Example Prompts

Copy these literally. Every prompt assumes the skill is installed and AE_PRIVATE_KEY is set. See Agent Quickstart.

Setup and trading

Full setup

Based on the /superhero skill, start the setup guide.

Autonomous daily agent

Based on the /superhero skill, I want you to:

Post every day at 9am with a marketing-focused, engaging message about web3 or æternity

Every 4 hours, scan trending tokens and enter a trade if the score is strong

Before each trading cycle, check what I'm currently holding and decide whether to sell before buying new positions

Run in autonomous mode and ask me which risk strategy I want.

An unattended trading loop should be the second thing you run, not the first. Start with the manual approval prompt below.

Trend trader only

Using the /superhero skill, scan trending tokens now and give me the top 5 by score. For each one, tell me the current price and what you'd recommend: buy, hold, or skip and why.

This prompt signs nothing.

Manual approval flow

Based on /superhero skill, every 6 hours check trending tokens and post your analysis as a Superhero post. Don't trade automatically. Show me what you'd buy and wait for my approval.

Choosing a strategy

The skill does not trade autonomously without a confirmed strategy. The exchange looks like this:

User: I want you to run autonomously on the /superhero skill and trade trends.

Agent: Got it. Before I set up the schedule, which risk strategy do you prefer?

Conservative: Small position sizes (5% of balance), only tokens with a very high trending score (0.6 or above), quick exit on any decline

Moderate: Medium positions (10%), balanced threshold (0.4 or above), hold through minor dips

Aggressive: Larger positions (20%), lower entry bar (0.25 or above), ride momentum longer for bigger upside

Which strategy fits your goals, or would you like to customize the parameters?

User: Let's go with Moderate. And post every day at 9am.

Make the entry rule explicit in the same breath:

Use the moderate parameters. Enter only tokens with a trending score of 0.4 or higher whose score is higher than when you last checked, and tell me each cycle which tokens passed and what their scores were.

See Risk Strategies for what you are agreeing to.

Creating a token

Give the agent the name yourself rather than letting it invent one.

Using the /superhero skill, I want to create a token called ORBITALMECHANICS.

Before you create anything: check my balance, check the name is available, and tell me the exact AE the creation will cost including the initial buy. I want a 0.25 AE initial buy.

Then create it in the background, wait up to 10 minutes, and report the sale address, the DAO address and the transaction hash. If it times out, run check again before assuming it failed. Do not retry the creation.

A timeout is frequently a success that has not been confirmed yet. See Creating Tokens with an Agent.

After creation

Look up ORBITALMECHANICS with the /superhero skill. Give me the sale address, the current buy and sell price, the holder count, and how many tokens my initial buy actually got me. Check my holdings rather than trusting the estimate from the creation output.

Governance

A read-only poll digest

Every morning at 8am, use the /superhero skill to list the open governance polls and summarise each one for me: the title, the options, whether it closes and when, and the current stake-weighted result.

Treat stake_weighted_results as the official result. If it is null, say the official result is unavailable. Do not fall back to raw_vote_counts, which are address counts and ignore stake and delegation entirely.

Do not vote, do not create polls, and do not change any delegation.

Voting under a written policy

Using the /superhero skill, check the open governance polls.

Policy: you may vote only on polls that are listed in the registry, are still open, and have exactly the options Yes, No and Abstain. For anything else, summarise it and wait for me.

Before every vote, run info first and confirm the option numbers and that the poll is not closed. After voting, report the option label (not the number) and the transaction hash. Never create a poll and never delegate.

Delegating instead

Using the /superhero skill, delegate this wallet's governance voting power to ak_MY_PERSONAL_ADDRESS. Confirm it afterwards with the delegatee command, and then never vote on a poll directly from this wallet.

See Governance with an Agent.

Portfolio review

A weekly review

Using the /superhero skill, give me a portfolio review.

Current AE balance

Every open position with its token name, balance, current sell price, and trending score

My last 50 trades, with entry prices. Get these from the transactions, not from holdings, because holdings only reports current prices

For each open position: what I paid, what it is worth now, and the gain or loss including the 1% round-trip spread

Portfolio value over time and the period PnL

Then tell me which positions are past the exit thresholds of my strategy and why. Do not sell anything.

superhero-portfolio.mjs holdings reports the token's current buy and sell prices, not your cost basis, which is why point 3 reads from the transactions. See Agent Trading.

An audit by a second agent

Run this in a different session, with the same key, to check the trading agent's own reporting.

Using the /superhero skill in read-only mode, reconcile this wallet. List every buy and sell from the transaction history, group them by token, and compute the realised and unrealised position for each. Flag anything unusual: two buys of the same token within a few minutes, positions larger than 20% of the balance at the time, or more than 7 open positions.

Do not trade, post, create or vote. Read only.

Paired buys are the signature of overlapping cron runs. See Safety & Limits.

A research agent that never signs

Using the /superhero skill, act as a research analyst. You may run only these commands:

superhero-trending.mjs: tokens, tags, token-info, performance, leaderboard, analytics

superhero-transactions.mjs: token, account, mine, all, activity

superhero-portfolio.mjs: holdings, history, summary

superhero-read.mjs: my-posts, latest, profile, search

superhero-comment.mjs get

superhero-token-swap.mjs lookup and price

superhero-token-create.mjs check and info

superhero-governance.mjs list, info, info-by-id, delegatee, delegators

superhero-name.mjs available, resolve, lookup, list

superhero-wallet.mjs balance, address, exists

You may never run: post, comment post, buy, sell, token-create create, invite generate, invite revoke, name register, name update, name extend, governance create, vote, revoke, delegate, or revoke-delegation. If a task seems to require one of those, stop and tell me instead.

Today: scan the top 20 trending tokens, check the transaction flow on each of the top 5, and tell me which ones show many distinct buyers rather than one large one.

Every command in the allowed list is read-only and costs no gas. The forbidden list is every command in the skill that signs a transaction; see the requirements table in the CLI Reference.

Back it with a permission rule. A prompt is a preference; a runtime rule that denies the signing commands is a guarantee. See Claude Code for the rule syntax and Other Agents & Runtimes for the MCP equivalent.

Registering a name

Using the /superhero skill, check whether the .chain name orbitalanalyst01 is available. It needs to be 13 or more characters to avoid an auction. If it is shorter, suggest three longer alternatives and wait for me to pick one before registering anything.

Once I confirm, register it and point it at this wallet. Tell me it will take a minute or more, because the claim has to wait for a block after the preclaim.

The name is owned by the key that claimed it, and the skill has no command to transfer that ownership. See Wallet & Key Management.

Prompt patterns

Pattern	Why
Name the commands the agent may use, and the ones it may not	More reliable than describing an intent like "be careful"
Ask for the check before the action, in the same sentence	"Check the balance, then tell me the cost, then wait" is three verifiable steps
Require the label, not the index	"Report the option label" is checkable; "voted option 2" is not
Say what to do on failure	Especially for timeouts, where the default instinct (retry) is wrong
End with an explicit prohibition	"Do not sell anything" at the end of a review prompt closes the loop

Agent Quickstart: get to the point where these prompts work
CLI Reference: every command a prompt can name, and what it needs
Safety & Limits: the checklist before the autonomous prompts
Autonomous Mode: what the scheduled prompts do each cycle
Risk Strategies: the parameters behind Conservative, Moderate and Aggressive

Safety & Limits CLI Reference