Loading ToolBench
Durable, scalable AI assistant framework with Postgres backend and Temporal workflows. Supports multiple chat channels (WhatsApp, Telegram, Zalo, Slack, etc.) with agent loop, memory, skills, and MCP tool integration.
miu-bot exhibits significant definition quality gaps. Tool descriptions are present but often generic (10-50 chars). Parameter schemas are visible but lack depth: most parameters have only type and description, missing constraints like enums, min/max, or format specifications. Critical issues: (1) exec tool accepts arbitrary shell commands with no validation guidance; (2) spawn and cron tools lack clear output schemas; (3) no tool declares permissions or security scopes; (4) error handling is not documented; (5) several tools (message, react, spawn, cron) have minimal descriptions (<50 chars). Naming is generally clear (verb_noun pattern), but composition issues exist: spawn and cron are high-risk tools with minimal safeguards. No evidence of idempotency guarantees or dry-run patterns for destructive operations.
Execute code using Claude Code SDK for complex computations
Schedule a recurring task using cron expressions
Edit a file by replacing a section with new content
Execute shell commands in the workspace
List directory contents
Send a message to a chat channel
Add emoji reactions to messages in chat channels
Read file contents from the workspace
exec tool accepts arbitrary shell commands with no input validation, sanitization guidance, or error recovery hints. No mention of command injection risks or safe usage patterns.
spawn and cron tools lack output schema documentation and clear error handling. spawn spawns subagents asynchronously but does not document return value, success/failure indicators, or how to track completion.
No tool declares permissions, scopes, or security requirements. Destructive tools (write_file, edit_file, exec, message, spawn, cron) lack permission gates or audit trail guidance.
message, spawn, cron, and react tools have descriptions under 50 characters, providing insufficient context for LLM tool selection.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 45 | <=2025-11-25 | v2 |
Spawn a subagent to handle a task asynchronously
Fetch and parse web page content
Search the web using Brave Search API
Write content to a file in the workspace
No evidence of idempotency guarantees, dry-run patterns, or confirmation steps for irreversible operations (exec, write_file, edit_file, message, spawn, cron). Agents cannot safely retry without risk of duplicate side effects.