papers (/research).
Use when: 'compile memos', training-data diff, parallel research dispatch, generate/execute lanes. NOT one-shot questions — use /research. Loop conductor: /observe maintain.
Use when: deciding WHETHER and HOW to show data — sentence vs table vs chart, which form (incl. distributions, uncertainty, multiverse/sensitivity), which tool (Plot/d3/matplotlib/TikZ/three.js), and checking a figure with a blind reader. Styling of HTML…
NATO Admiralty System for grading source reliability and information credibility. Apply automatically during OSINT, forensic investigations, legal research, entity audits, or fraud analysis — NOT for general software or casual research. Every claim gets a…
Use when: biomedical source lookup (ClinVar, gnomAD, Ensembl, PubMed, OpenTargets…). Route APIs before synthesis. NOT general memos (/research) unless life-science topic.
Bio/medical/scientific evidence and anti-hallucination reference for claim-heavy medical research, genomics interpretation, supplement evaluation, pharmacogenomics, or clinical synthesis. Not a workflow for casual health questions or other technical domains.
Use when: vendor/agent landscape scan, 'what's new', trending repos, weekly update check. Filtered vs known memos — NOT deep topic research (/research).
Build apps with the Claude API or Anthropic SDK. TRIGGER when: code imports `anthropic`/`@anthropic-ai/sdk`/`claude_agent_sdk`, or user asks to use Claude API, Anthropic SDKs, or Agent SDK. DO NOT TRIGGER when: code imports `openai`/other AI SDK, general…
Use when: Manim/3B1B-style math animation, equation visuals, render manim scenes. NOT static figures (/scientific-drawing).
Use when: /scientific-drawing, architecture/ERD/math diagram, plot a function. Typst/CeTZ default; TikZ/D2/Asymptote when needed. NOT paid illustration SVGs (/illustration-gen) or Manim (/manim-animations).
Repository README source: README.md
Read the source README