Start on your own laptop
Your next attempt
starts here.
Get better at working with AI. Choose a real problem, sharpen your prompt and make a deliberate next attempt in Claude Code or Codex.
First time? Install the local skill.
Follow the source-install guide for your host, then copy a prompt below. Private coaching needs no Better Loop account.
Install from the main branch using the source-install guide. The packages are available from source; there is no npm registry release.
One prompt. One clear next step.
Make it your loop.
Start where your work is.
Open your chosen repository in your host. The skill helps you pick a useful task and a dedicated local checkpoint.
Fill any placeholders in Codex. Keep local paths, task text and evidence there. This website does not ask for them.
Give the next attempt a clear purpose.
Copying does not run a command or send work to a model.
What should I get back?
A concrete experiment and acceptance check, plus a local checkpoint you can recall. A first assessment establishes a baseline; it does not demonstrate improvement. Missing measurements stay unknown.
Shape how you work with AI
A prompt and a working agreement you control.
Start with explicit preferences for your audience, feedback, delegation and verification. The local helper prepares a prompt and a scoped Markdown diff for your chosen project. Your task’s exact output requirements take precedence.
Selected task context: general. Make the prompt and acceptance check fit your work.
Your chosen challenge
Change one habit and check the next attempt
What to check on your next attempt
Freeze the outcome check and quality floor before trying the change. Compare a later compatible attempt, include known orchestration overhead, and keep neutral, adverse and missing findings. Lower token use alone is not better quality.
Save the copied prompt and check in a new local text file. Adapt them to your selected task, then use that file in place of EXACT_PROMPT.txt in the plan command below. Copying does not save or run a file.
1. Choose your preferences
Save the neutral starter in your chosen task directory. Edit it locally; keep personal context and project details off this website.
Download the preference starter2. Review the proposal
Choose your exact directory, preference file, prompt and a new private plan path. Inspect the proposed prompt, acceptance check, target and complete diff before applying.
3. Check a next attempt
Applying an agreement is preparation. Record a later human decision and checked outcome in your existing local checkpoint. Keep neutral and adverse findings too.
Replace the capitalized placeholders with explicit local paths; quote paths containing spaces. Use the installed source CLI from the pinned install guide. Planning makes no model or network call and does not edit AGENTS.md. The deterministic template is not a semantic diagnosis.
Apply only the reviewed diff, or roll it back
The plan binds the chosen root, host, input bytes and exact AGENTS.md change. Changed scope, inputs or target bytes require a new plan and approval. Only the managed project section is changed; unrelated Markdown is preserved.
Rollback checks the original root and installed bytes and refuses to overwrite later edits. It does not require unchanged preferences. Explicit model profiles describe preferences; they never switch or train a model. This website cannot read, apply or upload your files.
Bring the lesson back
Keep your place.
See what changed.
Save your checkpoint on your laptop. Next time, use “Return to my saved checkpoint” with that explicit path. Your before/after work and practice history stay local; a new chat does not have to remember you.
Raw work stays out of Better Loop. Your configured model provider may process the evidence you select. Share only a minimized story after reviewing every field and purpose.
Already have a prepared story? Preview it →