These skills turn open-ended AI work into bounded procedures with explicit inputs, verification, stopping conditions, and useful outputs.
Every page includes the complete instruction file, links to the original source, and distinguishes creator-authored or reviewed packages from independent XSkills adaptations.
Andrej Karpathy
Karpathy Autoresearch Skill for Claude Code & Codex
Turn a measurable optimization problem into a bounded agent loop that proposes one change, runs the same evaluation, keeps improvements, and records every result.
Tom Crawshaw
Agent-graph research workflow
Convert a research question into a dependency-aware agent graph with independent research branches, a separate skeptic, evidence-gated synthesis, and manual approval before delivery.
Theo
Model-strength task router
Apply Theo's source-era model rankings and routing defaults, use the named Codex execution paths, judge the result, and escalate when the first route misses the bar.
Atai Barkai
User-correction learning loop
Pair an agent's miss with the user's corrective action, distill the pair into readable procedural memory, review it separately, and make it available to the next matching run.
Matt Shumer
Build, Critique, and Improve Against a Hard Bar
A source-grounded Fable workflow for broad-goal building, independent grading against real output, and repeated biggest-gap improvement. XSkills adds authorization, privacy, budget, and consequential-action gates.
Dax Raad
Derive a Read-Only Client from HAR Traffic
A narrowly scoped reconstruction of Dax Raad’s demonstrated browser-control, HAR-capture, and derived-client sequence for authorized read-only retrieval. It excludes ordering, payment, account mutation, and access-control evasion.
Dara A.
Weaver Runtime Principles and Budget Diagnostics
A planning and review skill for Weaver SDK widget-runtime work: understand before building, use receipt-backed tripwires, and expose budget failures as actionable agent-facing diagnostics.
Thijs
Visual Wardrobe and Outfit Concepts
A consent-scoped workflow for turning authorized clothing photos into a reviewable wardrobe collection, proposing traceable outfit combinations, and optionally preparing GPT-image outfit concepts.
zostaff
Football Video Performance Analytics
A source-faithful implementation or review framework for a claimed YOLO, OpenCV, and Python match-video CV pipeline. XSkills adaptations add authorization, calibration, uncertainty tracking, and conservative metric rules.
Browse another skill category
- Software engineering agent skills
- Growth & SEO agent skills
- Content creation agent skills
- Product & UX agent skills
Or return to the complete source-backed agent skill library.