Crosswalk has 20 well-named tools with complete input schemas and descriptions. Naming follows verb_noun convention (setup_profile, add_resume, fetch_jobs, score_fit, tailor_resume, draft_application, submit_application, set_status, add_note, list_pipeline, schedule_workflow, run_workflow, list_workflows, delete_workflow, preview_application, apply_application). All tools have descriptions (avg ~100 chars, within baseline 34-392). Input schemas are properly typed with required fields marked. However, output schemas are NOT documented, the code returns JSON.stringify(result) with no schema specification, forcing LLMs to infer structure. Parameter descriptions are minimal (e.g., 'Resume ID to delete' lacks context on format or lifecycle). Error handling is absent, no recovery guidance, no categorization of retryable vs fatal errors. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk levels (WRITE, DESTRUCTIVE, READ_ONLY). Missing dependency hints (e.g., 'Call setup_profile first' or 'Requires resume to exist'). Composition is strong, tools chain well (draft_application uses resumeId, which list_resumes returns). Idempotency unclear for stateful operations like submit_application and set_status.

• Document output schemas for all 20 tools. For each tool, add a 'returns' section specifying field names, types, and descriptions. E.g., list_resumes returns [{id: string, label: string, createdAt: ISO8601}]. Use JSON Schema format in tool definitions.
• Add error handling with recovery guidance. Wrap tool implementations in try-catch; return structured errors like {error: 'resume_not_found', message: 'Resume ABC not found.', suggestion: 'Call list_resumes() to see available resumes or add_resume() to create one.'}
• Implement tool annotations in input schemas. Add readOnlyHint: true to fetch_jobs, list_resumes, list_pipeline, list_workflows, score_fit, top_matches, explain_fit, preview_application. Add destructiveHint: true to delete_resume, delete_workflow. Add idempotentHint: true to set_status, submit_application if they are idempotent.
• Enhance parameter descriptions with constraints and examples. E.g., 'resumeId (string, UUID format): Unique identifier of the resume. Deleting cascades to all applications referencing this resume.' For schedule_workflow, add 'schedule (string, cron format): Valid cron expression (e.g., '0 9 * * MON' for 9 AM Mondays).'
• Add pagination to list tools. Modify fetch_jobs, list_resumes, list_pipeline, list_workflows, top_matches to accept limit (1-100, default 20) and offset/cursor parameters. Return total count and next_cursor in responses. Document in tool descriptions.