v0.3.8

Fixes from Mira's Session 3 attempt: agents couldn't find the right tools.

Fixed

• Session 3 tool discoverability: Agents guessed brand_voice and brand_messaging (don't exist) instead of the real tools brand_extract_messaging and brand_compile_messaging. Both descriptions now include natural trigger phrases ("define brand voice", "brand messaging", "brand story", "start Session 3").
• Status next steps: When Session 3 is next, brand_status shows exact tool names and the recommended order rather than a generic pointer.

The root cause: agents search for tools by natural language matching, and "brand voice" is what humans say when they mean "brand_compile_messaging." The tool names are technical; the descriptions need to speak the agent's language.