Magic Context

A reusable library of prompt templates, skill definitions, and reference knowledge for AI-assisted software engineering. Designed to be embedded as a submodule and consumed by AI coding agents (Claude Code, Codex, Copilot, etc.).

Structure

instructions/

Standalone prompt templates for common development tasks. Drop into a conversation or reference directly.

• Code quality: write-code.md, check-code.md, review-code-*.md, strongly-typed.md, validate-types.md
• Debugging: debug/debug-code.md, debug/debug-deeply-py.md
• Documentation: document-python-code.md, explain/, revise-doc.md
• Testing: make-unit-test-python.md, gui-test-screenshots.md, headless-testing.md, nicegui-interactive-test.md
• Planning: planning/ — roadmaps, task breakdowns, refactor plans, open-question discussions
• Project setup: dir-setup/ — Python project init, experiment dirs, context dirs
• Research/reading: reading/ — paper conversion, survey, algorithm blocks, figure insertion
• Environment: envs/ — Claude Code, Codex, Copilot setup guides; linux-env.md, mac-env.md, win32-env.md
• Memory/context: make-memory.md, mem-as-*.md
• Web research: find-online.md, check-online.md, search-proactively.md, tavily-guide.md

knowledge-base/

Reference documentation for specific tools and technologies.

• blender-plugin/ — Blender addon development (background mode, data persistence, testing)
• general/ — Project profiles, Mermaid CLI, LiteLLM proxy, APT mirrors, DNN timing
• mcp-dev/ — MCP server development and programmatic invocation
• github/ — Secure CI secrets management
• tensorrt/ — TensorRT setup with Pixi, ONNX QDQ → INT8 conversion
• onnx/ — Parallel ONNX inference
• ngc-dockers/ — NVIDIA NGC Docker environments, SSH setup

skills/

Self-contained skill definitions consumed by AI agents. Each skill has SKILL.md, agent prompts, reference materials, and templates.

• CLI agents: Claude Code install/invoke, Codex CLI, Copilot
• Air-gapped: Offline Docker/Pixi projects, offline VSCode installer
• Development: Git worktrees, interactive testing, Pixi project init, MkDocs serving
• Tools/Pixi: NVIDIA CUDA setup, CUDA build env, offline package channels
• Tools/Conan: C++ package management basics
• OpenSpec extensions: explain, test-driven dev, respond-to-review, revise-by-decision
• Research: DNN model exploration (experiment report, stakeholder report templates)
• Writing: Program tutorial creation, Mermaid diagram generation
• Libraries: DeepFace basic usage

scripts/

Utility scripts for document conversion and environment setup.

• convert-html-to-markdown.py — HTML → Markdown with image handling
• convert-pdf-to-markdown.py — PDF → Markdown with LLM image descriptions
• download-html.py — HTML downloading via Playwright
• extract_references.py — Reference extraction from papers
• configure-submodule-https-ssh.sh — Git submodule HTTPS/SSH switching
• setup-envs.sh / setup-envs.ps1 — Environment initialization
• setup-proxy.sh, setup-ssh-reverse-tunnel.sh — Network setup

configs/

Configuration file examples for AI agents and tools (Codex, tmux).

roles/

AI persona definitions. Currently: academic_survey_agent.md — a structured persona for literature surveys with systematic search, synthesis, and academic integrity rules.

spec-driven-dev/

OpenSpec integration tools and specification kits for Python and C++ projects.

Usage

This repo is intended to be used as a git submodule. Reference files directly by path when constructing AI prompts, or copy templates into your project's context/ directory.

# Add as submodule
git submodule add <repo-url> magic-context

# Update to latest
git submodule update --remote magic-context