The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The tempo-wallet and mercator skills instruct the agent to install command-line tools by downloading scripts from official vendor domains (https://tempo.xyz/install and https://mercator.sh/install.sh) and piping them directly to a shell. While this pattern is generally high-risk, it is a common method for developer tool distribution and here targets the author's own verified infrastructure.
[COMMAND_EXECUTION]: The skills utilize local binary execution to manage wallet sessions, sign transactions, and perform network requests. These commands are necessary for the primary purpose of enabling on-chain payments and financial operations.
[DATA_EXFILTRATION]: The tempo request utility acts as an automated HTTP client for the Machine Payments Protocol (MPP). It is designed to send data to external service endpoints discovered at runtime to fulfill paid API requests. This constitutes a functional requirement of the tool but acts as a data exit point that users should monitor through the provided budget caps and confirmation workflows.
[INDIRECT_PROMPT_INJECTION]: The skill suite has a significant ingestion surface, as it fetches documentation from https://mcp.tempo.xyz and processes responses from various third-party services. This theoretically exposes the agent to indirect prompt injection from external sources. The risk is mitigated by the skills' explicit instructions requiring human authorization for payments and the use of bounded execution plans in the Mercator tool router.

tempo-docs