Agent Pack

Published specification, JSON Schema, and validator for Agent Packs.

An Agent Pack is a versioned, deployable unit of agent — its prompt, its tools, its engine, and its permissions.
It’s what your IDE writes, your CI validates, and your infrastructure runs.

Canonical HTML	https://packs.agenthippo.ai/spec/v1
Schema $id	https://packs.agenthippo.ai/schema/v1/agent.json
Pitch	https://agenthippo.ai/why-agent-pack
License	CC BY 4.0 (spec text + schema)
Validator	npx @agenthippo-ai/pack-validate · Action agenthippoai/pack

This repository is the machine-readable / citable source for the format. The website hosts the same schema and a rendered spec page. The CLI and GitHub Action validate against schema/v1/agent.json in this repo.

What is an Agent Pack

Most agent code never leaves a chat window or a laptop. Once it has to run against real company data, for more than one person, someone asks: which version is this, what can it touch, and whose permissions is it using?

An Agent Pack answers that as a versioned artifact: prompt, skills, tools, engine, and permissions — small enough to review and roll back like any other production change. This repo defines the manifest (agent.yaml) and schema. Deploy targets, runtimes, and infrastructure are generated from a pack by tooling, not declared in it.

Full write-up: packs.agenthippo.ai/spec/v1 · SPEC.md

Example

Read-only warehouse analyst with per-user (OBO) SQL access:

# yaml-language-server: $schema=https://packs.agenthippo.ai/schema/v1/agent.json
apiVersion: agenthippo.ai/v1
kind: Agent
metadata:
  name: warehouse-analyst
  version: 0.1.0
  author: AgentHippo
  description: Read-only warehouse analyst — answers questions against the warehouse via per-user (OBO) SQL access.
spec:
  engine: claude
  engineVersion: 2.1.209
  model: claude-fable-5
  permissions:
    fileAccess: read-only
  mcp:
    - name: databricks
      config: ./mcp/databricks.mcp.json
  skills:
    - path: ./skills/sql-query

Only agent.yaml is required by the schema. The pack prompt is always AGENTS.md at the pack root — a fixed path, not a manifest field (same idea as Dockerfile). Companion dirs skills/ and mcp/ are declared via spec.skills / spec.mcp because those can be many. Subagents should be other Agent Packs (own engine, model, tools), not nested mini-configs. Details: SPEC.md · Anatomy.

Layout

SPEC.md                 # Agent Pack Specification v1 (markdown)
schema/v1/agent.json    # JSON Schema (identical to packs.agenthippo.ai)
bin/pack-validate.mjs   # CLI entry
src/validate.mjs        # Ajv2020 + js-yaml
action.yml              # GitHub Action (thin CLI wrapper)
package.json            # @agenthippo-ai/pack-validate
LICENSE                 # CC BY 4.0

Validate

# current directory (./agent.yaml) or a pack / store packs dir
npx @agenthippo-ai/pack-validate
npx @agenthippo-ai/pack-validate ./path/to/agent.yaml
npx @agenthippo-ai/pack-validate ./agent-packs

GitHub Actions:

- uses: agenthippoai/pack@v0.1.0
  with:
    path: .

Editor validation

In agent.yaml:

# yaml-language-server: $schema=https://packs.agenthippo.ai/schema/v1/agent.json

Contributing

Open issues for clarifications and RFCs.