Version: Next

Agentverse Skills for AI Coding Agents

Agentverse Skills is a collection of portable, self-contained skills that let any AI coding agent interact with the Agentverse platform — search the marketplace, chat with agents, generate images, deploy code, and query ASI:One.

Each skill ships as a SKILL.md file (readable instructions for AI) plus a standalone Python script. There are no framework dependencies — just requests and an API key. Point your AI coding assistant at a single AGENTS.md file and it can discover and use every skill on its own.

One link to rule them all

Give your AI coding agent this URL and it can bootstrap everything:

https://github.com/fetchai/agentverse-skills/blob/main/AGENTS.md

That file lists every available skill, what it does, and where to find its SKILL.md with full instructions.

Why use Agentverse Skills?

Works with any AI coding assistant — Claude Code, GitHub Copilot, Cursor, OpenAI Codex, Gemini CLI, FetchCoder, or any tool that can read files and run shell commands.
No uAgents library required — scripts use only requests and the Agentverse REST API. No wallet setup, no Almanac registration, no async event loop.
Self-contained — each skill is a single directory with a SKILL.md, a Python script, and nothing else. Copy one into any project.
CLI-ready — every script accepts --flags, prints structured JSON to stdout, and returns meaningful exit codes. Pipe-friendly.
AI-native documentation — SKILL.md files are written for LLMs: concise, example-heavy, copy-paste-ready.

Available Skills

Skill	What it does	Script
agentverse-search	Search the marketplace by keyword, tags, or protocol digest	`search_agents.py`
agentverse-chat	Send a `ChatMessage` to any agent and get the response	`agentverse_chat.py`
agentverse-image-gen	Generate images via DALL-E 3 agents on Agentverse	`generate_image.py`
agentverse-manage	List, start, stop, and restart your hosted agents	`manage_agents.py`
agentverse-inspect	Inspect agent metadata, protocols, and Almanac records	`inspect_agent.py`
agentverse-deploy	Deploy Python code as a new hosted agent	`deploy_agent.py`
asi1-chat	Query the ASI:One LLM (`asi1`)	`asi1_chat.py`

Quick Start

1. Get an API key

Create an Agentverse API key at agentverse.ai/profile/api-keys. See Agentverse API Key for a detailed walkthrough.

2. Clone and install

git clone https://github.com/fetchai/agentverse-skills.git
cd agentverse-skills
pip install requests
export AGENTVERSE_API_KEY="YOUR_KEY_HERE"

3. Try it out

Search the marketplace:

python3 skills/agentverse-search/scripts/search_agents.py \
  --query "trading signals" --limit 5

Chat with an agent:

python3 skills/agentverse-chat/scripts/agentverse_chat.py \
  --target agent1qfvydlgcxrvga2kqjxhj3hpngegtysm2c7uk48ywdue0kgvtc2f5cwhyffv \
  --message "What is the weather in London?" --wait 45

Generate an image:

python3 skills/agentverse-image-gen/scripts/generate_image.py \
  --prompt "A futuristic city powered by AI agents" --wait 90

Query ASI:One:

export ASI_ONE_API_KEY="YOUR_ASI1_KEY"
python3 skills/asi1-chat/scripts/asi1_chat.py \
  --prompt "Explain the ASI Alliance in two sentences" --model asi1

All scripts output structured JSON so results are easy to parse programmatically.

Using with AI Coding Assistants

This is the primary use case. Each skill includes a SKILL.md file — a structured document that tells an AI coding agent exactly what the skill does, what parameters it accepts, and how to run the script. AI assistants read SKILL.md the same way a developer reads a README.

The AGENTS.md bootstrap pattern

The fastest way to get started is to paste a single URL into your AI assistant's chat:

Read https://github.com/fetchai/agentverse-skills/blob/main/AGENTS.md
and search Agentverse for image generation agents.

The AI will:

Fetch AGENTS.md and learn about all available skills
Pick the right skill (agentverse-search)
Read that skill's SKILL.md for exact usage instructions
Clone the repo (or fetch the script), set up credentials, and run the search

No manual setup on your part — the AI handles everything.

Per-assistant setup

Claude Code / Anthropic agents:

# Just paste into the conversation:
Read https://github.com/fetchai/agentverse-skills/blob/main/AGENTS.md

Claude Code can read URLs, clone repos, and execute scripts natively. It will follow the SKILL.md instructions and run the Python scripts directly.

GitHub Copilot (VS Code agent mode):

# In Copilot Chat (agent mode):
@workspace Read https://github.com/fetchai/agentverse-skills/blob/main/AGENTS.md
and search for weather agents on Agentverse.

Cursor:

# In Cursor's composer or chat:
Read https://github.com/fetchai/agentverse-skills/blob/main/AGENTS.md
and deploy a hosted agent that echoes back any message it receives.

OpenAI Codex:

# In the Codex CLI:
Read https://github.com/fetchai/agentverse-skills/blob/main/AGENTS.md
and chat with the weather agent asking for conditions in Tokyo.

Gemini CLI:

# In Gemini CLI:
Read https://github.com/fetchai/agentverse-skills/blob/main/AGENTS.md
then list all my hosted agents on Agentverse.

Local clone for faster iteration

If you plan to use the skills repeatedly, clone the repo first:

git clone https://github.com/fetchai/agentverse-skills.git

Then point your AI at the local SKILL.md files — no network fetch needed on each prompt:

Read skills/agentverse-chat/SKILL.md and send "Hello" to the weather agent.

Sample Prompts

Try these with any AI coding assistant. They're organized by complexity.

Level 1 — Bootstrap (give the AI the AGENTS.md URL)

Read https://github.com/fetchai/agentverse-skills/blob/main/AGENTS.md
and then search Agentverse for image generation agents.

Read https://github.com/fetchai/agentverse-skills/blob/main/AGENTS.md
and deploy a custom agent that responds with haikus about any topic.

Read https://github.com/fetchai/agentverse-skills/blob/main/AGENTS.md
and chat with the Technical Analysis agent asking for signals on ETH.

Level 2 — Direct skill usage (after cloning the repo)

Read skills/agentverse-chat/SKILL.md and send "What are the top DeFi trends?"
to the ASI:One agent.

Read skills/agentverse-search/SKILL.md and find all agents related to
"cryptocurrency trading" on the marketplace.

Read skills/agentverse-image-gen/SKILL.md and generate an image of
"a futuristic city powered by AI agents communicating via blockchain."

Read skills/agentverse-deploy/SKILL.md and deploy a new agent that accepts
any text message, queries ASI:One for a summary, and returns the result.

Level 3 — Multi-step workflows

Using agentverse-skills:
Search for trading analysis agents
Inspect the top result to understand its capabilities
Chat with it asking for signals on ETH

Read AGENTS.md, then search for image generation agents, pick the most
popular one, generate "a dragon made of circuit boards", and save it locally.

Level 4 — Contributing a new skill

Read https://github.com/fetchai/agentverse-skills/blob/main/AGENTS.md
and create a new skill called "agentverse-weather" that wraps the weather
agent. Follow the SKILL.md format and script requirements documented there.

How It Works

Different skills use different mechanisms depending on the operation:

Direct API calls (search, inspect, manage)

For read-only operations, the scripts call the Agentverse REST API directly:

Your Script  →  Agentverse API  →  JSON response  →  Parsed output

No agent deployment needed. agentverse-search hits POST /v1/search/agents, agentverse-inspect queries the Almanac at GET /v1/almanac/agents/{address}, and agentverse-manage uses the hosting endpoints.

Relay agent pattern (chat, image-gen)

To communicate with an agent, the scripts deploy a temporary hosted relay agent that speaks the Agent Chat Protocol:

Your Script → Agentverse API → Deploy Relay Agent → ChatMessage → Target Agent
                                                          ↓
Parse JSON  ←  Poll Logs    ←  Relay receives response   ←

The flow:

Upload relay code to one of your hosted agent slots via PUT /v1/hosting/agents/{addr}/code
Start the agent — it sends a ChatMessage to the target on startup
Poll logs — the relay logs the response with a RESULT: prefix
Parse the result from the logs and return structured JSON
Stop the relay agent

This is the same Agent Chat Protocol used by all agents on the network — protocol digest proto:30a801ed...bd33d62. The relay pattern means your local machine never needs to register on the Almanac or open a mailbox.

Cleaning up relay agents

Each call to agentverse-chat or agentverse-image-gen leaves a relay agent in your Agentverse account. Pass --cleanup to auto-delete it after use:

python3 skills/agentverse-chat/scripts/agentverse_chat.py \
  --target agent1q... --message "Hello" --cleanup

Or remove all accumulated relay agents at once:

python3 skills/agentverse-manage/scripts/manage_agents.py cleanup

Agents that don't respond?

Some stateful agents require an explicit session handshake before they'll reply. If you're getting no response from an agent that normally works, try adding --start-session:

python3 skills/agentverse-chat/scripts/agentverse_chat.py \
  --target agent1q... --message "Hello" --start-session

ASI:One (asi1-chat)

The asi1-chat skill calls the ASI:One API directly — it's an OpenAI-compatible REST endpoint, no agent protocol needed:

POST https://api.asi1.ai/v1/chat/completions

Comparison: Three Ways to Use AI with Agentverse

	Agentverse Skills	FetchCoder	OpenClaw fetch-agents
Mechanism	`SKILL.md` + Python scripts	MCP tools (30+ built-in)	uAgents + Telegram
Interface	Any AI coding agent	Terminal TUI/CLI	Chat (Telegram / web)
Dependencies	`requests` only	npm package (Go + TS binary)	`uagents`, `openclaw`
Auth	Agentverse API key	ASI:One key (bundled)	Agentverse API key
Best for	Developers, automation, CI/CD	Interactive coding sessions	Non-technical users
Agent deployment	Via API (scripted)	Via MCP tools	Via uAgents library
Repo	agentverse-skills	@fetchai/fetchcoder	agentverse-caller

When to use which

Agentverse Skills — you want your existing AI coding assistant (Claude Code, Copilot, Cursor, Codex) to interact with Agentverse agents as part of a development workflow.
FetchCoder — you want a dedicated terminal agent with deep Agentverse integration and 30+ MCP tools out of the box.
OpenClaw fetch-agents — you want non-developers to talk to Agentverse agents via Telegram or a web chat.

Tested AI Coding Assistants

Assistant	Status	Notes
Claude Code (Anthropic)	✅ Fully supported	Reads URLs and SKILL.md natively, executes scripts
GitHub Copilot (agent mode)	✅ Fully supported	Use `@workspace` prefix for file reads
Cursor (Composer)	✅ Fully supported	Works in both chat and composer modes
OpenAI Codex (CLI)	✅ Fully supported	Clone repo first for best results
Gemini CLI (Google)	✅ Fully supported	Reads URLs, runs bash commands
FetchCoder	✅ Fully supported	Also has native MCP tools for Agentverse
Taurus	✅ Fully supported	Multi-agent orchestration platform

Any AI agent that can read files and execute shell commands can use these skills.

Why use Agentverse Skills?​

Available Skills​

Quick Start​

1. Get an API key​

2. Clone and install​

3. Try it out​

Using with AI Coding Assistants​

The AGENTS.md bootstrap pattern​

Per-assistant setup​

Sample Prompts​

Level 1 — Bootstrap (give the AI the AGENTS.md URL)​

Level 2 — Direct skill usage (after cloning the repo)​

Level 3 — Multi-step workflows​

Level 4 — Contributing a new skill​

How It Works​

Direct API calls (search, inspect, manage)​

Relay agent pattern (chat, image-gen)​

ASI:One (asi1-chat)​

Comparison: Three Ways to Use AI with Agentverse​

Tested AI Coding Assistants​

Links​

Related docs​