AgentDesk provides 21 tools with complete input schemas and descriptions for all tools and parameters. Naming follows verb_noun conventions (open_workspace, read, write, grep, bash, browser_navigate, etc.). However, descriptions are often terse (10-50 chars) and lack actionable context for LLM selection. Output schemas are not documented, LLMs cannot predict response structure. Error handling is minimal; no recovery guidance or categorization. Tool composition has issues: bash and exec_command are near-duplicates; browser tools lack idempotency guarantees. Parameter descriptions exist but lack constraints (e.g., timeout max 300s mentioned in bash but not enforced in schema). No enum constraints on categorical parameters. Security: workspaceId is passed as a parameter (potential logging risk); no explicit permission gates documented in tool definitions.

• Document output schemas for all tools. Include field names, types, and examples. E.g., read() returns {content: string, size: number, offset: number}. This enables LLMs to plan downstream calls.
• Expand tool descriptions to 100-200 chars. Add WHEN to use it and WHAT it returns. E.g., 'List files matching a glob pattern. Use this to discover files before reading. Returns array of {path, size, type}.'
• Consolidate bash and exec_command into a single tool with clear naming (e.g., execute_shell_command). Document differences if both must exist (e.g., one supports stdin, one does not).
• Add error handling guidance to each tool. E.g., read() should document: 'If file not found, try glob() first to verify path. If permission denied, check workspace permissions.'
• Add enum constraints to categorical parameters. E.g., kill_process.force should be {type: boolean, enum: [true, false]} with description 'Use force=true only if normal termination fails.' Browser tools should document valid selector syntax.