fetch-agents

# Fetch Agents

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.
```bash
python3 {baseDir}/scripts/catalog.py
```

### Search the Agentverse marketplace
Runs in <5 seconds. Use when the user wants to find agents by topic.
```bash
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
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
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.