The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from various external bridge APIs and blockchain RPC providers, creating a surface for indirect prompt injection if those sources were compromised.
Ingestion points: External API responses from DefiLlama, Wormhole, LayerZero, Across, and public blockchain RPC endpoints are fetched in scripts/bridge_fetcher.py, scripts/protocol_adapters.py, and scripts/tx_tracker.py.
Boundary markers: The retrieved data is formatted into plain-text tables or JSON for the agent without specific delimiters or instructions to ignore potential commands embedded in the data.
Capability inventory: The agent is granted Read, Write, and Edit permissions, along with scoped Bash access to the bridge_monitor script.
Sanitization: The skill does not perform sanitization on string values returned from external APIs before displaying them to the agent.
[EXTERNAL_DOWNLOADS]: The skill connects to several well-known cryptocurrency monitoring services and blockchain infrastructure providers to fetch real-time data.
Targets include the DefiLlama Bridges API (bridges.llama.fi), Wormhole Scan (api.wormholescan.io), LayerZero Scan (scan.layerzero.network), and Across Protocol (across.to).
It also utilizes public RPC endpoints for multiple blockchain networks including Ethereum, Binance Smart Chain, Polygon, Arbitrum, Optimism, Base, and Avalanche.
[CREDENTIALS_UNSAFE]: The skill provides a secure framework for managing sensitive information.
The documentation in references/implementation.md correctly instructs users to store API credentials in a separate .env file (config/crypto-apis.env) rather than hardcoding them.
The configuration loader in scripts/config_loader.py prioritizes environment variables for RPC URLs, following security best practices.

monitoring-cross-chain-bridges