A multi-agent system with MCP (Model Context Protocol) support, featuring native tools for filesystem operations, shell execution, web fetching, and task scheduling. Integrates with multiple LLM providers and supports Telegram transport.
TuskBot exposes 4 tools with minimal schema documentation and generic descriptions. Tool names lack action verbs (filesystem, shell, fetch, schedule vs. read_file, execute_command, fetch_url, create_schedule). No visible input parameter schemas in the provided code, only tool names and brief descriptions. Descriptions are generic (10-80 chars) and do not explain WHEN to use each tool, what parameters are required, or what the output structure is. No evidence of parameter type definitions, enums, or constraints. Error handling and recovery guidance absent. Security-critical tools (shell, filesystem) lack permission gates or input validation documentation.
Web fetching tool for retrieving content from HTTP/HTTPS URLs
Filesystem operations tool for reading, writing, and managing files within the runtime path
Task scheduling tool for creating and managing scheduled tasks within the swarm service
Shell execution tool for running system commands within the runtime path
No input schemas visible for any tool. Cannot verify parameter types, required fields, enums, or constraints.
Tool names lack action verbs. 'filesystem' and 'shell' are nouns, not verbs. Should be 'read_file', 'write_file', 'execute_command', 'fetch_url', 'create_schedule'. LLMs infer intent from verb-noun naming.
Descriptions are generic and under 100 chars. Do not explain WHEN to use each tool, what parameters are required, or output structure. Baseline for A+ tools is 194 chars with clear WHAT/WHEN/RETURN guidance.
Destructive tools (shell, filesystem) lack permission gates, input validation guidance, or confirmation patterns. No evidence of audit logging or scope declarations.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 29 | 2026-07-28+ | v2 |
No error handling or recovery guidance documented. LLMs cannot determine if failures are retryable, user-fixable, or fatal.