Valdr Agent Packs
Ship agent workflows you can trust — in days, not weeks
Ad-hoc prompts crumble once you have more than a couple agents — reviews conflict, permissions drift, and no one remembers why decisions were made. Valdr packs give you a portable, governed unit of agents — prompts, capabilities, and context — so you can reuse, share, and evolve workflows without starting over.
Tip
Repeatable + auditable: treat agents like productized infrastructure, not improvisation.
Note
Recommended source locations for pack authoring are ~/.valdr/valdr-packs/<pack-name> for user-wide packs or ./.valdr/valdr-packs/<pack-name> for repo-local packs. Nikol defaults to checking those directories when helping you discover, validate, or generate packs.
Valdr packs are portable bundles defined by YAML manifests and annotated Markdown. A pack.yaml declares the pack, .agent.yaml files define each agent, and <!--<capability>--> annotations make prompts machine-discoverable. Import a pack through the UI and get a complete agent roster — orchestrators, reviewers, planners, executors, and auditors — ready to work.
What’s inside a pack
| Layer | Format | Purpose |
|---|---|---|
| Pack manifest | pack.yaml | Pack identity, version, included directories |
| Agent definitions | .agent.yaml | Agent identity, kind, capability bindings |
| Capabilities | Annotated Markdown | Structured agent behavior with role-based loading |
| Prompts | Annotated Markdown | Reusable instructions with tags and metadata |
Why not “just prompts”?
| Just prompts | Valdr packs |
|---|---|
| Hard to reuse | YAML specs make packs portable and importable |
| No governance | Capabilities and permissions are explicit |
| Drift over time | Content-hash comparison detects changes |
| People-dependent | System-enforced structure |
| “Trust me, it worked” | Multi-dimensional scoring with evidence |
| Fine for demos | Built for real systems |
Who this is for
- Engineering leads tired of agents that work in dev but break in prod
- Platform teams distributing governed AI capabilities across the org
- Compliance-conscious orgs that need auditable decision trails
- Teams scaling past 3 agents who’ve felt the pain of prompt sprawl
If you’re experimenting with a single agent, you may not need packs yet. But once you’ve said “we need this to work the same way every time” — you’re ready.
The Valdr core pack
The Valdr core pack ships with 26 agents, 156 capabilities, and 168 prompts — including named orchestrators (Gunnar, Nikol, Skadi), specialists (Sigrid, Freya, Tyr), and task executors for TypeScript, Go, Java, Python, Rust, and more. See Included agents for the full roster.
Use this section to:
- Understand pack structure and YAML specs
- Learn the tag system for scoring, versioning, and model targeting
- Build team-specific capabilities that encode your institutional knowledge
- Validate and generate archives from the source pack directories Nikol already knows how to find
- Author or adapt packs with the authoring workflow
- Import packs through the Valdr UI
Product view
For the platform-level story behind packs and capabilities, read Team Capabilities for AI Agents. It explains how packs turn repeated team instructions into reusable operational behavior for agents.