Skip to main content

Agent author guide

Compose effective agents from reusable parts — without writing Rust.

1. Write a skill​

In the UI → Skills → Create skill. Use the Monaco editor to write structured instructions:

name: Rust API Implementation
version: 1
description: How to implement backend API tasks in Rust.
applies_to: [backend, rust]
required_tools: [cargo, git]
content: |
  When implementing Rust APIs:
  - use axum
  - use serde DTOs
  - add tracing instrumentation
  - write integration tests
  - never commit secrets

Test the skill against a sample task before publishing. See Concepts: Skills.

2. Compose an agent​

Create an agent and attach your skills. Pick a model, a runtime image (e.g. nexus-agent-rust), and Kubernetes resources.

3. Choose memory scopes​

Decide what the agent may read and write:

memory_policy:
  read_scopes: [global, project, agent_private]
  write_scopes: [project, agent_private]
  auto_save: true

Remember: written memory goes to a review queue until approved. See Concepts: Memory.

4. Set permissions​

Grant only what's needed, and gate the dangerous actions:

permissions:
  can_commit: true
  can_open_pr: true
  can_merge: false
  requires_human_approval_for: [database_migration, dependency_upgrade]

5. Test, enable, iterate​

Test on a sample task (use nexus-examples), then enable. Watch success rate; clone-and-tune for new variants. Consider a blueprint once you have several similar agents.