Make intent durable
Put goals, constraints, and success signals where contributors and agents can find them.
A practical learning guide
Make repositories that keep moving.
AutoRepo.ai helps teams turn clear goals into a self-sustaining practice: work is understood, planned, built, checked, reviewed, and improved without needing a person to push every small step.
- Goals are explicit
- Changes carry evidence
- Humans keep judgment
The simple version
Autonomy is a practice, not a switch.
A self-sustaining repository does not mean unsupervised software. It means the next useful action is easy to identify, safely bounded, and backed by feedback.
Start with a written goal. Keep work small. Let tests and review inform the next move. Reserve irreversible or high-impact choices for people.
Make progress inspectable
Use focused work, checks, and decisions to show what happened and why.
Make intervention normal
Define the moments to pause, ask, redirect, or stop before they become emergencies.
The operating rhythm
One loop. Seven useful moves.
Each step leaves the repository in a better position to take the next one. Skip steps only when their answer is already clear and recorded.
Understand
Turn an outcome into a clear problem, constraints, and success signals.
Plan
Choose the smallest safe path and name the decisions that need a person.
Build
Make one bounded change at a time with the relevant context nearby.
Verify
Run checks that demonstrate the change works, not just that it looks done.
Review
Inspect the result, trade-offs, and evidence before accepting the next step.
Learn
Record what changed the plan so the repository gets easier to operate.
Continue
Promote the next clear piece of work while goals and limits still apply.
A 45-minute quickstart
Give your repository a starting rhythm.
You do not need an elaborate control system to begin. Pick one active goal and create the four files below in a place your team already uses.
- 01
Name one outcome
Write the user-facing result, how you will recognize it, and what is out of scope.
- 02
Break it into safe work
Make tasks small enough to verify and explicit about when someone must decide.
- 03
Set the proof
Choose the tests, review questions, or observations that make a task genuinely complete.
- 04
Run the next loop
After review, record what changed and promote the next clear piece of work.
Copyable repository templates
Start with four plain-text files.
These templates are intentionally small. Copy them from this page or open the versioned files in the repository and adapt them to your project.
GOAL
Open file Goal template
# Goal: <short outcome>
## Why now
<user, team, or repository need>
## Success looks like
- <observable result>
- <observable result>
## Boundaries
- Do not: <excluded work>
- Ask before: <decision or risk>
## Evidence
- <test, review, or user-facing proof>WORK
Open file Work template
# Work: <short task>
## Goal
<link or restatement of the goal>
## Scope
- Change: <paths or behavior>
- Do not change: <explicit exclusions>
## Done when
- [ ] <verification>
- [ ] <verification>
## Escalate when
- <a decision, risk, or blocked dependency appears>DECISION
Open file Decision template
# Decision: <title>
**Status:** proposed | accepted | superseded
**Date:** YYYY-MM-DD
## Context
<what requires a decision>
## Decision
<what we will do and why>
## Consequences
- Positive: <effect>
- Trade-off: <effect>
- Revisit when: <condition>INSTRUCTIONS
Open file Agent instructions
# Agent instructions
## Purpose
Move this repository toward its stated goals with small, reviewable changes.
## Before changing anything
1. Read the relevant goal, work item, and local instructions.
2. State the intended change and the checks you will run.
3. Stop for missing context, conflicting instructions, or high-impact decisions.
## While working
- Prefer the smallest reversible change.
- Keep documentation and tests aligned with behavior.
- Record assumptions and unresolved risks.
## Deployment and approval boundaries
- Routine, reversible preview or staging deployments may continue when repository policy explicitly permits them.
- Ask for human approval before a production or public release, an irreversible or destructive action, an access or credential change, spending, a legal or security posture change, or an unclear material trade-off.
- Never override a failing check or invent evidence.Starting points
Three prompts for the work in front of you.
Choose the prompt that helps you clarify the next useful move. They are not a role curriculum or a prescribed sequence.
Frame the outcome
Name the changeWrite the user value, constraints, and the signal that says the work helped.
Start hereBuild the feedback loop
Make proof easyChoose small changes and checks that show what happened without relying on memory.
Start hereMaintain the rhythm
Keep the next step clearRecord decisions and intervention points so useful work can continue.
Start hereSafety and intervention
A useful system knows when to stop.
Automation can propose, prepare, build, and verify within clear limits. Routine, reversible preview or staging deployments may continue when repository policy explicitly permits them; high-impact decisions stay with people.
When the work is clear
Scope is bounded, checks are available, and the result is reversible or easy to review. Routine preview or staging deployments may continue when policy permits them.
When judgment is needed
Ask for human approval when requirements conflict, a material trade-off is unclear, or a dependency makes the plan uncertain.
Before high-impact action
Production or public releases, destructive actions, access or credential changes, spending, and legal or security posture changes require explicit human approval.