MasihMoafi
UserTested markdown rules, constraints, and guidelines for steering AI coding agents (Claude Code, Codex)
Categories
Indexed Skills (9)
clean-up
Procedural guide for auditing and safely reclaiming disk space on the system. Covers package manager caches (uv, pip, npm, conda, pnpm), Docker caches/images, transient agent files (.ES), and temp/backup files. Never deletes without explicit user confirmation.
doc-style
Enforces a brutalist, evidence-first documentation and launch-writing standard — structure, tone, and distribution outputs, with no unproven claims. Use when writing or reviewing a README.md, SKILL.md, specification, or launch material.
experiment-workflow
Canonical Experiment & Testing Workflow. Use to rigorously evaluate systems by isolating variables, unit-testing, and running 3x10 query stress-tests.
first-principles
Executable first-principles engineering method for AI agents. Run this on every task to question requirements, delete components, simplify/optimize, accelerate cycle time, and automate.
gitlab
GitLab workflow guidelines for agents. Use when working on repositories hosted on GitLab to coordinate branches, commits, MRs, and permissions.
harness-eng
Use when designing, evaluating, or simplifying an agent project harness: AGENTS.md/CLAUDE.md rules, startup scripts, progress logs, feature trackers, handoffs, evaluator rubrics, quality documents, repo-local knowledge maps, and mechanical guardrails for coding agents. Especially useful when converting raw agent-workflow notes into a concise, verifiable project control layer.
Use when tasks involve reading, creating, or reviewing PDF files where rendering and layout matter; prefer visual checks by rendering pages (Poppler) and use Python tools such as `reportlab`, `pdfplumber`, and `pypdf` for generation and extraction.
product-design
Turns an early product idea into explicit user flows, screen architecture, stack choices, and a frontend-first implementation plan. Use before coding a new product, redesigning a product flow, choosing a web/app stack, or translating a product concept into an AI-assisted build plan.
training-nn
Karpathy-style recipe and common-pitfalls checklist for training neural networks from scratch. Use when designing, debugging, or reviewing a custom training pipeline (data loading, loss, optimizer, regularization) — especially when training silently fails to converge or underperforms with no errors thrown.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.