LINTelligent exposes 2 tools with basic descriptions but critical structural gaps. SubmitReviewRequest and GetReviewDetails have descriptions (194 and 156 chars respectively, within baseline 34-392 range), but parameter schemas lack proper type definitions in the visible code. Input parameters are documented via C# [Description] attributes, but JSON Schema structure is not explicitly visible. Both tools return dynamic/string types without documented output schemas. Error handling is minimal (bare catch blocks returning generic strings). No tool annotations (readOnlyHint, destructiveHint, idempotentHint). The mutual exclusivity constraint between codeSnippet and gitHubUserContentFileUrl is documented in descriptions but not enforced via schema. No pagination, no structured output, no recovery guidance in errors.

• Document output schemas explicitly. For SubmitReviewRequest, define: {reviewId: string (UUID), status: 'Pending'|'Processing'|'Completed'|'Failed'}. For GetReviewDetails, define: {reviewId: string, status: string, report?: string, language: string, codeSnippet?: string}.
• Enforce mutual exclusivity in schema: mark codeSnippet and gitHubUserContentFileUrl as required: false, then add a schema-level constraint or validation that rejects requests where both are null or both are non-null.
• Replace generic error strings with structured error responses: {error: string, code: 'INVALID_INPUT'|'REVIEW_NOT_FOUND'|'SERVICE_ERROR', recoveryHint: string}. E.g., 'Invalid input: both codeSnippet and gitHubUserContentFileUrl provided. Provide exactly one.'
• Add tool annotations: mark SubmitReviewRequest with destructiveHint=true (creates resource, triggers webhooks); mark GetReviewDetails with readOnlyHint=true.
• Refactor polling logic: wrap GetReviewDetails polling inside SubmitReviewRequest or create a separate WaitForReviewCompletion tool that handles retries internally. Expose a simple 'submit and wait' interface to the agent.