AgentSpec Lab: AI Agent Manifest Builder

Draft a portable YAML or JSON manifest for an AI agent's identity, capabilities, tools, and human-approval guardrails. Experimental and processed locally.

Validation build: AgentSpec 0.1 is an ASAPUtils experiment, not a published standard. We are testing whether a portable, human-readable agent manifest solves a real workflow.

Describe the agent

Use non-secret examples. This experimental format is not yet an industry standard.

Agent namePurposeModel referenceCapabilities (comma-separated)Tool names (comma-separated) Require human approval before external actions

Portable manifest draft

agentspec: 0.1-experimental
agent:
  name: support-triage-agent
  description: Classifies support requests and drafts a response for human review.
  model: provider/model-name
capabilities:
  - classify tickets
  - retrieve documentation
  - draft responses
tools:
  - name: knowledge_search
  - name: ticket_lookup
guardrails:
  human_approval_required: true

100% in your browser — nothing is uploaded.

What this experiment is testing

Agent definitions are often scattered across prompts, code, tool configuration, and deployment settings. This lab tests whether teams want one reviewable manifest for the agent’s purpose, model reference, declared capabilities, tools, and approval boundaries.

What comes next only if people return

Repeat browser usage will guide whether AgentSpec should become a schema validator, CLI, npm package, GitHub Action, embeddable editor, or browser extension. Keeping the experiment at asaputils.com/labs/agentspec lets it earn those distribution surfaces before becoming a separate product.

Tell us where an agent manifest would fit in your workflow →