Loading ToolBench
Autonomous AI Mesh Network — Multi-provider LLM routing with MCP integration. Unified control center exposing routing, memory, shell commands, browser crawls, screen OCR, anti-bot web scraping, security scans, and Composio tools.
M4STCLAW exposes 18 tools with basic schemas and descriptions, but exhibits critical gaps in parameter documentation, output schema clarity, and error handling guidance. Most tools lack descriptions of what they return, making it difficult for LLMs to plan downstream calls. Parameter descriptions are present but often generic (e.g., 'System command to execute' for shell_execute lacks safety constraints). No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite high-risk operations like shell_execute and mesh_run. Error handling is absent, no guidance on retryability, recovery steps, or actionable error messages. Naming is generally clear (verb_noun pattern), but composition issues exist: shell_execute and mesh_run are irreversible with no confirmation pattern; memory tools lack output schema documentation.
Navigates to websites, interacts with elements, and captures screenshots using Playwright.
Executes a Composio action (connects to Slack, Linear, Notion, Jira) with parameters.
Queries the optimal LLM task chain with automatic provider fallback routing.
Records a task execution log into episodic memory (T2).
Stores a permanent document embedding into ChromaDB semantic memory (T3).
Gets a value from current working memory (T1).
Queries ChromaDB semantic memory (T3) for relevant context.
No output schemas documented for any tool. LLMs cannot infer what fields to expect in responses, forcing them to guess at downstream tool inputs and risking broken tool chains.
Irreversible operations (shell_execute, mesh_run) lack confirmation or dry-run patterns. Agents can execute destructive commands without safeguards, risking data loss or system compromise.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) present. LLMs cannot distinguish safe read-only tools from destructive ones, leading to unsafe retry logic and unintended side effects.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 57 | <=2025-11-25 | v2 |
Searches cross-session episodic memory logs (T2) for matching keywords.
Sets a value in the current working memory (T1).
Executes the multi-agent mesh loop to generate, audit, and test a solution.
Runs a port scan against a whitelisted in-scope local target.
Runs nuclei vulnerability checks against a whitelisted local target.
Queries Shodan for public information about the host IP.
Classifies a user query into one of the 9 specialized task chains (speed, reasoning, code, etc.).
Returns the rotation and cooldown status of configured provider API keys.
Performs an anti-bot static crawl of raw text contents from target URLs.
Executes developer system commands securely within the sandbox boundary.
Captures and analyzes display monitor screenshots with OCR and multimodal vision model routes.
Error handling absent across all tools. No guidance on retryability, recovery steps, or actionable error messages. LLMs receive raw errors with no context for next steps.
shell_execute parameter 'command' lacks safety constraints in description. No mention of sandbox boundaries, allowed commands, or injection prevention. LLMs may attempt arbitrary system calls.