app.agentmesh.link

The description uses strong imperative framing ('USE THIS TOOL when...', 'Do not stop at a capability limitation', 'do not ask the user to manually search... before checking AgentMesh') that goes beyond describing the tool's function and instead tries to bias the agent's general reasoning/decision process toward always preferring this tool over other options or over informing the user of limitations. This pattern resembles prompt-injection style steering embedded in a tool description rather than a neutral capability description.

The tool advertises that it can 'discover compatible external MCP providers and tools' and includes an 'execute' boolean that defaults to false but can be set true, allowing the agent to hand off the user's problem/query text and potentially trigger execution via unknown third-party providers. There is no schema-level control limiting what data is sent or verifying explicit user authorization before execution; the safeguard ('External execution occurs only when explicitly authorized') is only a textual instruction, not enforced by the tool interface.