Loading ToolBench
Four tools with strong, detailed descriptions (50-200 char range per pattern baseline). All tools have explicit input schemas with types and descriptions. Tool names are action-verb based (roll, validate). However, output schemas are declared as `output_schema=None`, which violates the documented-output-schema pattern. Error handling returns `{"error": str(e)}` but lacks recovery guidance or categorization. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite all being read-only operations. Parameter descriptions are thorough but could be more concise for LLM parsing.
Roll dice using standard RPG notation and return the total result. PURPOSE: Use this tool when you need a quick dice roll with just the total numeric result and a formatted output string. This is the most common tool for standard RPG scenarios like attack rolls, ability checks, saving throws, and damage calculations. WHEN TO USE: - Making attack rolls (1d20+5) - Rolling damage (2d6+3, 8d6) - Ability checks and saves (1d20+modifier) - Any roll where you just need the total - Quick single rolls without needing detailed breakdown WHEN NOT TO USE: - If you need to see individual die results, use roll_detailed instead - If you need to make multiple different rolls, use roll_batch instead - If you just want to validate syntax without rolling, use validate_syntax instead SUPPORTED NOTATION: - Basic: 1d20, 3d6, d20 (equivalent to 1d20) - Arithmetic: 1d20+5, 2d6-1, 3d6*2, 1d20/2 - Keep highest/lowest: 4d6kh3 (keep highest 3), 4d6kl1 (keep lowest 1) - Drop: 4d6p1 (drop lowest 1, equivalent to kh3) - Reroll: 1d20rr<10 (reroll until ≥10), 1d20ro1 (reroll 1s once) - Exploding: 1d6e (explode on max), 1d6e6 (explode on 6) - Min/Max: 1d20mi10 (minimum 10), 1d20ma20 (maximum 20) - Complex: (1d4+1)*2, (1d6, 3, 2d4) COMMON EXAMPLES: - D&D Attack Roll: "1d20+5" → Check if hits AC - Ability Score: "4d6kh3" → Roll 4d6, keep highest 3 - Fireball Damage: "8d6" → Total fire damage - Sneak Attack: "1d20+7" for attack, then "1d6+3d6" for damage - Critical Hit: "2d8+6" (doubled dice) - With Advantage: "2d20kh1" (roll 2d20, keep highest)
Roll multiple different dice expressions in a single efficient operation. PURPOSE: Use this tool when you need to make several different dice rolls at once. Instead of calling the roll tool multiple times, batch them together for efficiency and to receive all results in a single structured response. Perfect for combat rounds, character creation, or any scenario requiring multiple distinct rolls. WHEN TO USE: - Combat rounds: Roll attack, damage, and saving throws together - Character creation: Roll all 6 ability scores at once (6x "4d6kh3") - Multiple checks: Roll several different ability checks simultaneously - Efficiency: When you need 3+ different rolls, batching is more efficient - Organized results: Get all rolls back in a structured array - Mixed roll types: Different expressions with different modifiers/mechanics WHEN NOT TO USE: - Single roll (use roll instead) - Need detailed AST for each roll (call roll_detailed multiple times) - Rolls are dependent on each other (roll one, check result, then decide next roll) ERROR HANDLING: Each roll in the batch is evaluated independently. If one expression fails, it returns an error entry, but other rolls continue. Check the "success" field for each result: - success: true → Roll succeeded, has "total" and "result" - success: false → Roll failed, has "error" message COMMON EXAMPLES: Combat Round: ["1d20+5", "2d6+3", "1d20+2"] → Attack roll, damage roll, saving throw Character Creation (6 ability scores): ["4d6kh3", "4d6kh3", "4d6kh3", "4d6kh3", "4d6kh3", "4d6kh3"] → Roll all six ability scores at once Party Ability Checks: ["1d20+3", "1d20+1", "1d20+5", "1d20+0"] → Perception checks for 4 party members Mixed Damage Types: ["2d6", "1d8", "3d4+2"] → Sword damage, fire damage, poison damage With Advantage/Disadvantage: ["2d20kh1+5", "2d20kl1+2"] → Attack with advantage, save with disadvantage
Output schemas declared as `output_schema=None` for all tools. LLMs cannot infer return structure; must parse responses blindly. Violates pattern:tool and pattern:response-shaper.
Error handling returns bare `{"error": str(e)}` without recovery guidance. LLMs cannot determine if error is retryable, user-fixable, or fatal. Missing pattern:recovery-guide and pattern:error-classification.
No tool annotations despite all tools being read-only. Missing readOnlyHint declarations that would signal to agents these are safe to call without side effects.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 62 | 2026-07-28+ | v2 |
Roll dice with comprehensive breakdown including AST structure and individual die values. PURPOSE: Use this tool when you need to see exactly what happened during a dice roll, including which specific values each die showed, which dice were kept or dropped, and the complete Abstract Syntax Tree (AST) structure of the roll. Perfect for debugging, transparency, and understanding complex roll mechanics. WHEN TO USE: - Character creation: See which specific values you rolled for ability scores (4d6kh3) - Debugging: Understand why a complex roll gave a particular result - Transparency: Show players exactly what each die rolled - Learning: Understand how keep/drop/reroll mechanics work - Complex rolls: Analyze multi-step expressions with nested operations - Statistical analysis: See individual die distributions WHEN NOT TO USE: - Simple rolls where you only need the total (use roll instead) - Multiple different rolls (use roll_batch instead) - You don't need the AST complexity for basic gameplay AST STRUCTURE EXPLANATION: The "ast" field contains a tree structure representing the roll: - "type": Type of node (Expression, Dice, Number, etc.) - "total": Numeric value of this node - "number": Number of dice rolled (for Dice nodes) - "size": Dice size, e.g., 6 for d6 (for Dice nodes) - "values": Array of individual die results with "value" and "kept" (true/false) - "children": Nested operations (for complex expressions) INDIVIDUAL DIE RESULTS: The "values" array shows each die roll: - {"value": 4, "kept": true} → This die was kept in the final result - {"value": 2, "kept": false} → This die was dropped (e.g., in 4d6kh3) COMMON EXAMPLES: - Character Stats: "4d6kh3" → See which 3 of 4 dice were kept - Advantage: "2d20kh1" → See both d20 results, know which was kept - Reroll Analysis: "4d6rr1" → See which dice were rerolled - Exploding Dice: "3d6e" → See which dice exploded and their chain - Complex Expression: "(1d4+1)*2d6kh1" → Full breakdown of nested operations
Validate a dice expression's syntax without actually rolling any dice. PURPOSE: Use this tool to check if a dice expression is syntactically correct and can be evaluated before actually rolling it. Perfect for validating user input, testing complex expressions, or checking syntax in application development. This is a read-only operation that doesn't generate random results.
Parameter descriptions are verbose (100+ chars) and include extensive examples. Trim to 50-100 chars and move examples to separate documentation.
roll_batch lacks per-item error handling. If one expression fails, response structure unclear, does it return partial results with success/failure flags, or a blanket error? Code shows independent evaluation but response format not documented.