Loading ToolBench
Fusion 插件生态 — 基于 fusion-cowork 的标准化插件注册中心、生命周期管理、Claude 全链路适配
8 tools with basic schemas and descriptions, but significant gaps in LLM-optimization and error guidance. Tool names lack consistent verb-noun patterns (e.g., 'mlx_chat_plugin', 'heavy_compute', 'slow_worker' are vague). Descriptions are present but generic (avg ~60 chars, well below 194-char baseline). Parameters have types and descriptions, but lack constraints (enums, ranges, format specs). No documented output schemas. Error handling is absent, tools provide no recovery guidance. Composition is reasonable (single responsibility), but parameter naming could be more explicit (e.g., 'label' vs 'label_text'). No security annotations visible.
Async entry point demo: sleeps then returns a label.
Subagent-driven batch refactor of a target symbol.
Echoes input text uppercased.
Formats a title + body into a markdown report.
VRAM-consuming compute (auto acquire/release on enable/disable).
Runs in a PROCESS sandbox subprocess (crash-isolated).
Local MLX inference via desk.mlx_chat().
Tool names lack action verbs and are vague. 'mlx_chat_plugin', 'heavy_compute', 'slow_worker' do not follow verb_noun pattern (get_, create_, search_, etc.). LLMs cannot infer intent from names alone.
No output schemas documented. Tools return results but LLMs cannot plan downstream calls or extract fields. Violates pattern:tool and pattern:response-shaper.
Descriptions are generic and under 100 chars (avg ~60). Lack WHAT/WHEN/WHY context. E.g., 'Echoes input text uppercased' does not explain when to use it vs other tools or what the output format is.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 46 | <=2025-11-25 | v2 |
Long-running worker; demonstrates timeout meltdown + restart.
No error handling or recovery guidance. Tools provide no actionable error messages. If a tool fails, LLM has no guidance on retry, fallback, or next steps.
Parameters lack constraints. No enums for 'action' in batch_refactor (has enum but others lack them). No ranges for numeric params (matrix_size, work_seconds, max_tokens). No format specs for strings.