Loading ToolBench
oktsec plugin for OpenClaw - runtime security for AI agent tool calls
Single tool 'oktsec' with minimal definition. Tool name lacks action verb (should be 'get_security_status' or similar). Description is generic (11 chars: 'Security status and controls'), below 20-char minimum, scoring 0. Input schema exists but is severely underspecified: single 'args' string parameter with no enum constraint, no validation, no clear guidance on valid values ('status', 'logs', 'dashboard' mentioned only in description, not in schema). No output schema documented. No error handling guidance. Tool appears to be a plugin hook wrapper rather than a proper MCP tool, it intercepts OpenClaw events but does not expose a well-defined tool interface for LLM consumption. This is a security monitoring plugin, not an agent-callable tool.
Security status and controls
Tool name 'oktsec' lacks action verb. Should be 'get_security_status', 'check_security_controls', or similar to signal intent to LLM.
Tool description is 11 characters ('Security status and controls'), below 20-char minimum. Provides no context on WHEN to call, WHAT it returns, or HOW to use it.
Input schema has single 'args' parameter as free-form string with no enum constraint. Valid values ('status', 'logs', 'dashboard') are mentioned only in description, not enforced in schema. LLM will hallucinate invalid values.
No output schema documented. LLM cannot plan downstream actions or extract structured data from responses.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 20 | <=2025-11-25 | v2 |
Tool is primarily a plugin hook system (before_tool_call, after_tool_call, message_received, message_sending) rather than a callable MCP tool. It intercepts events but does not expose a proper tool interface for agent invocation.