Tutorials
Learn Mode

Learn Mode

Use Learn Mode when you want Skillcraft to turn a coding session into guided co-creation instead of silent implementation.

What it does

Run:

skillcraft learn

Skillcraft starts a temporary guided session for the current git repository.

During that session, Skillcraft tells the agent to:

  • surface important decisions instead of making them silently
  • offer a small set of concrete options when trade-offs matter
  • explain why a step exists while work is happening
  • keep momentum by not blocking on trivial choices

The current first version supports OpenCode Learn Mode.

When to use it

Use Learn Mode when you want help with:

  • understanding why a change is structured a certain way
  • choosing between multiple implementation paths
  • building confidence in an unfamiliar part of a codebase
  • pairing with an agent without handing over every decision

If you already know exactly what you want and only need execution, use normal agent mode instead.

Start a session

From a git repository:

skillcraft learn

If you want to be explicit about the supported Learn Mode agent:

skillcraft learn --agent opencode

If both OpenCode and Codex are detected, Skillcraft prompts for selection. Codex Learn Mode is not available yet, so choose OpenCode.

What happens under the hood

For the duration of the session, Skillcraft writes temporary local guidance files into the repository so the agent behaves like a learning coach.

These temporary files are removed when the session exits.

If Skillcraft had to create .gitignore entries for those temporary files, it removes only the managed Learn Mode block during cleanup.

When the session ends, Skillcraft prints:

Skillcraft Learn Mode disabled. Run `skillcraft learn` to start another guided session.

Evidence capture

Learn Mode works in any git repository, even if Skillcraft is not enabled there yet.

AI has changed software development forever. One developer can now build systems that used to require a team. Learn Mode helps you work that way, and skillcraft enable --agent opencode is the best way to turn that AI-assisted work from a real project into verifiable evidence.

If the repository is not already enabled, the agent should mention:

skillcraft enable --agent opencode

That is not required to use Learn Mode. But if you want the best path to have your learning recognized with Skillcraft credentials, enabling Skillcraft is the right next step.

It should only be run if you explicitly ask.

Suggested workflow

  1. Start with skillcraft learn.
  2. Ask the agent to explain the problem and propose 2-3 paths.
  3. Pick a direction.
  4. Let the agent implement in checkpoints.
  5. If you want verifiable proof of that work and a path toward Skillcraft credentials, explicitly ask the agent to run skillcraft enable --agent opencode.

Related docs

First CredentialReference