Protocol-aware MCP runtime firewall with verified execution, approvals, and tamper-evident evidence.
Three tools with complete input schemas and risk annotations, but critical gaps in parameter descriptions and output documentation. Tool names follow verb_noun convention (kb.search, repo.write_file, ops.restart_service). All have descriptions (10-60 chars), but descriptions lack actionable context about when to use each tool and what they return. Input schemas are well-formed with type constraints and length limits, but parameter descriptions are entirely absent, LLMs cannot infer what 'query', 'path', 'content', or 'service' mean beyond the name. No output schemas documented. Error handling and recovery guidance missing. Security annotations present (risk levels) but no scope declarations or permission gates visible in tool definitions.
Search the approved internal operations knowledge base.
Request a production service restart.
Stage a file change in the isolated portfolio sandbox.
All three tools lack parameter descriptions. 'query', 'path', 'content', 'service' parameters have no guidance on format, constraints, or expected values. LLMs cannot infer meaning from names alone.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields (e.g., does kb.search return result IDs for follow-up calls?).
Tool descriptions are too brief (25-60 chars) and lack context. 'Search the approved internal operations knowledge base' does not explain when to use kb.search vs other discovery tools, what format results take, or prerequisites.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 56 | 2025-06-18+ | v2 |
Destructive tool (ops.restart_service) lacks confirmation/dry-run pattern. No guidance on recovery if restart fails or causes cascading outages.
repo.write_file accepts 'path' and 'content' with length limits (180, 8000 chars) but no description of path format (relative? absolute? sandboxed?), file type restrictions, or what 'isolated portfolio sandbox' means to the LLM.