Strong foundation with well-structured tool definitions and comprehensive parameter documentation. All 7 tools have clear descriptions (194-400+ chars, well above 20-char minimum). Input schemas are complete with typed parameters and descriptions. Tool naming follows verb_noun convention (project_use, memory_remember, memory_search, etc.). However, output schemas are not explicitly documented in the visible code, and error handling lacks recovery guidance. No tool annotations (readOnlyHint/destructiveHint) despite clear risk classifications. Descriptions are production-quality and load-bearing (as noted in code comments), but could be more concise for LLM token efficiency.

• Add explicit output schema documentation for all 7 tools. Include field names, types, and what each field represents. Example: 'Returns {project_id: string, slug: string, display_name: string, created_at: ISO8601}'.
• Expose tool annotations in schema: add readOnlyHint=true to memory_search/memory_history/memory_context; add destructiveHint=true to memory_forget; add idempotentHint=true to memory_remember (deduplication behavior).
• Expand memory_search and memory_context descriptions to 150+ chars. Explain when to use each: 'memory_search finds specific memories by keyword/tag; memory_context generates a token-budgeted brief of the most relevant memories for context injection.'
• Add error recovery guidance in tool descriptions. E.g., 'If project_use returns not found, verify the slug or git_remote. Call with create=true and an explicit slug to create a new project.'
• Implement a confirm_memory_forget tool or add a dry_run parameter to memory_forget to prevent accidental permanent deletion.

Inferred effective spec: 2025-06-18+.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry