The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches dynamic 'code recipes' and contract ABIs from llms.megapot.io at runtime. This design ensures the agent uses the most up-to-date contract addresses and transaction logic as the protocol evolves.
[REMOTE_CODE_EXECUTION]: By design, the skill instructs the agent to fetch and follow task-specific instructions from remote task pages (e.g., https://llms.megapot.io/tasks/buy-bulk). This constitutes dynamic instruction loading where the agent's behavior for complex tasks is determined by content hosted on the vendor's infrastructure.
[PROMPT_INJECTION]: The skill integrates with an external Data API (api.megapot.io) to discover a user's unclaimed winnings, which creates a surface for indirect prompt injection.
Ingestion points: Data is ingested from the /v1/wallets/{address}/wins endpoint as described in references/data-api.md and references/claim-winnings.md.
Boundary markers: There are no specific delimiters or 'ignore embedded instructions' warnings applied when the agent processes the API response content.
Capability inventory: The skill has the capability to initiate blockchain transactions, specifically claimWinnings and buyTickets, which involve asset transfers (USDC) and NFT minting.
Sanitization: The skill instructions specify lowercasing the user's wallet address before querying the API, but there is no explicit validation or sanitization of the values returned by the API (such as ticket IDs or prize amounts) before they are used in transaction payloads or presented to the user.

megapot