Skill-Plus

The open standard for Agent skill dispatch metadata.

Skill-Plus defines how AI Agents declare, discover, and schedule capabilities — with resource awareness, cost profiles, and safety declarations that go beyond simple function-calling.

Created by Point Eight AI — the team behind V-Box, the Agent-native social network.

What is Skill-Plus?

MCP and OpenAI function-calling solve "what can an Agent call". Skill-Plus solves the harder problem: "when should an Agent call it, at what cost, and when should it not?"

Key Innovations

1. Resource-Aware Declarations — Skills declare required context window, model capabilities, latency budget, GPU/network needs.
2. Dispatch Hints — invoke_when / do_not_invoke_when conditions for intelligent scheduling.
3. Cost Profile — Token consumption, P50/P99 latency estimates.
4. Safety Declarations — Data retention policy, user consent requirements, content categories.

Repository Structure

├── spec/                    # Skill-Plus specification + JSON Schema
├── skills/                  # Official & community-contributed Skills
│   └── {slug}/
│       ├── skill.yaml       # Skill manifest
│       ├── main.py          # Implementation (or main.go / main.ts)
│       ├── README.md
│       └── tests/
├── sdk-python/              # pip install skillplus
├── sdk-typescript/          # npm install @skillplus/sdk
├── sdk-go/                  # Go SDK
├── cli/                     # skillplus validate / publish
└── examples/                # Integration examples

Quick Start

# Install the CLI
pip install skillplus

# Validate a skill
skillplus validate ./my-skill/

# Run tests
skillplus test ./my-skill/

Writing a Skill

A Skill-Plus skill is defined by a skill.yaml manifest:

slug: my-skill
version: 1.0.0
name: My Custom Skill
description: A skill that does something useful

dispatch_hints:
  invoke_when:
    - content_type: image
    - content_type: text
      min_length: 100
  do_not_invoke_when:
    - content_type: video

resources:
  context_window: 4096
  model_capability: vision
  latency_budget_ms: 5000
  memory_mb: 256

cost_profile:
  avg_tokens: 500
  p50_latency_ms: 1200
  p99_latency_ms: 4500

safety:
  data_retention: none
  user_consent_required: false
  content_categories: [general]

Contributing

We welcome contributions! All contributors must sign our CLA before their PR can be merged.

1. Fork this repository
2. Create your skill in skills/{your-slug}/
3. Include skill.yaml, implementation, README, and tests
4. Submit a Pull Request

See CONTRIBUTING.md for detailed guidelines.

License

MIT — see LICENSE.

Copyright 2026 Point Eight AI Pte. Ltd.