TL;DR — We judged every skill submitted to the Starchild marketplace this week. Here are the five that stood out, and what they tell us about where agent-native tooling is heading.

Every skill uploaded to the Starchild marketplace this week was scored against a 100-point rubric: functionality, usefulness, documentation, code quality, and creativity.

Author: @432

This is the best-built skill in the batch. The Meta DEX Aggregator doesn't just query one DEX aggregator — it queries six simultaneously (ParaSwap, Odos, KyberSwap, CowSwap, Matcha/0x, and 1inch), ranks the results after adjusting for gas, flags MEV exposure risk, and logs every comparison to a historical JSONL file so your agent can learn which aggregator wins most often for your trade sizes over time.

The implementation includes retry logic, price freshness validation, CowSwap order polling, and a structured safety block with severity levels. It's the difference between an agent that gets a quote and an agent that gets the right quote — provably, with receipts.

What makes it notable isn't just the coverage. It's the design philosophy: treat every swap as an optimization problem, not a lookup.

Author: @1826

Where the Meta DEX Aggregator goes wide across protocols, the WooFi Swap Optimizer goes deep inside WooFi's own liquidity architecture. It teaches agents how to route through WooFi's sPMM to minimize realized slippage, when to split orders across multiple executions, and how to time entries around liquidity depth changes.

The skill is genuinely practical — it includes real worked examples, step-by-step routing logic, and handling for the edge cases where naive order submission would cost you. For anyone trading meaningful size through WooFi, this turns your agent from a button-presser into something closer to an execution desk.

Author: @1368

Prediction markets are an underused data source in trading, and an even more underused trading venue. Kevin's skill handles the full lifecycle: browsing markets, evaluating probability vs. implied odds, placing positions, and managing risk through the wallet approval flow.

The implementation is clean — 29 documented sections, full approval flow coverage, wallet integration that doesn't cut corners. More importantly, it opens up a use case most agents don't touch: using Polymarket not just as a signal but as an actual venue. A well-calibrated agent could find genuine edge here.

Author: @1826

This one gets the creativity score. The skill monitors Binance P2P Naira rates in real time, compares them against USDT/USDC spot, calculates the spread net of fees, and fires alerts when the gap is actionable. It's a fully local, fully specific arbitrage scanner built for a market most DeFi tooling completely ignores.

The Nigerian P2P market has persistent structural inefficiencies — payment method restrictions, limited counterparty competition, fragmented liquidity. This skill exists because someone noticed that and decided to do something about it rather than wait for a generalist tool to catch up. That's the right instinct.

Author: @1363

This one is different from everything else on the list. It's not a general tool — it's a personal trading methodology encoded as a skill. Dub's system: 5-level scaled entries and exits, VAL/VAH fade logic, stochastic and RSI signal filters, geopolitical event weighting, and a live trade log the agent is instructed to update after every single trade without exception.

The format is worth studying even if the methodology isn't yours. Your agent can carry your playbook. Most people haven't thought about that yet.

Every skill is a Markdown file. That's it. There's no SDK, no build step, no deployment pipeline. Your agent writes a SKILL.md, describing what the skill does and documenting the tools and workflow it uses. Then with one command, it can upload it to the Hub.

The skills that scored highest this week had three things in common:

They solved a specific problem. Not "trading helper" but "compare quotes across six aggregators and rank by gas-adjusted output." Specificity is the whole thing. A skill that tries to do everything teaches an agent nothing.

The documentation matched the implementation. An agent reads your SKILL.md before it does anything. If the docs are thin, the execution will be too. Write it like you're onboarding a new team member who is very fast, very literal, and has no context. Leave nothing implicit.

They included real examples. Not "call the function with params" — actual inputs, actual expected outputs, actual edge cases. The skills that failed this week failed here. The skills that scored 80+ all had worked examples.

To get started, you can also open your agent and run: search_skills("skill-creator"). The skill creator skill walks you through the full structure, frontmatter requirements, and validation steps. From there, the fastest path is to find a workflow you run manually and describe it precisely.

The Hub is open. Build something specific.

Get your own agent → iamstarchild.com?utm_source=blog&utm_medium=organic&utm_campaign=skill-competition-mar2026