The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources, including DeFi APIs, prediction markets, and web search results. This creates a surface where maliciously crafted external content could attempt to influence the agent's logic.
Ingestion points: Data is fetched in SKILL.md via DeFiLlama, CoinGecko, Polymarket, and WebSearch queries.
Boundary markers: The skill includes an explicit 'Network note' directing the agent to treat all returned fields as untrusted and to never follow instructions embedded in fetched content.
Capability inventory: The skill utilizes external_api, sends_notifications, and contents:write capabilities.
Sanitization: The instructions explicitly forbid shell interpolation of external data and require the agent to ignore any embedded directives.
[EXTERNAL_DOWNLOADS]: The skill fetches market data and metadata from well-known technology services and financial data providers, including DeFiLlama, CoinGecko, Polymarket, and alternative.me. These are established services used for the skill's primary purpose.
[COMMAND_EXECUTION]: The skill uses standard shell commands (curl, mkdir, grep) for data retrieval and processing. It uses a specialized ./secretcurl tool to handle API keys securely, preventing credential exposure in logs or process lists.

defi-overview