Loading ToolBench
Reusable AI agent skills, coding standards, and safety runner toolkit for multi-engine Prolog
This Prolog toolkit exposes 17 tools with significant definition gaps. Most tools lack proper descriptions (many under 20 chars or generic). Input schemas are present but incomplete: many parameters use non-standard types ('chars', 'atom', 'term', 'any') that are Prolog-specific and not JSON Schema compliant. No output schemas documented. Tools like 'mcp_call_tool' and 'mcp_handle_request' are meta-tools that expose internal dispatch logic rather than user-facing capabilities. Error handling is minimal. The toolkit is domain-specific (Prolog) but not optimized for LLM agent consumption.
Dispatches a structured request to the corresponding skill handler
Dynamically registers a new capability bundle for the agent
Find all skills that provide a given capability
Generates content for a new Prolog module stub
Generates file specifications for standard project template layout
Generates abstract file specification list for a new Prolog project
Generates POSIX-compatible bash initialization script specification
Executes an MCP tool call with arguments and returns structured output
Non-JSON-Schema parameter types: 'chars', 'atom', 'term', 'any' are Prolog-native types, not JSON Schema primitives. LLMs cannot validate or reason about these types. Must map to JSON Schema: string, object, array, etc.
No output schemas documented. Tools return Prolog terms (e.g., 'result(Answer)', 'file(Path, ContentChars)') without specifying what fields LLMs should expect. Agents cannot plan downstream calls or extract data reliably.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 35 | 2026-07-28+ | v2 |
Top-level JSON-RPC / MCP request dispatcher
Returns all registered MCP tools in the server
Evaluates a pure arithmetic constraint or expression
Tests declarative unification between two terms
Registered skills and their respective capability tags
Query skill capabilities
Relative path to the skill's SKILL.md definition file
Validates a relational claim generated by an LLM against the ground truth
Validates and solves task scheduling constraints proposed by an LLM within MaxHours
Meta-tools expose internal dispatch logic (mcp_call_tool, mcp_handle_request, mcp_list_tools) instead of domain-specific capabilities. These are framework plumbing, not user-facing tools. Remove or hide them from the MCP tool list.
Descriptions are generic or incomplete. Examples: 'Registered skills and their respective capability tags' (tool 'skill'), 'Relative path to the skill's SKILL.md definition file' (tool 'skill_path'). No guidance on WHEN to call these tools or WHAT to do with results.
No error handling guidance. Tools like 'verify_fact' and 'verify_schedule' do not document what happens on validation failure, what error messages are returned, or how agents should recover. Agents cannot self-correct.