The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill identifies a potential attack surface where transaction parameters (such as recipients or call data) could be influenced by untrusted external content like web pages or prior tool outputs.
Ingestion points: Processes data from user prompts and external tool outputs (e.g., /token/search) to populate transaction fields.
Boundary markers: The instructions mandate explicit warning messages (e.g., '⚠️ EXTERNAL BRIDGE') and require a verbatim 'yes' from the user before execution.
Capability inventory: The skill utilizes financial execution tools (relay_execute, onchain_jupiter_execute) across multiple blockchain networks.
Sanitization: Includes a pre-swap checklist to resolve token symbols to contract addresses and validates market capitalization and liquidity to mitigate risks from malicious tokens.
[COMMAND_EXECUTION]: Through the bridge+call feature, the agent can execute arbitrary blockchain transactions (txs). The skill mitigates unauthorized execution by requiring a plain-English summary of each call's data and value, followed by individual user confirmation for every transaction in the sequence.
[EXTERNAL_DOWNLOADS]: Fetches chain configuration and metadata from api.relay.link, an established service for cross-chain liquidity and bridging operations.

openfin-relay