OPS-1.0 · Open Standard

Portfolio Format Specification

A Portfolio is a downloadable, versioned, tested AI agent configuration package. It is the atomic product unit of the AgentFolio marketplace.

Version 1.0 · Published 2026-04-10 · Open standard — no IP claims

Design Goals

✓

Complete

Everything needed to deploy and run the agent is included in one package

✓

Verifiable

Test cases prove the agent works before purchase

✓

Portable

Clear platform targeting with migration guides to other platforms

✓

Versioned

Each release has a changelog so buyers can update

Portfolio Package Structure

Every portfolio is delivered as a ZIP archive with the following structure:

{slug}/
├── manifest.json          # Required: Metadata, version, platform
├── system-prompt.md        # Required: The core system prompt
├── tools/
│   └── *.md               # Optional: Per-tool configuration docs
├── workflows/
│   └── workflow.md        # Required: Sequential workflow definition
├── test-cases/
│   └── cases.json         # Required: Test input/output pairs
├── deployment/
│   ├── README.md          # Required: Step-by-step deployment guide
│   └── platform/
│       └── *.md           # Platform-specific import files
├── examples/
│   └── *.md               # Optional: Sample inputs and expected outputs
└── changelog.md            # Required for v1.1+: What changed since last release

manifest.json Required

Metadata file with all required portfolio information.

{
  "format_version": "1.0",
  "portfolio_id": "customer-support-agent-openclaw",
  "version": "1.0.0",
  "title": "Customer Support Agent",
  "tagline": "Handles Tier-1 customer inquiries with escalation logic",
  "description": "A production-tested customer support agent that handles FAQ responses, ticket routing, and escalation to human agents. Built on OpenClaw with GPT-4o.",
  "platform": "openclaw",
  "platform_version": ">= 1.4.0",
  "category": "customer-support",
  "subcategory": "tickets-and-escalation",
  "tags": ["support", "tickets", "escalation", "faq"],
  "price_cents": 4900,
  "license": "single-buyer",
  "language": "en",
  "min_required_user_skill": "intermediate",
  "requires_api_keys": ["openai"],
  "created_at": "2026-03-15T00:00:00Z",
  "updated_at": "2026-03-15T00:00:00Z",
  "test_cases_version": "1.0.0"
}

Field Type Required Notes

format_versionstring✅Must be "1.0"

portfolio_idstring✅kebab-case, unique across marketplace

versionsemver✅"major.minor.patch"

titlestring✅Max 80 chars

taglinestring✅Max 120 chars

descriptionstring✅200–800 chars

platformenum✅openclaw, n8n, make, zapier, langchain

platform_versionsemver range✅Minimum platform version

categoryenum✅See category list below

subcategorystring✅Free-form sub-category

tagsstring[]✅3–8 tags

price_centsinteger✅Price in cents (USD)

licenseenum✅single-buyer, team (up to 5), enterprise

languageISO 639-1✅Primary language

min_required_user_skillenum✅beginner, intermediate, advanced

requires_api_keysstring[]✅List of required API keys

created_atISO 8601✅UTC timestamp

updated_atISO 8601✅UTC timestamp

test_cases_versionsemver✅Must match test file version

system-prompt.md Required

The core system prompt that defines the agent's behavior, personality, guardrails, and output format.

# System Prompt: {Title}

## Role Definition
[Who the agent is, its authority level, what it can and cannot do]

## Behavioral Rules
[3–8 specific rules the agent must follow in every interaction]

## Output Format
[Structured output format — JSON schema or markdown template]

## Guardrails
[What the agent must refuse to do, sensitive topics, content limits]

## Escalation Triggers
[Conditions under which the agent must escalate to human]

## Tone and Style
[Writing style, formality level, example phrases]

workflows/workflow.md Required

Sequential workflow definition showing the agent's decision tree and action sequence.

# Workflow: {Title}

## Trigger Conditions
[What starts this workflow — user message, webhook, scheduled time, etc.]

## Step 1: [Name]
- Action: [What the agent does]
- Tool used: [Tool name]
- Decision point: [If X, go to Step 2a; if Y, go to Step 2b]

## Step 2: [Name]
[... continue for all steps]

## Fallback Path
[What happens if something goes wrong]

## Success Criteria
[How to know the workflow completed successfully]

test-cases/cases.json Required

Test input/output pairs that prove the agent works correctly.

{
  "test_suite_version": "1.0.0",
  "test_cases": [
    {
      "id": "tc-001",
      "name": "Handles FAQ response",
      "description": "Given a common FAQ question, the agent returns an accurate response",
      "input": {
        "user_message": "How do I reset my password?",
        "context": {}
      },
      "expected_output_type": "structured",
      "expected_output": {
        "response_type": "faq_answer",
        "confidence": "high",
        "answer": "To reset your password, go to Settings > Security > Reset Password...",
        "escalate": false
      },
      "pass_criteria": ["escalate === false", "answer contains password reset instructions"],
      "platform": "openclaw",
      "tags": ["faq", "password"]
    }
  ]
}

Field Type Required Notes

test_suite_versionsemver✅Version of this test suite

test_cases[].idstring✅Unique test case ID

test_cases[].namestring✅Human-readable test name

test_cases[].descriptionstring✅What this test validates

test_cases[].inputobject✅Test input parameters

test_cases[].expected_outputobject✅Expected output structure

test_cases[].pass_criteriastring[]✅Validation criteria strings

deployment/README.md Required

Step-by-step deployment instructions for the target platform.

# Deployment Guide: {Title}

## Prerequisites
- OpenClaw account (free tier works)
- OpenAI API key with GPT-4o access
- [Any other required accounts or keys]

## Step 1: Create a new agent
[Screenshot or step description]

## Step 2: Import the system prompt
[Copy-paste from system-prompt.md — specific instructions]

## Step 3: Configure tools
[Tool configuration, including any required API keys]

## Step 4: Import test cases
[How to load test-cases/cases.json into the platform]

## Step 5: Test your agent
[How to run the test cases to verify it works]

## Expected Setup Time
[Estimated time for a complete setup]

## Troubleshooting
[Common issues and fixes]

Platform Identifiers

Platform ID Notes

OpenClawopenclawPrimary target for MVP

n8nn8nSelf-hosted workflow automation

MakemakeVisual automation platform

ZapierzapierAutomated workflows

LangChainlangchainProgrammatic AI agent framework

Quality Bar

A portfolio listing on the marketplace must pass ALL of the following before going live:

manifest.json validates against this spec

system-prompt.md exists and is non-empty

workflows/workflow.md exists and covers all decision paths

test-cases/cases.json has ≥3 test cases

deployment/README.md exists with ≥4 steps

All test cases pass on the target platform (creator self-reports; spot-checked)

Creator bio is complete (name, photo, previous work)

Description is unique — no copy-pasted AI tool documentation

Versioning

When updating a portfolio to a new version:

1 Increment manifest.json version (semver)

2 Add entries to changelog.md

3 Update test-cases/cases.json if test logic changed

4 Publish as a new release on the marketplace

5 Notify existing buyers via email with upgrade link