Loading ToolBench
Navigate any Claude Code (and future Codex/Gemini) session as a numbered tree in your terminal and resume from any node.
Five tools with complete schemas and descriptions, but quality is uneven. All tools have descriptions (10 - 200 chars, within baseline range) and input schemas with typed parameters. However, descriptions lack LLM-optimized guidance on WHEN to use each tool vs. alternatives, and error handling is minimal. Output schemas are not documented. Parameter descriptions are present but terse (avg ~40 chars vs. baseline 72). No tool annotations (readOnlyHint/destructiveHint). agent_tree_unstar is the only write operation but lacks confirmation/dry-run pattern. Naming is clear (verb_noun style: list, snapshot, picks, diff, unstar) and distinguishes tools well. No security issues detected (no secrets in params). Composition is sound, tools chain naturally (list → snapshot/diff → unstar). Overall: functional but below production baseline for LLM-optimized agent tools.
Summarise what happened between two nodes — event span, segments crossed, files touched, tools used.
Render a numbered ASCII tree of an agent (Claude Code, Codex, Gemini) session. Phase headers (user prompts) at depth 1, sub-actions at depth 2. Returns text suitable for chat display.
List every recorded pick across every session. Shows session id × node × mode × timestamp.
Get the resume markdown for a specific node so the user can paste it into a new claude session. Records the pick for future ⭐ display.
Remove ⭐ from a previously picked node.
Output schemas not documented. LLMs cannot infer what fields each tool returns, forcing them to guess downstream field names and risk broken chains.
Descriptions lack WHEN-to-use guidance. E.g., agent_tree_snapshot says 'Get resume markdown' but doesn't explain when to call it vs. agent_tree_list or agent_tree_diff. LLMs waste reasoning cycles deciding between similar tools.
agent_tree_unstar (write operation) lacks confirmation/dry-run pattern. No way for agents to preview what will be removed before executing. Risks accidental data loss.
No tool annotations (readOnlyHint/destructiveHint/idempotentHint). Agents cannot distinguish safe (list, snapshot, picks, diff) from unsafe (unstar) operations without reading descriptions.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 57 | <=2025-11-25 | v2 |
Parameter descriptions are terse (avg ~40 chars vs. baseline 72). E.g., 'cwd' is described as 'Caller cwd, used for smart-default session pick' but doesn't explain what happens if omitted or how it interacts with sessionId.