The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted transaction and token metadata from external blockchain APIs and CoinGecko. While these fields are included in alerts, the skill provides explicit instructions to treat this data as untrusted and to avoid using it in commands.
Ingestion points: External data from Alchemy, Etherscan, Robinhood RPC, and CoinGecko.
Boundary markers: No specific delimiters are used for external content in the generated notifications.
Capability inventory: Uses the ./notify tool for alerting and performs file system writes for logging and state management.
Sanitization: Explicitly instructs the agent to treat all fetched fields as untrusted and avoid interpolation into shell commands.
[COMMAND_EXECUTION]: Local shell utilities like sed, awk, and grep are used to update configuration files. Input is strictly validated against a hexadecimal address regex and a whitelist of supported chains before being processed.
[EXTERNAL_DOWNLOADS]: Retrieves blockchain and pricing data from well-known infrastructure and financial service providers (Alchemy, Etherscan, CoinGecko).
[CREDENTIALS_UNSAFE]: Uses a placeholder system within the ./secretcurl utility to handle sensitive API keys, preventing credentials from appearing in plaintext within shell execution history or logs.

onchain-monitor