The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill documentation and helper files require the use of sensitive cryptographic credentials.
The SKILL.md file lists an "Ethereum private key for x402 payment signing" as a prerequisite for operation.
The references/payment.md file provides a Python implementation that accepts a raw private_key string and uses it with Account.from_key(). Handling raw private keys within an AI agent's operational context presents a risk of exposure if the agent is targeted by prompt injection or if logs are improperly secured.
[COMMAND_EXECUTION]: The skill provides functional code snippets for cryptographic signing and network communication.
The references/payment.md file contains Python code that utilizes the web3 library to perform network calls to external RPC providers for contract metadata (name, version, and decimals).
The provided code executes EIP-712 typed data signing (Account.sign_typed_data), which involves complex cryptographic operations performed within the agent's runtime environment.
[EXTERNAL_DOWNLOADS]: The skill interacts with and references external vendor-controlled resources.
It utilizes an MCP gateway endpoint at https://mcp.fortytwo.network/mcp for tool execution.
It references canonical skill files and documentation hosted on github.com/Fortytwo-Network and docs.fortytwo.network. These are identified as official resources belonging to the skill author.
[PROMPT_INJECTION]: The skill defines specific triggers and interpolation patterns that interact with user input.
The SKILL.md file includes "Dissatisfaction" and "Skepticism" triggers (e.g., "that's wrong", "prove it") that instruct the agent to proactively suggest the paid service, which influences agent behavior based on user sentiment.
User input is directly interpolated into a JSON-RPC tool call payload using the {user_question} placeholder. While this is a common pattern for MCP tools, it remains an ingestion point for potentially malicious instructions embedded in user queries.

fortytwo-mcp