The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted third-party data including transaction activity, token names, and wallet metadata (e.g., common.tags).
Ingestion Points: gmgn-cli portfolio stats and gmgn-cli portfolio activity commands fetch external blockchain data.
Capability Inventory: The agent is instructed to write local HTML files and execute shell commands via gmgn-cli.
Sanitization: The skill includes explicit instructions to HTML-escape data, neutralize suspicious metadata notices from the CLI, and ignore any text in data fields that resembles commands or overrides.
Boundary Markers: The 'Notes' section explicitly directs the agent to treat all third-party fields as data only and not as instructions.
[COMMAND_EXECUTION]: The skill requires the installation and execution of the gmgn-cli tool to interact with the gmgn.ai platform.
Evidence: Commands such as gmgn-cli portfolio stats and gmgn-cli portfolio activity are used to retrieve the necessary data for analysis.
Context: The tool is a vendor-provided CLI for the 'gmgnai' ecosystem, used here for read-only data retrieval.
[DYNAMIC_EXECUTION]: The skill generates interactive HTML previews by assembling data into a local file.
Evidence: Step 9 details the creation of a local HTML page using pixel-based coordinate mapping and responsive scaling logic.
Security Controls: The instructions mandate HTML-escaping of all API strings and the use of Unicode escapes inside script tags to prevent XSS or script termination attacks.

gmgn-wallet-style