Three tools with minimal schema documentation and generic descriptions. Tool names lack action verbs (budget, clean, explain are weak). Input schemas visible but lack type constraints and parameter descriptions. No output schemas documented. Descriptions are present but generic (34-72 chars, below 194-char baseline). No error handling guidance, no parameter validation rules, no idempotency markers. This is a CLI tool wrapped as MCP with minimal agent-optimization.

• Rename tools to verb_noun pattern: 'budget' → 'estimate_context_usage', 'clean' → 'remove_generated_files', 'explain' → 'describe_agent_types'. This lets LLMs parse intent from names.
• Add output schemas for each tool. E.g., budget returns {total_tokens: number, breakdown: {root: number, sidecars: object}, targets: string[]}. Document in tool description.
• Expand parameter descriptions to 50-100 chars with constraints. E.g., 'target: Array of target names to limit scope (e.g., ["web", "mobile"]). If empty, all targets included.'
• Add error handling guidance: 'If target not found, returns available targets. If clean fails mid-operation, returns partial_success with list of deleted files and failures.'
• Add destructive hint to 'clean' tool and implement dry-run parameter: 'dry_run: boolean (default true), preview files to be deleted without removing them. Set to false to execute.'