Server has 11 tools with basic descriptions and input schemas, but significant quality gaps. Most tool descriptions are present but lack actionable detail for LLM selection. Parameter descriptions are minimal or absent. No output schemas documented. Error handling is generic (500 errors with error.message). Tools like stream_command expose dangerous capabilities without confirmation patterns. Naming is generally clear (verb_noun), but composition issues exist: browse_directory and browse_fs are near-duplicates; preview_configuration and apply_configuration share identical parameter structures but lack dependency documentation.

• Document output schemas for all 11 tools. Example: 'Returns {ok: boolean, data: {environment: {node: string, python: string, ...}}, error?: string}'. This enables LLM planning.
• Add confirmation/dry-run pattern to stream_command. Require explicit confirmation=true or return a preview of the command before execution. Include timeout and error recovery guidance.
• Consolidate browse_directory and browse_fs into a single browse_project_directory tool with a mode parameter (native_dialog | in_app_browser). Reduce LLM decision overhead.
• Document parameter structures for agentSkills, ponytail, headroom, graphify, designSuite, cybersecurity, projectRules. Specify required vs optional keys, valid values, and examples.
• Add error recovery guidance to all tools. E.g., 'If targetDir does not exist, call browse_fs to select a valid directory.' Replace generic 500 errors with actionable messages.