Open Registry · WardPack Format v1

The Universal Registry for Intelligent Agents

Write once. Deploy everywhere. Never lose what you've learned. Pull pre-configured agents, skills, and pipelines. Deploy them across any LLM provider.

$ ward pull research-agent@latest
ClaudeGPT-4oKimiGeminiMistral

◆ Official Agents · wardscribe.io/agents

research-agentv2.4.1

0.91
Confidence

Deep research, synthesis and citation across multiple sources.

researchweb-searchsynthesis+1
48.2k pulls

code-review-agentv1.8.0

0.91
Confidence

Static analysis, style checking, PR comment writer.

code-reviewanalysisgit
31.7k pulls

data-analyst-agentv3.1.0

0.89
Confidence

SQL, statistical analysis, chart generation, insights.

datasqlanalysis
27.1k pulls

Section 04 — Usage Modes

1
Mode 01

Pull & Deploy

Analogous to: docker run

Find in registry → select LLM + project → live in <10s. Runtime verifies signature, resolves dependencies, compiles LLM profile, loads confidence baseline.

$ ward deploy research-agent@2.4.1 \ --project=my-project \ --llm=claude
2
Mode 02

Fork & Customize

Analogous to: FROM base-image

Pull any agent as a template. Modify persona, swap skills, tune LLM profiles. Save as your own agent. Lineage recorded. Inherits parent confidence tensor as baseline.

$ ward fork research-agent@2.4.1 \ --name=my-researcher # edit persona.yaml, add skills $ ward publish ./my-researcher
3
Mode 03

Compose Pipelines

Analogous to: docker compose

Select multiple agents, define collaboration: sequential, parallel, branching, human checkpoints. The pipeline is itself a publishable registry object.

$ ward compose ./pipeline.yaml \ --project=my-project $ ward pipeline run \ content-production