RobloxForge demonstrates solid naming conventions (verb_noun pattern: rb_doctor, rb_docs_search, rb_project_analyze) and comprehensive descriptions (avg ~120 chars, all present). Schemas are well-formed with proper JSON Schema structure and type definitions. However, parameter descriptions are sparse or missing entirely, most parameters lack guidance on format, constraints, or when to use them. Output schemas are undocumented; LLMs cannot predict response structure. Error handling is minimal; tools return raw results without recovery guidance. No tool annotations (readOnlyHint/destructiveHint) despite clear risk classifications in metadata. The rb_verify_receipt tool accepts multiple boolean flags without explaining their interdependencies or what 'UNVERIFIED' means in practice.

• Document output schema for every tool. Example for rb_docs_search: 'Returns {results: [{title: string, path: string, excerpt: string}], total: integer, next_cursor?: string}'. This enables LLMs to extract path for rb_docs_read without guessing.
• Add parameter descriptions with format/constraint guidance. For 'path': 'Filesystem path to project root (absolute or relative to cwd). Must be a directory containing Rojo project file or Luau scripts. Returns error if not found or not readable.'
• Add tool annotations to schema. For rb_docs_update: add 'readOnlyHint: false' and 'destructiveHint: false' (it's safe to retry). For rb_verify_receipt: add 'idempotentHint: true' (same task+checks = same receipt).
• Add error recovery guidance. For rb_docs_search: 'If no results, try: (1) broader query without API names, (2) call rb_docs_update to refresh cache, (3) check rb_capabilities to confirm docs are available.'
• Clarify rb_verify_receipt semantics. Document: 'playtest_performed=true + screenshot_captured=true + screenshot_inspected=true → VERIFIED. Any false → UNVERIFIED. Unverified receipts are recorded but flagged for manual review.'