A Kotlin-based agent runtime system that manages tool execution including builtin tools, custom Python tools, and MCP servers. Provides tool resolution, execution, and result classification for agentic workflows.
Critical gaps across all dimensions. Tool definitions lack input schemas (cannot verify parameter types/constraints), descriptions are minimal or generic (10-50 chars), and no output schemas are documented. The py_meta_tools description is an exception at ~400 chars but lacks parameter-level detail. No error handling guidance, no security annotations, no idempotency markers. Tools like execute_python, terminal, and screen_operation_* are high-risk (WRITE/DESTRUCTIVE) with no permission gates, validation rules, or recovery guidance. Naming is inconsistent: 'screen_operation_accessibility' vs 'screen_operation_shell' are vague; 'py_meta_tools' is cryptic. No evidence of input validation, rate limiting, or audit logging in the source. This server is not production-ready.
Execute Python code
Find installed applications
Launch an application
Load a skill
Memory management tool
Send notifications
Open a URI
Meta-tool for creating Python tools: a tool used to build other tools. Each created tool is a Python source defining `main(**kwargs)`; parameters come from basic type annotations on main (str/int/float/bool), the docstring becomes the description. Actions: list (show registered tools), read (return full code of one tool), write (create or replace a tool, validates syntax, main, and annotations), delete (remove a tool), test (run draft code or an existing tool with sample args without saving). Store the result of a run by printing from main; stdout is returned.
No input schemas visible for 12 of 13 tools. Cannot verify parameter types, constraints, or descriptions.
Descriptions are generic or missing (10-50 chars). LLMs cannot determine when to select these tools or what they return.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 22 | <=2025-11-25 | v2 |
Perform screen operations using accessibility APIs
Perform screen operations using shell commands
Take a screenshot
Execute terminal commands
View an image
High-risk tools (execute_python, terminal, screen_operation_*) marked WRITE/DESTRUCTIVE with no permission gates, dry-run support, or confirmation patterns. No audit logging visible.
Ambiguous tool names: 'screen_operation_accessibility' vs 'screen_operation_shell' do not clearly distinguish their purposes. 'py_meta_tools' is cryptic. Naming should start with action verbs and be self-documenting.
No output schemas documented. LLMs cannot plan downstream tool calls or extract required fields. No pagination, error classification, or recovery guidance visible.