Fetch Agents

Call Fetch.ai Agentverse agents by address. Search the Agentverse marketplace, browse a curated catalog of top agents (Tavily Search, ASI1-Mini, DALL-E 3, Te...

Security Scan

The skill's code, declared requirements, and runtime instructions are coherent with its stated purpose of calling Fetch.ai Agentverse agents; required env/binaries and installs are proportional, though it writes local state and spawns background processes which you should be aware of.

Purpose & Capability

Name/description match the actual behavior: the scripts call Agentverse endpoints, send ChatMessages to agents, and return responses. The declared requirements (python3, AGENTVERSE_API_KEY) and install of uagents/uagents-core are appropriate for this purpose.

Instruction Scope

SKILL.md instructs the agent to run bundled scripts (catalog.py, search.py, call.py, fire.sh, result.sh). Those scripts perform network requests to agentverse.ai, create a local mailbox, and read/write local files (.seed in the skill directory; /tmp/fetch-agents-<user>-response.txt). This is expected for the skill but does expand scope to local file I/O and background process management.

Install Mechanism

Install specifies uagents and uagents-core packages (uv/pip). No arbitrary URL downloads or extract-from-unknown-host steps are present. This is consistent with the Python-based implementation.

Credentials

Only AGENTVERSE_API_KEY is required and declared as the primary credential, which is reasonable. The code also optionally reads FETCH_AGENT_SEED (not declared) and persists/uses a local .seed file; this is functional for mailbox/identity persistence but worth noting because it creates a persistent local identity and can be overridden by an env var.

Persistence & Privilege

The skill does not request always:true and does not modify other skills. It does persist a seed file (./scripts/.seed path) and writes response files to /tmp, and fire.sh launches a background Python process (nohup + disown). These behaviors are normal for this kind of caller but mean the skill will leave local artifacts and background processes while waiting for agent responses.

Assessment

This skill appears to do exactly what it says: it needs AGENTVERSE_API_KEY, will install uagents and uagents-core, call agentverse.ai endpoints, and run bundled Python scripts. Before installing, consider: (1) Provide a scoped Agentverse API key you trust for mailbox registration; (2) the skill will create a .seed file in its directory (persistent agent identity) and a /tmp/fetch-agents-<youruser>-response.txt file for results; (3) when you call agents the skill spawns a background Python process (via nohup/disown) that will run for the call duration; (4) if you want to limit persistence or network exposure, inspect or run the scripts in an isolated environment or container, or supply a dedicated seed via the FETCH_AGENT_SEED env var. If any of these behaviors are unacceptable, do not install or run the skill until you review/modify the scripts.

Send queries to Fetch.ai Agentverse agents and return the agent's reply in natural language.

CRITICAL RULES — READ FIRST

1. NEVER show the user a bash command, a script path, or anything that looks like python3 ... or bash .... Those are for YOU to run internally. The user is chatting in Telegram/CLI and wants plain language.
2. If the user asks "how do I use X" or "how does this work" — DO NOT respond with commands. Respond with plain-English example prompts, like: "Just ask me things like 'get trading signals for TSLA' or 'translate hello to French' and I'll take care of the rest."
3. When the user makes an actual request (get signals, translate, etc.), run the script yourself and reply with the agent's answer. Format it nicely. Do not tell them to run anything.
4. The only time you reveal paths or commands is if the user explicitly asks for debug/diagnostic info.

How to answer common user questions

• "How do I use this?" / "What can this do?" → Explain in plain English: it calls Fetch.ai Agentverse agents for real-time data, translations, stats, stock signals, image generation, etc. Give 3-4 example prompts the user can try (in natural language, NOT bash).
• "What agents are available?" / "Show me the catalog" → Run catalog.py and format the result as a friendly list.
• "Find a [topic] agent" → Run search.py and return the top matches in plain text.
• Any actual task (signals, translation, stats, etc.) → Run fire.sh, tell the user to hold on ~40 seconds, then run result.sh and reply with the agent's answer in your own words.

Natural-language prompts the user might send

• "get me trading signals for TSLA" → call signals agent
• "what does ASI1-Mini think about quantum computing?" → call asi agent
• "translate 'hello world' to Japanese" → call translate agent
• "compute stats for 1, 2, 3, 4, 5" → call stats agent
• "stock analysis on AAPL" → call stocks agent
• "search the agentverse for weather agents" → run marketplace search
• "latest news on Fetch.ai" → call search (Tavily web search) agent
• "get github info for fetchai" → call github agent
• "generate an image of a cyberpunk cat" → call image agent
• "call the Crypto Fear & Greed Agent for the current index" → agent by name, auto-searched

Agent Shortcuts

Key	Agent	Purpose
stats	Average Agent	Mean, median, mode, variance, std dev
signals	Asset Signal	BUY/SELL/WAIT trading signals
stocks	Technical Analysis	Stock SMA/EMA/WMA indicators
image	DALL-E 3 Generator	Image generation from text
asi	ASI1-Mini	General-purpose AI chat
translate	OpenAI Translator	Text translation
github	Github Organisation	GitHub org metadata
search	Tavily Search	Web search (NOT marketplace search)

---

Internal Execution (for you, the agent, not the user)

The following are the commands YOU run with the exec tool. Never show these to the user.

Show the curated catalog

Runs in <5 seconds. Use for "what agents are available?" type questions.

python3 {baseDir}/scripts/catalog.py

Search the Agentverse marketplace

Runs in <5 seconds. Use when the user wants to find agents by topic.

python3 {baseDir}/scripts/search.py "query" -n 10

Call an agent (two-step, ~40 seconds total)

Calling an Agentverse agent takes 30-60 seconds. Use the two-step pattern:

Step 1 — Fire (returns instantly):

bash {baseDir}/scripts/fire.sh <shortcut-or-address-or-name> "user's query"

Accepts: a shortcut key from the table above, a full agent1q... address, or an agent name (auto-searched).

Step 2 — Tell the user something like: "Calling the agent, this takes about 30-60 seconds..."

Step 3 — After 30-60 seconds, read the result:

bash {baseDir}/scripts/result.sh

Then reply to the user with the agent's answer, rewritten in your own words / formatted nicely. Never show the raw command.

Rules

1. Resolve {baseDir} to the absolute path of this SKILL.md's parent directory.
2. Single absolute-path command per exec call. No cd, no &&.
3. The user never sees the commands or paths. Just give them the agent's answer.