Loading ToolBench
A Personal AI OS scaffold implementing a zero-context kernel with operational store, security gates, and process management. Milestone M0/M1 implementation with frozen cross-module contracts.
RAG-OS exposes only 1 tool (gate-check) with a minimal schema. The tool has a basic description ('Ask the gate whether a path is readable') and a single string parameter with description. However, the schema lacks critical details: no output schema is documented, no error handling guidance, no enum constraints, and the description is only 45 characters, below the 50-200 char baseline for LLM-optimized descriptions. The tool name 'gate-check' uses a hyphen (non-standard; verb_noun convention prefers 'check_gate' or 'gate_check_access'). No tool annotations (readOnlyHint, destructiveHint, idempotentHint) are present. The server appears to be a kernel/daemon scaffold with minimal MCP integration, tool definitions are inferred from code rather than explicitly registered in an MCP handler.
Ask the gate whether a path is readable
No output schema documented. LLMs cannot infer what gate-check returns (decision, reason, metadata, etc.), forcing them to guess downstream behavior.
Tool description is only 45 characters. Baseline is 50-200 chars. Lacks WHEN to use it, WHAT it returns, and prerequisites. LLMs cannot determine when to select this tool over alternatives.
No error handling guidance. If the path is unresolvable or access is denied, what does the tool return? How should the LLM recover? No actionable error messages.
Tool name uses hyphen (gate-check) instead of verb_noun convention (check_gate or gate_check_access). Inconsistent with 90% of production tools.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 25 | 2026-07-28+ | v2 |
No tool annotations. The tool is read-only (Risk: READ_ONLY noted in metadata), but no readOnlyHint is declared in the schema. Agents cannot infer safety properties.