Overview

What is a Valdr pack?

A Valdr pack is portable operational behavior for agents: agents, prompts, capabilities, and metadata packaged so behavior is versionable, reviewable, and reproducible.

Packs are the distribution unit for Valdr workflows. Instead of rebuilding an agent system project by project, you import a pack and get the same governed behavior in each workspace: orchestrators, workflow agents, executors, reviewers, auditors, prompts, capabilities, and the relationships between them.

Official Valdr packs live in projectviking-ai/valdr-packs. Use that repository for the current pack source, release archives, and examples you can adapt for your own workspace.

Tip

Treat packs as source control for agent behavior. They make the instructions that drive agent systems visible, portable, and reviewable.

Why packs exist

Most teams do not start with packs. They start with a useful prompt, a specialist agent, a review checklist, a shell script, or a set of repeated operating instructions. At first, that is fine.

Then the system grows. Prompts get copied into different repos. Agents behave differently across machines. Workflow instructions live in chat history. Review criteria drift. Nobody knows which version of the operating instructions is correct.

Valdr packs exist to stop that drift. A pack turns agent behavior into a real artifact your team can inspect, version, import, improve, and reuse.

Instead of asking every project to rediscover the right agent setup, you ship the setup as a pack.

What changes when you use packs

Problem	Without packs	With packs
Agent setup	Onboard agents manually, one by one	Import a pack and get a complete roster
Prompt sprawl	Instructions live across files, chats, and local configs	Prompts and capabilities are structured and discoverable
Drift	Each workspace tunes agents differently	The same pack can be reviewed and reused across workspaces
Workflow memory	People remember how the system should work	The pack records how the system is supposed to work
Change control	Updates are copied by hand	Behavior changes are made in source and regenerated
Trust	“It worked for me”	Preflight import shows what will be created, changed, or skipped

The cost of not using packs

Prompt-only systems often feel fast until they need to survive real use.

Stage	What happens	Impact
Week 1	A prompt works in one repo but fails in another	Users lose time rediscovering context
Month 2	Teams copy and edit prompts independently	Outputs become inconsistent
Month 6	Nobody knows why an agent behaves a certain way	Review and audit get harder
Year 1	Prompt sprawl makes changes risky	The agent system stops evolving

Packs short-circuit that path. They make the agent system explicit enough to improve.

What a pack gives you

A pack can include:

Agent definitions that declare handles, names, kinds, default roles, prompts, and capability bindings
Capabilities that encode reusable operating behavior for agents
Prompts that can be tagged, scored, ordered, and reused
Orchestrators that help users navigate projects, knowledge, agent design, and sprint execution
Workflow agents that plan work, review implementation, and audit session evidence
Starter agents that demonstrate language, security, CI/CD, infrastructure, documentation, and dependency-audit patterns you can run or copy
Provider and launcher references that help agents run consistently in the target workspace

That is why packs are more than a file format. They are the packaging model for reusable agent operations.

How packs fit into the Valdr workflow

Author the pack as ordinary source files.
Validate the pack so broken references and malformed definitions fail before import.
Generate a .valdr-pack.tar.gz archive.
Import the archive through the Valdr UI and review the preflight action plan.
Run agents from the imported roster.
Improve the pack source as your operating model matures.

This keeps experimentation local while still giving teams a repeatable way to share what works.