LearnPrompt AI Practice Wiki

LearnPrompt is not a glossary of tool terms, nor a textbook you read cover to cover. It’s a practice-first tutorial series for real projects: you pick a task that’s already happening, then take the shortest path to a result you can actually check.

First time here? Start from your task

The problem you’re solving right now	Start here	Your first deliverable
You haven’t had AI actually change your project yet	The minimum agentic coding workflow	A small change that includes a plan, a diff, checks, and a retro
You’re already using Claude Code or Codex but aren’t sure which to reach for	Choosing between Claude Code and Codex	A decision table based on environment, feedback, risk, and async needs
Your agent keeps missing context, overstepping scope, or finishing work you can’t verify	The five components of a harness	A check across instructions, capabilities, constraints, state, and orchestration
You do the same workflow over and over and want an agent to reuse it reliably	What an Agent Skill actually is	A clear call: “keep using a prompt” or “upgrade to a Skill”
You want project knowledge to survive across sessions and agents	When Markdown counts as agent-handoff memory	A handoff packet a fresh agent can pick up and continue
You’re building on Hermes or OpenClaw	Hermes’s learning loop or The OpenClaw architecture guide	An observable write-approval flow or message-routing diagram

Still not sure where to start? See the AI Practice Map. It turns “I want to learn AI” into an actionable learning path.

How every tutorial proves it’s useful

An in-depth tutorial never gives you just concepts and screenshots. You’ll find four kinds of evidence:

1. A concrete reader task: what you’ll be able to do after reading, not just “understand a feature.”
2. A replayable showcase: valid scenarios alongside counter-examples that get rejected.
3. Executable verification: commands, exit codes, diffs, receipts, or manual checkpoints.
4. Real sources with clear limits: primary sources, secondary topic maps, and field experiments, each labeled separately.

verified only means an article has passed the project’s own research, showcase, independent review, diagram, and build gates — it doesn’t mean the tool’s capabilities will never change. For anything involving versions, pricing, permissions, or security, always trace back to the official source linked at the bottom of the article.

The shortest way to use this wiki

Pick one real task
  → Read only the 1–3 articles needed to unblock it
  → Copy the template and adapt it to your own input
  → Run the command or do the manual check
  → Write what failed back into your project rules, memory, or a Skill

If you have 15 minutes, finish the exercise in one tutorial. If you have a week, run the full loop with the 7-Day Practice Path. Don’t mistake “I bookmarked it” for “I can use it.”

Set boundaries before you start

Before you let an agent write files, run commands, or reach external systems for the first time, read Safety basics for AI tool setup. When secrets, deletions, releases, paid calls, or real user data are involved, no tutorial can make the authorization call for you.

Sources and maintenance

• How to use this wiki: reading status, evidence tiers, and learning receipts.
• Task, context, and verification: freezing a vague request into a contract an agent can execute.
• Source Index: primary sources, secondary topic maps, and licensing boundaries by section.