Loading ToolBench
26 tools with basic schemas and descriptions, but significant gaps in parameter documentation and output schema clarity. Most tools have 1-4 parameters with type definitions (good), but descriptions are often generic or missing context. No tool annotations (readOnlyHint/destructiveHint) despite clear risk classifications. Output schemas are not documented, LLMs cannot predict response structure. Error handling is minimal; most tools return raw error strings without recovery guidance. Naming is verb-forward (list-, create-, kill-, send-) which is correct, but some tools conflate concerns (run-agent, recover-codex-pane mix agent selection with execution).
Capture content from a tmux pane with configurable lines count and optional color preservation
Change directory in a tmux pane
Copy files from repository root into a worktree
Create a new tmux session
Create a new window in a tmux session
Ensure a git worktree exists, creating it if necessary
Execute a command in a tmux pane and capture output
Find a tmux session by name
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite explicit risk classifications. Tools marked DESTRUCTIVE (kill-session, kill-window, kill-pane, remove-worktree) should declare destructiveHint=true to warn agents.
Output schemas not documented. LLMs cannot predict response structure (e.g., what fields does list-sessions return? Is it {sessions: [...]} or {data: [...]}?). All tools return JSON but structure is opaque.
Error handling returns raw error strings without recovery guidance. E.g., 'Error listing tmux sessions: <error>' tells LLM nothing about whether to retry, ask user, or give up. No error classification (retryable vs fatal).
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 59 | <=2025-11-25 | v2 |
Get the ID of the currently active tmux pane
Get the result of a previously executed command
Get the address (session, window, pane indices) of a tmux pane
Kill a tmux pane by ID
Kill a tmux session by ID
Kill a tmux window by ID
List panes in a tmux window
List all active tmux sessions
List windows in a tmux session
List all git worktrees in a repository
Detect and recover from codex startup issues
Remove a git worktree
Rename a tmux pane
Run an agent (codex, claude, or gemini) in a tmux pane with optional prompt
Select/focus a tmux pane
Send keys/text to a tmux pane
Set environment variables in a tmux pane
Split a tmux pane horizontally or vertically
run-agent and recover-codex-pane conflate agent selection with execution. 'run-agent' accepts optional agent type but defaults to undefined behavior. Should split into separate tools or require agent parameter.
Parameter descriptions lack format/constraint details. E.g., 'lines' in capture-pane is described as 'Number of lines to capture' but no min/max bounds. 'direction' in split-pane says 'horizontal or vertical' but not as enum constraint.