Loading ToolBench
Live codebase visualization and structural quality gate for AI-agent-written code
Sentrux provides 7 tools with complete input schemas and descriptions. Naming follows verb_noun convention (scan, health, session_start, etc.). Descriptions are substantive (100-200 chars), explaining WHAT and WHEN. However, parameter descriptions are sparse or missing for optional params (e.g., 'days' in git_stats lacks guidance on valid range; 'format' in dsm lacks enum values). Output schemas are not documented, LLMs cannot predict return structure. Error handling is minimal; no recovery guidance. Tool composition is sound (each does one thing), but missing pagination/limit guidance for potentially large results (dsm, test_gaps). No tool annotations (readOnlyHint/destructiveHint) despite clear risk levels declared in metadata.
Get the Design Structure Matrix: NxN dependency matrix showing file relationships, clusters, and inversions.
Git history analysis: code churn, hotspots (churn x complexity), bus factor, change coupling. Raw data — not a score. Requires git history.
Get quality signal (0-1) with root cause breakdown (modularity, acyclicity, depth, equality, redundancy). Quality signal = geometric mean — maximize this ONE number.
Scan a directory and compute all metrics. Must be called before other tools. Returns quality_signal.
Re-scan and compare current state against session baseline. Returns diff showing what degraded.
Save current health metrics as baseline for later comparison via 'gate' or 'session_end'.
Output schemas not documented. LLMs cannot predict return structure (fields, types, nesting) for any tool. Breaks downstream tool chaining and forces agents to guess field names.
Optional parameters lack descriptions and constraints. 'days' in git_stats has no range guidance (default 90, but min/max unknown). 'format' in dsm lacks enum values ('text' vs 'stats' mentioned but not formalized). 'limit' in test_gaps (default 20) has no bounds.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk levels in metadata. session_start is WRITE; others are READ_ONLY. Annotations enable LLM safety reasoning.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 63 | 2026-07-28+ | v2 |
Find high-risk source files with zero test coverage. Cross-references test file detection with import graph and complexity.
No error handling guidance. Tools return success or fail silently. No recovery hints (e.g., 'scan must be called before health' is stated in health description but not enforced or explained in error context).
Potential result explosion not addressed. dsm and test_gaps return potentially large matrices/lists without pagination. No mention of result limits, truncation, or next_cursor patterns.