Loading ToolBench
swe-grep provides 4 read-only code search tools with basic schemas and descriptions. All tools have input schemas with typed parameters and descriptions, meeting minimum requirements. However, descriptions are terse (10-60 chars), lacking context on WHEN to use each tool or HOW they differ. No output schemas documented. Parameter descriptions are minimal, e.g., 'Regex pattern to search for' lacks format guidance or examples. No error handling guidance. Tools follow verb_noun naming (readfile, tree, glob, rg) but lack the depth expected of production tools. Naming is clear but descriptions fall short of the 50-200 char LLM-optimized baseline (94% of A+ tools have rich docstrings; these average ~30 chars).
Find files matching glob patterns with optional type filtering
Read file contents with optional line range limiting
Search code using ripgrep with pattern matching and file inclusion filters
Display directory tree structure with configurable depth and filtering
Descriptions are too terse (10-60 chars vs. 50-200 char baseline). LLMs cannot determine WHEN to use readfile vs. tree vs. glob vs. rg without richer context.
No output schemas documented. LLMs cannot plan downstream operations or extract structured data from results.
Parameter descriptions lack format guidance, constraints, or examples. E.g., 'Regex pattern to search for' does not explain regex dialect, escaping, or performance limits.
No error handling guidance. If a file is not found, regex is invalid, or ripgrep is unavailable, LLMs receive raw errors with no recovery path.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 48 | 2026-07-28+ | v2 |
Tool composition unclear. LLMs cannot determine whether to use glob + readfile or rg for a given task without richer descriptions explaining use cases.