The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes JSON data from external HTTP 402 payment challenges, which represents a potential attack surface for indirect prompt injection.
Ingestion points: The agent parses the 402 challenge body received from merchant servers (e.g., the accepts array and extra fields containing token metadata).
Boundary markers: The skill relies on structured JSON parsing rather than natural language delimiters, though it includes technical instructions to treat certain fields as raw bytes to ensure signature validity.
Capability inventory: The skill utilizes shell-based tools (curl, cast, node, npx) and WebFetch to interact with external APIs and the blockchain.
Sanitization: The skill implements comprehensive 'Input Validation Rules', using regex for addresses and amounts, and explicitly rejecting any free-text values containing shell metacharacters.
[EXTERNAL_DOWNLOADS]: The skill handles the installation of the 'viem' Node.js library to facilitate off-chain signing.
Evidence: The skill documentation describes running npm install viem into a local cached directory at ~/.cache/uniswap-pay-with-app/signer/.
Mitigation: The installation is explicitly gated behind a mandatory AskUserQuestion call, ensuring the user confirms the dependency download before it occurs.
[COMMAND_EXECUTION]: The skill executes shell commands using cast for blockchain interaction and curl for interacting with the Uniswap Trading API.
Mitigation: Risks of command injection are mitigated by strict input validation rules that prohibit shell metacharacters in all fields interpolated into shell scripts and use regex to validate structured data like Ethereum addresses.

pay-with-app