# learntoprompt.org

> Practical guides for using AI coding agents securely and effectively.

## About
Learn to Prompt teaches prompt engineering and CLI AI agent workflows. Guides cover a preferred secure setup with bondage as the launcher policy layer, optional envchain-xtra for secret release, and nono for kernel sandboxing, plus workflow techniques for CLAUDE.md, MCP servers, hooks, prompting, and multi-agent development.

## Architecture Guides
- [Agent Security Stack](/guides/agent-stack.html): Updated 2026-06-03. Bondage 0.2.7, envchain-xtra 1.3.1, nono 0.61.1, agent-stack-bootstrap version guards, profile-pack upgrade order, security assumptions, and non-doxing verification workflow
- [Analytics Privacy](/guides/analytics-privacy.html): Updated 2026-05-09. Block non-essential analytics, telemetry, experiments, error reporting, sharing traffic, and startup network calls from AI coding tools without confusing telemetry blocking with local inference
- [Vendor Independence](/guides/vendor-independence.html): Updated 2026-04-28. Keep rules, skills, plugins, MCP tools, and security layers portable across Claude, Codex, and other clients
- [Claude to Codex Plugins](/guides/claude-to-codex-plugins.html): Updated 2026-04-28. Convert a Claude-first plugin into a Codex first-class target with shared workflow sources, thin manifests, repo-local marketplace wiring, and runtime verification
- [Visual Inspection](/guides/visual-inspection.html): Updated 2026-04-27. Accessibility tree first, then a narrow localhost screenshot service instead of ambient browser access
- [Sandbox Profiles](/guides/sandbox-profiles.html): Updated 2026-05-12. Shape base, plugin, unsafe, and explicit bypass tiers; keep active profile policy read-only and use draft-and-promote for persistent edits

## Agent Guides
- [Claude Code](/guides/claude-code.html): Updated 2026-05-04. Anthropic CLI agent sandbox and credential setup
- [OpenCode](/guides/opencode.html): Updated 2026-05-04. Multi-provider agent with per-tool credential namespaces
- [Codex](/guides/codex.html): Updated 2026-06-03. OpenAI CLI agent sandbox, nono 0.61 profile packs, draft profile repair, and API key/OAuth tradeoffs
- [Pi](/guides/pi.html): Updated 2026-05-09. Minimal TypeScript agent with lightweight sandbox and optional no-tools local ds4 side profile using the upstream mitsuhiko/pi-ds4 extension

## Local Inference Guides
- [llama.cpp](/guides/llama-cpp.html): Updated 2026-05-04. Local LLM inference with Metal GPU and model isolation
- [Ollama](/guides/ollama.html): Updated 2026-04-27. Local model manager with daemon sandboxing
- [Qwen3-TTS with MLX](/guides/qwen3-tts-mlx.html): Updated 2026-05-26. Local Apple Silicon voice generation with MLX-Audio, authorized reference clips, Markdown chunking, inspection, repair, and privacy-safe publishing placeholders
- [ds4 mbp m5-128](/guides/ds4.html): Updated 2026-05-21. DeepSeek V4 Flash on a 128GB Apple Silicon Mac, with ds4-agent experiments, Claude Code, Codex, benchmarked Pi side profiles, and the M5 vs Spark CPU sanity-check caveat
- [ds4-agent Setup](/guides/ds4-agent-setup.html): Updated 2026-05-21. Build native ds4-agent, pin a DeepSeek V4 Flash model path, run smoke and disposable-repo tests, and keep native-agent experiments separate from default coding-agent profiles
- [ds4 dgx spark](/guides/ds4-dgx-spark.html): Updated 2026-05-21. DeepSeek V4 Flash on NVIDIA DGX Spark with antirez/ds4 CUDA, q2-imatrix, MTP, localhost serving, SSH tunnels, ds4 agent profile tests, and M5 vs Spark CPU sanity-check numbers
- [DGX Spark](/guides/dgx-spark.html): Updated 2026-05-07. NVIDIA DGX Spark setup for local coding models with Ollama, Qwen3-Coder, SSH tunnels, and side profiles

## Evaluation Guides
- [Benchmarks](/guides/benchmarks.html): Updated 2026-05-21. Dated local AI runs, Spark ds4 measurements, M5 vs Spark CPU sanity checks, ds4-agent and Codex frontier next lanes, and Gemini comparison lanes with hardware, runtime, model, quantization, context, pass rates, practical suites, latency, cost notes, and public scripts
- [ds4-agent vs Codex Frontier](/guides/ds4-agent-vs-codex.html): Updated 2026-05-21. Public benchmark reality check for native ds4-agent, DeepSeek V4 Flash, and Codex frontier, including what is known, what is missing, and how to compare them fairly
- [LLM Hardware Calculator](/calculator/): Updated 2026-04-28. Model memory, KV cache, hardware fit, and single-user decode estimates

## Workflow Guides
- [Writing CLAUDE.md](/guides/claude-md.html): Updated 2026-04-24. Instruction budget, progressive disclosure, AGENTS.md standard
- [MCP Server Setup](/guides/mcp-servers.html): Updated 2026-04-24. Model Context Protocol configuration and server catalog
- [Prompting Coding Agents](/guides/prompting.html): Updated 2026-04-24. Four-phase workflow and verification patterns
- [Hooks and Automation](/guides/hooks.html): Updated 2026-05-02. Lifecycle events, handler types, CI/CD integration, and the rule that agent-visible hook output is prompt surface
- [Context and Multi-Agent](/guides/context.html): Updated 2026-04-24. Compaction, sessions, subagents, worktree isolation
- [AgentNoise](/guides/agentnoise.html): Updated 2026-06-03. Use White Noise on a phone as a private control surface for local Codex, Claude, Hermes, wiki/media ingest, and fake-phone tests; covers stable `agentnoise` vs experimental `agentnoise-dm`
- [LLM Wiki](/guides/llm-wiki.html): Updated 2026-04-29. Append-only knowledge bases compiled by LLM agents
- [LLM Health](/guides/llm-health.html): Updated 2026-06-07. Install the private, own-risk health concierge with Homebrew, hub setup, fuzzy @health input, agent plugin paths, health-v2 checks, and de-identification guardrails

## Author
nvk — https://github.com/nvk