Loading ToolBench
Autonomous coding agent in your terminal — background vex tasks, 7-phase pipeline, multi-provider LLMs, live dashboard. Rust TUI.
d3vx has 25 tools with significant definition gaps. Most tools lack visible input schemas in the provided source code, only 6 tools (Write, Edit, LaunchAgent, SendNudge, KillSession, GetStatus, BatchLaunch) show explicit schema definitions. Descriptions are present but generic (10-50 chars for many tools), failing to explain WHEN to use each tool or what it returns. No output schemas documented. Error handling is absent, no recovery guidance or actionable error messages visible. Tool composition is poor: Bash and Write are destructive but lack confirmation patterns; coordination tools (coord_*) lack clear parameter documentation. The codebase shows approval flows and tool execution logic, but this is not reflected in tool definitions exposed to LLMs.
Execute shell commands
Launch multiple agent sessions in parallel to process a batch of issues.
Mark a task as complete
Edit file contents by replacing text
Get detailed status for a specific agent session.
Find files matching a glob pattern
Search for text patterns in files
Terminate a stuck or unwanted agent session with a reason.
19 of 25 tools have no visible input schema. Only Write, Edit, LaunchAgent, SendNudge, KillSession, GetStatus, BatchLaunch show explicit schemas. This violates the fundamental requirement that LLMs must know what parameters a tool accepts.
Destructive tools (Bash, Write, Edit, MultiEdit, KillSession) lack confirmation or dry-run patterns. No evidence of approval flow exposure to LLM. Agents can delete files or kill sessions without user confirmation.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 32 | <=2025-11-25 | v2 |
Spawn a new agent session with a prompt. Optionally specify a branch name.
List all active agent sessions and their current status.
Edit multiple files in a single operation
Read file contents
Send a nudge message to a running agent session to guide or redirect it.
Record output from a task execution
Stop a running task
Internal reasoning tool for agent planning
Write task items to a todo list
Fetch content from web URLs
Search the web for information
Write content to a file
Claim a task from the coordination board
Mark a task as complete on the coordination board
Retrieve and drain messages from this agent's coordination inbox
List all ready tasks from the coordination board
Send a message to another agent via the coordination inbox
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields. E.g., does ListSessions return session_id, status, created_at? Unknown.
Descriptions are generic and under 50 chars for 19 tools. E.g., 'Execute shell commands' (23 chars) does not explain when to use Bash vs Think, what it returns, or error handling. Fails LLM-optimization baseline of 50-200 chars.
Coordination tools (coord_*) use snake_case naming inconsistent with camelCase tools (LaunchAgent, KillSession). LLMs conflate similar names, this inconsistency increases selection errors.