Forest Soul Forge exposes 11 tools via fastmcp/STDIO. All tools have descriptions (10-80 chars, mostly adequate) and input schemas with typed parameters. However, critical gaps undermine production readiness: (1) Tool names lack action verbs, 'health', 'trust', 'route', 'bounties', 'quarantined', 'verify', 'why', 'nodes', 'agents', 'training_tasks', 'run_training' are mostly nouns or domain-specific jargon, not verb_noun patterns like 'get_health', 'list_trust_scores', 'recommend_route'. LLMs struggle to infer intent from noun-only names. (2) Parameter descriptions are minimal (5-40 chars) and lack actionable constraints, e.g., 'Optional node identifier' does not explain what a node is, valid formats, or when to use it. (3) Output schemas are not documented anywhere in the source, no indication of what fields each tool returns, forcing LLMs to guess. (4) No error handling guidance, tools return HTTP status codes but no recovery hints. (5) No pagination support despite tools like 'bounties' and 'quarantined' returning lists. Baseline: 90% of A+ tools start with action verbs; 100% document return types; 94% have descriptions; 100% of params have descriptions. This server meets ~40% of those baselines.

• Rename all tools to verb_noun pattern: get_health, list_trust_scores, recommend_route, list_bounties, list_quarantined_nodes, verify_trust_ledger, explain_trust_value, list_nodes, list_agents, list_training_tasks, run_training_suite. This matches LLM expectations and the 90% baseline of A+ tools.
• Document output schemas for every tool. Add a 'Returns' section to each tool description or a separate schema file. Example: 'Returns: {node: string, problem_class: string, trust_score: number (0-1), sample_count: integer}'. This enables LLMs to plan downstream calls.
• Expand parameter descriptions to 50-200 chars with explicit constraints. E.g., 'node: Optional node identifier (string, 1-100 chars, alphanumeric + underscore). If omitted, returns trust scores for all nodes.' Include format, range, enum, and when required.
• Add pagination to list-returning tools. Accept limit (1-100, default 20) and offset (default 0). Return {items: [...], total: integer, has_more: boolean, next_offset: integer}. This prevents context window exhaustion and matches pattern:paginated-result.
• Add error handling guidance to tool descriptions. E.g., 'If node not found, returns 404 with message: "Node 'X' not found. Available nodes: [list]. Call list_nodes() to discover." This guides LLM recovery per pattern:recovery-guide.