Loading ToolBench
Persistent, structured memory for AI coding agents. Stores and retrieves decisions, facts, goals, tasks, error resolutions, and manages memory branches.
Server has 13 well-named tools with consistent verb_noun patterns (memory_get_*, memory_create_*, memory_list_*, etc.). All tools have descriptions (10 - 200 chars, within baseline range). Input schemas are present with typed properties and descriptions for most parameters. However, output schemas are NOT documented, critical gap per pattern:tool. Parameter descriptions lack constraint details (enums, ranges, formats). No error handling guidance visible. Missing tool annotations (readOnlyHint, destructiveHint). Risk classification present in metadata but not in tool definitions.
Compare two memory branches. Shows entities only in A, only in B, and conflicts.
Record an architectural or technical decision.
Store a learned fact or piece of knowledge.
Create a high-level goal or objective.
Record how an error was resolved. Prevents solving the same error twice.
Create a concrete task or action item.
Get recent agent activity log.
Output schemas not documented. LLMs cannot plan downstream calls or extract required fields without knowing response structure.
Parameter descriptions lack constraint details. 'mode' accepts 'compact, full, survival, delta' but no enum constraint. 'priority' and 'confidence' accept specific values but not formalized. LLMs guess at valid options.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 54 | 2026-07-28+ | v2 |
Get assembled memory context for the current project. Returns structured knowledge: decisions, facts, goals, tasks, resolutions. Use this at session start to load project context.
Get memory health — entity counts, stale items, conflicts.
Get chronological timeline of memory events.
List all memory branches with entity counts and current branch.
Merge a source branch into target. Moves all entities from source to target.
Switch to a different memory branch. New entities will be stored on this branch.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Risk metadata exists but not exposed in tool definitions. Agents cannot distinguish safe reads from destructive writes.
No error handling guidance. Tools delegate to REST API but no recovery hints documented. If API returns 404 or validation error, LLM has no guidance on next steps.
Parameter 'goal_id' in memory_create_task accepts integer but no description of what happens when goal_id=0 or invalid ID. Undocumented dependency on goal existence.