Loading ToolBench
Two tools with minimal documentation. Both have basic input schemas with type and description fields, but tool descriptions are extremely brief (under 50 chars). Parameter descriptions are present but generic. No output schemas documented. No error handling guidance. No tool annotations. Naming follows verb_noun pattern (search_context, enhance_prompt) but descriptions lack context on when/why to use each tool or what they return. Missing idempotency, pagination, and composition guidance.
Enhance a prompt using AugmentCode with conversation history
Semantic code search using AugmentCode
Tool descriptions are too brief (under 50 chars). LLMs cannot determine when to select search_context vs enhance_prompt or what each returns. Descriptions must explain WHAT, WHEN, and WHAT IT RETURNS (10-1024 chars recommended).
No output schemas documented. LLMs cannot plan downstream calls or extract results. Both tools must declare their return type (e.g., {results: [{file: string, score: number, snippet: string}], total: number}).
No error handling guidance. If search_context fails (invalid path, API timeout, no results), the tool provides no recovery hints. Errors must guide the LLM: 'Project path not found. Verify project_root_path exists and is readable.'
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 42 | <=2025-11-25 | v2 |
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Both tools are READ_ONLY but this is not declared in the schema. Agents cannot determine if tools are safe to retry or have side effects.
Parameter 'conversation_history' in enhance_prompt requires JSON stringified array but lacks format/pattern constraint. LLMs may pass invalid JSON. Add pattern or enum constraint, or accept structured array directly.