Free instruction workspace

Project & Agent Instructions Builder

Choose an AI platform and a job. Create clear instructions for coding agents, business workflows, or the Microsoft ecosystem across Copilot Studio, Power BI, Power Automate, Power Apps, Excel, Word, and PowerPoint.

Choose where it will run.

The builder changes the file name and setup steps for the selected platform.

Define the project once.

Stable facts, boundaries, sources, commands, and evidence belong in project instructions.

Give the agent one job.

A focused assignment defines the workflow, deliverables, checks, and response format.

Private browser workspace

Build your instructions

Your entries stay on this page. Nothing is sent to an AI provider or AEVUS.

1. AI platform

2. Agent job

3. Project contextProject or team nameGoal and usersApproved sources of truthPermissions and boundariesCommands or workflow cadenceDefinition of done

Setup for OpenAI Codex

AGENTS.md

Save the project instructions as AGENTS.md in the repository root. Paste the agent assignment into a new Codex task.

Codex automatically discovers AGENTS.md files from the repository hierarchy.

Always-on context

Project instructions

# [Add project name] — Project Instructions

## Purpose
[Add the project goal and the people it serves]

## Sources of truth
[Add approved repositories, documents, mailboxes, databases, or work systems]

## Operating rules
- Read the applicable project instructions, source material, and existing patterns before acting.
- Confirm identity, workspace, environment, permissions, and current state before any consequential action.
- Use the smallest change or workflow that satisfies the approved goal. Preserve unrelated behavior and data.
- Distinguish proposed work, local validation, deployed behavior, and human-approved completion.
- Do not invent facts, owners, deadlines, statuses, customer data, test results, or approvals.
- Keep credentials, private personal data, confidential material, and regulated information out of prompts, logs, screenshots, and reports unless explicitly approved and protected.
- Prepare drafts for review. Require human confirmation before sending external messages, changing production, deleting data, spending money, publishing, merging, or writing back to another system.

## Project-specific boundaries
[Add actions the AI may take, actions requiring approval, and actions it must never take]

## Approved commands or operating cadence
[Add build, test, lint, reporting, or workflow commands and when to use them]

## Definition of done
[Add observable acceptance criteria, evidence, persistence checks, and receiving-side results]

## Reporting
- Lead with the result and its practical impact.
- Report changed files or records, validation performed, environment, remaining risks, and blocked proof separately.
- Cite source IDs, file locations, message links, test artifacts, or system records so another person can verify the conclusion.

Focused task

Agent instructions

# Agent Assignment — Code implementation agent

## Role
Act as a software implementation agent.

## Objective
Use the project instructions and approved sources to complete this bounded assignment: [Add describe the requested outcome].

## Workflow
- Read repository instructions, architecture, contracts, and relevant tests before editing.
- Restate the bounded goal, affected surfaces, exclusions, and acceptance criteria.
- Make the smallest coherent change using existing patterns and components.
- Run focused checks, inspect the diff, and report remaining runtime validation.

## Required deliverables
- Implementation plan
- Working code and necessary tests
- Validation evidence
- Change summary and remaining risks

## Acceptance checks
- Authorization precedes UI exposure.
- No secret, unrelated file, or protected environment is changed.
- Passing tests are not described as production proof.

## Response format
1. **Result:** What is ready or what was found.
2. **Evidence:** Sources, records, files, environments, or test artifacts used.
3. **Actions prepared or completed:** Clearly distinguish drafts from executed actions.
4. **Decisions needed:** Questions and approval points that need a person.
5. **Remaining risk:** Failed, blocked, untested, stale, or conflicting items.

If required context is missing, continue safe read-only preparation and ask one concise question that unlocks the next dependent step. Never treat elapsed time or silence as approval.

Ready to copy or download. Replace bracketed placeholders before using the instructions.

Launch Studio member guide

Connect the full technology stack.

Move from instructions to a maintained setup for Git and GitHub, PostgreSQL and DBeaver, Supabase or another managed database, Render, Resend, Stripe, and OpenAI. The member workspace turns your confirmed architecture into a saved, versioned implementation plan.

Explore Tech Stack Setup →