Skip to content

Starter agents

Starter agents are examples you can actually use. They are not meant to be a perfect fit for every codebase on day one; they are pack-ready agent shapes that show how Valdr turns a role, a handle, a core system capability, and hot-loaded domain guidance into runnable behavior.

Use them in two ways:

  • Run them as built-in executors or reviewers when the default behavior matches the task.
  • Copy the pattern when you want to build a team-specific Java, Go, security, CI/CD, infrastructure, or documentation agent.

Tip

Treat starter agents like project templates. Keep the structure, then replace generic capability guidance with your repository conventions, build commands, review gates, and operational standards.

What ships as starter agents

The Valdr Vanguard pack includes language-focused task agents and utility agents that demonstrate the most common extension points.

AgentHandleDefault roleUse when
Java Task Agentjava-task-agentexecutorJava 17+, Maven/Gradle, package design, JUnit validation
Go Task Agentgo-task-agentexecutorGo modules, idiomatic package boundaries, error handling, table tests
Python Task Agentpython-task-agentexecutorPython typing, package tooling, pytest validation
Rust Task Agentrust-task-agentexecutorCargo workspaces, ownership, traits, property-oriented testing
CI/CD Pipeline Agentcicd-agentexecutorGitHub Actions, deployment workflows, quality gates
Infrastructure-as-Code Agentinfra-agentexecutorTerraform/OpenTofu, containers, cloud, networking, observability
Security Audit Agentsecurity-audit-agentreviewerVulnerability review, dependency audit, secrets checks, control-alignment notes
Refactoring Scoutrefactor-scoutplannerCode health discovery and refactoring task shaping
Hugo Documentation Stewardhugo-docs-agentplannerDocs drift audits, source-truth checks, screenshot and shortcode upkeep
Dependency Audit Scoutdependency-audit-scoutplannerMulti-ecosystem dependency reachability, unused dependency, and upgrade planning

How starter agents are built

Every starter agent follows the same pack pattern:

  1. A stable .agent.yaml file declares the handle, name, kind, default role, tags, and capability bindings.
  2. Exactly one role: core capability defines the agent identity and baseline workflow.
  3. Domain capabilities attach deeper guidance as constraints, workflow, validation, context, or integration.
  4. Most domain capabilities are hot-load: true, so the agent starts lean and loads deeper instructions only when the task needs them.
  5. The same agent can be reused across workspaces because its behavior is source-controlled pack content.

That structure is the important part. The built-in Java and Go agents both use four hot-loaded domain capabilities (build, code.rules, design.solid, and testing). The Security Audit Agent uses scanning and policy capabilities (vulnerability-patterns, dependency-audit, secrets-detection, and compliance). Your own agents can use the same model with names and capabilities that match your stack.

When to use starter agents

Use a starter agent when the task already has a clear implementation or review target:

  • /valdr-agent java-task-agent implement this JVM bug fix and run the affected Maven tests.
  • /valdr-agent go-task-agent implement the module boundary cleanup and run package tests.
  • /valdr-agent security-audit-agent review this auth change for exploitable paths.

Do not use starter agents as general command centers. For navigation and task creation, start with Gunnar. For pack design, use Nikol. For sprint staffing and launch readiness, use Skadi.

Starter agent docs

Start with one of the examples, then adapt the structure in Build your own when your codebase needs stricter conventions than the generic starter pack can safely assume.