Desktop AI intelligent agent workbench: coding / documentation / automation, supporting multi-agent scheduling and plugin extension
Auraxis is a desktop agent workbench with 55 tools, but definition quality is severely compromised. Most tools lack visible input schemas in the source code provided. Only 2 tools (Pty, ReadDocument, WriteDocument) have explicit schema definitions shown; the remaining 52 are inferred from file references without visible parameter documentation. Descriptions are uniformly terse (5-15 chars), far below the 50-200 char baseline for LLM-optimized tools. No output schemas are documented. Error handling guidance is absent. The codebase shows tool registration in electron/contracts/tools.ts but the actual schema definitions are not visible in the provided source excerpt, making it impossible to verify parameter types, constraints, or descriptions for the majority of tools.
Invoke another agent
Ask the user for input
Execute shell commands
Create a new goal
Create a cron job
Delete a cron job
List cron jobs
List Google Drive files
Read Google Drive file
52 of 55 tools lack visible input schemas in source code. Only Pty, ReadDocument, WriteDocument show explicit parameter definitions. Remaining tools inferred from file references without verifiable schema structure.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 34 | 2026-07-28+ | v2 |
Edit file contents
Enter plan mode
Enter a worktree session
Exit plan mode
Get current goal
Find files matching glob patterns
Search file contents with pattern matching
Inspect runtime state
Interrupt another agent
Language Server Protocol operations
List available agents
List available skills
Mount a plugin
Edit notebook cells
Create a Notion page
Search Notion database
Persistent PTY session management (create, write, read, close, list, clear)
PowerShell execution
Ralph tool (purpose unclear from source)
Read file contents
Read document files (.docx, .xlsx, .pptx, .pdf)
Read a skill definition
Read spilled tool output from disk
Replan task execution
Generate a report
Review an artifact
Run code
Run a workflow
Send message to another agent
Read session event details
Search session events
Query session data
Get session trace
List Slack channels
Post a message to Slack
List tasks
Get task output
Stop a task
Write todo items
Unmount a plugin
Update current goal
Fetch content from web URLs
Search the web
Write file contents
Write document files (.docx, .xlsx, .pptx, .pdf)
Write a skill definition
All 55 tool descriptions are extremely terse (5-20 characters). Baseline for LLM-optimized tools is 50-200 chars. Descriptions like 'Execute shell commands' and 'Read file contents' lack context for when/why to use each tool and what they return.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields without knowing what each tool returns. This violates the response-shaper pattern.
Destructive tools (Bash, CronDelete, RunCode, Pwsh) lack confirmation/dry-run patterns and error recovery guidance. No indication of what happens on failure or how to undo side effects.
Tool 'Ralph' has unclear purpose ('purpose unclear from source'). Ambiguous tool names violate naming clarity patterns and force LLMs to guess intent.