The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains directive markers designed to override default agent behavior. Specifically, it instructs the agent to ignore summaries from web fetch tools and prioritize setup commands from a specific remote file at https://tempo.xyz/SKILL.md, while also directing the agent not to search for additional documentation during the setup process.
[REMOTE_CODE_EXECUTION]: The skill provides instructions to download and execute a setup script directly from the vendor's domain (https://tempo.xyz/install) using a piped shell command (| bash). As this resource belongs to the official vendor infrastructure, it is classified as a safe administrative task.
[EXTERNAL_DOWNLOADS]: The skill includes capabilities to download files from remote URLs returned by service responses (e.g., for retrieving generated images) using curl. This behavior is limited to specific utility functions within the wallet skill.
[COMMAND_EXECUTION]: The skill relies extensively on shell command execution via the tempo CLI and curl to manage wallets, discover blockchain services, and execute paid HTTP requests.
[PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection as it processes data from external documentation and service APIs to construct shell commands.
Ingestion points: read_web_page on https://tempo.xyz/developers/llms.txt, mcp__tempo_mcp__read_page (SKILL.md).
Boundary markers: None identified in the instructional text for separating untrusted content.
Capability inventory: Includes tempo request, curl, and bash execution (ai/plugins/tempo/skills/tempo-wallet/SKILL.md).
Sanitization: Not explicitly defined within the prompt construction logic.

tempo-docs