Loading ToolBench
Kiokuko has 9 tools with basic schemas and descriptions, but significant gaps limit production readiness. Tool names lack clear action verbs (e.g., 'capture_interaction_memory' vs 'store_interaction', 'recall_interaction_memory' vs 'search_interactions'). Descriptions are present but generic (avg ~60 chars, below 194-char baseline). Parameter descriptions exist but lack constraints, ranges, or format guidance. No output schemas documented. Error handling is minimal, no recovery guidance or actionable error messages. Security concerns: tools accept generic 'object' types for interaction/evidence/updates without validation. No tool annotations (readOnlyHint/destructiveHint) despite clear WRITE/READ_ONLY risk classifications.
Answer questions for agent task preparation
Capture interaction memory from agent interactions
Checkpoint scoped memory for task execution
Inspect and validate task assurance state
Prepare an agent task with profile and routing information
Recall interaction memory from stored interactions
Record execution evidence for task assurance
Refresh and update task memory state
No output schemas documented. LLMs cannot infer what fields to expect from tool responses, forcing unstructured parsing and downstream lookup calls.
Generic 'object' type parameters (interaction, evidence, updates, memory, profile) lack validation constraints. LLMs cannot determine valid structure; no schema validation prevents malformed input.
Tool names lack clear action verbs. 'capture_interaction_memory' and 'recall_interaction_memory' are vague; 'store_interaction' and 'search_interactions' would be clearer. LLMs struggle to disambiguate similar names.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 40 | 2025-06-18+ | v2 |
Review and validate task memory state
No tool annotations (readOnlyHint/destructiveHint/idempotentHint) despite explicit WRITE/READ_ONLY risk classifications in metadata. Agents cannot determine safe retry behavior.
Parameter descriptions lack format guidance, ranges, or constraints. E.g., 'limit' has no min/max; 'workspace' has no format; 'revision' has no pattern. LLMs cannot validate input before calling.