Loading ToolBench
Rebound is a resilience and chaos engineering platform with fault injection capabilities, combining a CLI, web backend, job execution, and network proxy components
Scoring was not performed
Tool name 'fault_index_source_code' lacks clear action verb at start. Should follow verb_noun pattern (e.g., 'index_source_code'). Current name is ambiguous about the action being performed.
Tool description is only 31 characters ('Index source code for analysis'). Too short to provide context for LLM selection. Should be 50-200 chars, explaining WHAT the tool does, WHEN to use it, and WHAT it returns.
No output schema documented. LLMs cannot know what fields to expect from the tool response, preventing downstream tool composition and forcing trial-and-error reasoning.
Parameter 'source_dir' description is vague ('Directory path containing source code to index'). Does not specify if path is absolute or relative, whether it must exist beforehand, or what error occurs if missing.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-21 | F | 31 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 43 | - | v1 |
Parameter 'lang' description is minimal ('Programming language of source code (e.g. 'python')'). Does not specify valid language values as an enum, leaving LLMs free to hallucinate unsupported languages. No constraint on case sensitivity or format.
No error handling guidance. Tool may fail for missing directories, unsupported languages, or permission errors, but provides no recovery instructions for the LLM.
Source code provided is primarily project configuration files (pyproject.toml, package.json, etc.) rather than actual MCP server implementation. Cannot fully assess tool registration patterns or MCP compliance from dependencies alone.