sentrux
3,344Real-time architectural sensor that scores codebase modularity, acyclicity, and redundancy as the agent writes.
Activity band details
Activity band A to F: a bin of a score that adds GitHub stars (log scale), a push to the upstream repository in the 12 months before `as_of`, the number of declared integration surfaces and, where the catalogue records them, citing papers and benchmark evaluations. It measures popularity and activity. It is not a level of evidence: no band means a paper cites the tool or that it was measured. Recency is judged as of 2026-11-22.
Computed band: D
(raw score 4.91).
Shown band: D
.
Facets:
documented_in=0evaluated_on=0integration_surfaces=1.5paper_citations=0recent_push=2stars=1.41
capped from C to D by maturity_status='experimental'
Source of truth: pipeline/maturity/score.py
Description
Pure-Rust CLI plus MCP server that watches a codebase's structural health in real time and emits a single 0–10000 quality signal computed from five root-cause metrics (modularity, acyclicity, depth, equality, redundancy). Ships an interactive treemap GUI, a CI-friendly rules engine (.sentrux/rules.toml declares max-cycles, layer order, forbidden boundaries, god-file thresholds) and a session-baseline gate (sentrux gate) that flags architectural regression between agent runs. MCP integration exposes nine tools (scan, health, session_start/end, rescan, check_rules, evolution, dsm, test_gaps), letting Claude Code, Cursor, Windsurf, and OpenCode read the sensor signal and self-correct without a human in the loop. 52 languages via tree-sitter plugins.
Why it's included
Occupies detection × post_generation × fully_autonomous for dependency_blindness: the redundancy metric flags duplicate function bodies (identical body hashes) inside the repo. Narrow coverage: it catches exact in-repo duplicates, not re-implementation of stdlib or library functionality. The session gate (sentrux gate) measures structural-quality regression between agent runs; the 2026-11-20 tag audit judged that this is not detection of cascading_failure (error recovery compounding errors), so that tag was removed (research/fm_tag_audit/final_table.csv). Distinct from semgrep (rule-or-pattern static analysis) by emitting a continuous live score plus session-baseline diff rather than discrete pass/fail findings. The CI-gate mode shifts the same binary toward prevention and graduated_hitl; if that use-case grows we may split it into a second Tool node.
Security notes
Pure Rust, single binary, runs fully local. Reads source files read-only; the MCP server exposes scan/session/rules tools but no filesystem write capability. Rules and constraints live as a human-authored .sentrux/rules.toml in the repo. No telemetry is documented in the README; verify before shipping into a regulated environment.
Addresses failure modes
These tags are the catalog's claim about what sentrux is built to catch or prevent. A tag is not a test result. SAICA-Bench has not measured sentrux, so none of these tags is tested here.
- Dependency blindness — Reimplementing functionality that already exists in reachable code.