Single tool 'stream_tool_loop' has a reasonable description (145 chars) but critical schema deficiencies. Input schema lists parameter names and types but lacks proper JSON Schema structure with 'type', 'description', and 'required' fields visible in the definition. Parameters like 'tier' reference a custom 'ModelTier' type without definition. No output schema documented. The tool description explains what it does but omits when to use it, prerequisites, or what structure the yielded events follow. Parameter descriptions are minimal (e.g., 'Model tier for LLM selection' is vague about valid values). No error handling guidance. Tool name 'stream_tool_loop' is generic and does not follow verb_noun convention clearly.

• Define input schema as proper JSON Schema with 'type', 'properties', 'required', and 'description' for each parameter. Define 'ModelTier' as an enum (e.g., ['claude-3-5-sonnet', 'claude-3-opus', 'claude-3-haiku']) or reference a schema definition.
• Document output schema: specify that the tool yields a stream of dicts, each with one of {text: string}, {tool: {name: string, input: object}}, {tool_result: {content: string}}, or {end: {stop_reason: string}}. Include field types and examples.
• Expand parameter descriptions with constraints and discovery hints. E.g., 'tier: Model tier for LLM selection (valid: claude-3-5-sonnet, claude-3-opus, claude-3-haiku)'. For tool_set, add 'Pass a tuple of tool names available to the agent; call list_available_tools() to discover options.'
• Rename tool to 'run_agentic_loop' or 'execute_tool_loop' to clarify the action. Add to description: 'Use this to run a multi-turn agentic loop where the LLM selects and invokes tools. Yields structured events for SSE encoding.'
• Add error handling section to description: 'On invalid tier, returns error with valid options. On missing tools, returns error listing available tools. On LLM failure, returns error with retry guidance.'