Three tools with complete input schemas and descriptions. Tool names follow verb_noun pattern (copilot_exec, codex_exec, consult). Descriptions are detailed (150-400 chars) and explain WHEN to use each tool. However, output schemas are entirely undocumented, LLMs cannot plan downstream calls or extract structured data. Parameter descriptions are present but lack format constraints (e.g., no regex for cwd paths, no min/max for timeout). Error handling is present in implementation but not declared in tool definitions. Security considerations (sandbox modes, read-only defaults) are well-documented in descriptions but no permission gates or audit trail patterns visible.

• Document output schemas for all three tools. Specify that copilot_exec and codex_exec return {text: string, metadata: {model, sandbox, exit_code}}, and consult returns {results: [{engine: string, text: string, error?: string}], failures: [{engine, reason}]}.
• Add format constraints to 'cwd' parameter: 'Absolute filesystem path (must start with / on Unix or C:\ on Windows). Defaults to server's cwd.'
• Enhance error descriptions in tool definitions. E.g., 'Returns error if Copilot CLI is not installed or not authenticated. Retryable: run 'copilot login' and retry. Non-retryable: ensure @github/copilot-cli is installed.'
• Add 'timeout_sec' as an optional parameter to copilot_exec and codex_exec so agents can override defaults per-call (e.g., for long-running tasks).
• Clarify consult's 'order' parameter: 'In mode "all", order is ignored (all engines run in parallel). In mode "first", order determines the fallback chain (first engine tried first, last engine tried last).'