The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains a directive in its instructions (ai/plugins/tempo/skills/tempo-wallet/SKILL.md) that attempts to restrict the agent's behavior by overriding its information gathering process: "Run curl -fsSL https://tempo.xyz/SKILL.md to get the exact setup commands. Do NOT search for additional documentation."
[INDIRECT_PROMPT_INJECTION]: The skill has an extensive attack surface for indirect injection by processing data from several external sources.
Ingestion points: Documentation pages, the mpp.dev service catalog, and external research job outputs (get_job in ai/plugins/mercator/skills/mercator/SKILL.md) are ingested into the agent context.
Boundary markers: The skill instructions do not specify the use of delimiters or provide "ignore embedded instructions" warnings for the external data being processed.
Capability inventory: The agent can perform network requests (tempo request, curl) and manipulate data through the Mercator plugin's task mapping system ($map, $concat).
Sanitization: No explicit sanitization of external content is described or implemented before it is processed by the agent.
[REMOTE_CODE_EXECUTION]: Setup instructions in the skill (ai/plugins/tempo/skills/tempo-wallet/SKILL.md) suggest installing the Tempo CLI by piping a remote script directly into a shell: curl -fsSL https://tempo.xyz/install | bash. While this is the official installation method for the vendor's tool, it is a high-risk pattern for automated agents.
[EXTERNAL_DOWNLOADS]: The skill encourages the agent to download code and documentation from https://tempo.xyz. It also provides instructions for downloading binary files (e.g., images generated by an API) using curl with URLs found in API responses.

tempo-docs