Two tools with clear verb-noun naming (codex_list_sessions, codex_read_session) and solid descriptions (180-220 chars, meeting baseline). Input schemas present with typed parameters and descriptions. However, output schemas are undocumented, LLMs cannot predict return structure. Error handling is minimal (only 'not found' cases). No pagination guidance despite list_sessions returning up to 20 items. Tool descriptions lack dependency hints and prerequisites. Parameters lack format constraints (e.g., session_id UUID format not enforced in description).

• Document output schemas for both tools. E.g., codex_list_sessions returns {sessions: [{id, title, model, cwd, created_at, updated_at}], total_count, has_more}. codex_read_session returns {session_id, messages: [{role, text, timestamp, ...}]}.
• Add format constraints to descriptions: 'session_id: UUID string (e.g., 019d6cdb-40ca-7d52-82e3-b13af6c88301)', 'limit: 1-100 (default 20)', 'max_messages: 1-500 (default 50)'.
• Add pagination guidance: 'Returns up to 20 most recent sessions. For older sessions, use keyword filter or implement offset-based pagination in a future version.'
• Enhance error handling with actionable messages: 'Session not found. Call codex_list_sessions to see available sessions.' 'Invalid session_id format, must be a UUID.'
• Add dependency hints in codex_read_session description: 'Requires a valid session_id from codex_list_sessions. Call that tool first to discover available sessions.'