Loading ToolBench
Dear User MCP has 11 tools with inconsistent quality. Most tools lack input schemas entirely (audit, onboard, security, history, share, wrapped, implement_recommendation, dismiss_recommendation, collab have no visible schema definitions in source). Descriptions are present but generic (avg ~80 chars, well below the 194-char baseline). Tool naming is reasonable (verb-noun pattern: analyze, audit, onboard, security, feedback, share, wrapped, implement_recommendation, dismiss_recommendation, history, collab) but several tools conflate multiple concerns (e.g., 'implement_recommendation' and 'dismiss_recommendation' are state-mutation tools without dry-run or confirmation patterns). No output schemas documented. Error handling is absent from visible code. The 'collab' tool is a direct alias of 'analyze', violates single-responsibility principle. Parameter descriptions exist for some tools (analyze, feedback, history, wrapped) but are minimal. No evidence of permission gates, audit trails, or secret injection patterns.
Scan your Claude Code setup and generate a collaboration report
Deep audit of CLAUDE.md rules, hooks, and skills for quality and safety
Alias for analyze tool - generate collaboration report
Dismiss a pending recommendation without implementing it
Send feedback, corrections, or insights to improve the collaboration system
Retrieve session history and past collaboration reports
Execute a pending recommendation (install MCP server, add hook, merge settings)
8 of 11 tools have no visible input schema definitions (audit, onboard, security, share, and others declared with empty {} input). Cannot verify parameter types, constraints, or LLM-parseable structure.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract structured data. Violates pattern:tool and pattern:response-shaper.
'collab' is a direct alias of 'analyze', violates single-responsibility principle. LLMs waste reasoning cycles deciding between two identical tools. Merge into one canonical tool.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 48 | 2026-07-28+ | v2 |
First-time setup wizard for Dear User preferences and configuration
Security audit of your setup for exposed secrets, unsafe patterns, and compliance issues
Generate a shareable report card with collaboration metrics
Spotify Wrapped-style collaboration summary with moments and archetypes
State-mutation tools (implement_recommendation, dismiss_recommendation, onboard, feedback) lack dry-run, confirmation, or idempotency hints. Agents can accidentally execute irreversible actions without safeguards.
No error handling guidance visible. Tools do not return actionable recovery steps (e.g., 'Try search_users() first' or 'Invalid status: must be one of...'). Violates pattern:recovery-guide.