The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data such as token names, symbols, and metadata retrieved from blockchain APIs.
Ingestion points: Data enters the agent's context through gmgn-cli token security responses and order status queries.
Boundary markers: The skill requires a structured trade summary and explicit user confirmation ("Reply 'confirm' to proceed") which serves as a human-review checkpoint before execution.
Capability inventory: The agent can submit on-chain transactions via swap, multi-swap, and order strategy create commands.
Sanitization: The instructions state that token metadata is sanitized for hidden control characters and prompt-injection framing, and the agent is explicitly told to ignore any instruction-like text found within these fields.
[COMMAND_EXECUTION]: Core functionality is performed through the gmgn-cli tool. The skill also suggests standard system commands for diagnosing network issues.
Evidence: Usage of gmgn-cli subcommands for financial operations and standard commands like ifconfig or ip addr show to check for IPv6 network configurations.
Scope: Execution is restricted to the specific CLI tool and local diagnostics, with no evidence of arbitrary shell execution or unsafe argument interpolation.
[EXTERNAL_DOWNLOADS]: Instructs the user to install and maintain the gmgn-cli utility via the npm registry.
Evidence: Recommends running npm install -g gmgn-cli for setup and updates.
Context: The package is the legitimate vendor-owned utility required for the skill's operation.
[DATA_EXFILTRATION]: Performs an external network request to a diagnostic service to identify the user's public IP version.
Evidence: The skill suggests contacting https://ipv6.icanhazip.com to check for IPv6 connectivity when troubleshooting authentication errors.
Context: This is a well-known service used for network diagnostics and does not involve the transfer of sensitive user data.

gmgn-swap