Blog
Engineering
Engineering
A Tool That Did Nothing Came Out 43% Cheaper Than the Bare Agent
A tool the agent never called measured 43% cheaper than a bare agent. Prompt caching and run order explain it, and why we report output tokens, not dollars.
Read6 min read
Engineering
How to Benchmark a Codebase Tool So the Number Survives a Rerun
JetBrains reran two token-saving claims and both collapsed. These eight rules build a codebase tool benchmark that holds up when someone else reruns it.
Read10 min read
Engineering
We Benchmarked Ourselves Against Four Open Source Tools and Came Last
A sealed retrieval benchmark on 112 ContextBench instances: repowise started last, fixed a query-time bug, and scored higher on 42 instances it never tuned on.
Read8 min read
Engineering
Does Code Health Predict Bugs? 21 Repos, 9 Languages, ROC AUC 0.74
A reproducible defect-prediction study: 21 repos, 9 languages, 2,826 labeled files, ROC AUC 0.74, and 2.3x more defects caught under a fixed review budget.
Read10 min read
Engineering
Is AI-written code buggier than human code? We blamed 112,000 commits to find out
We git-blamed 112,382 commits across 28 repos to test if AI-agent code is buggier than human code. Controlled for size, it isn't, and its lines last longer.
Read9 min read
Engineering
Process Metrics Beat Structural Metrics for Predicting Defects
Across 21 repositories, co-change, change entropy and ownership predicted bug-fix files better than complexity or code smells, with file size controlled.
Read10 min read
Engineering
How a 23-Marker Code Health Scorer Works
How repowise calculates code health: 23 deterministic markers, category caps, a 1 to 10 file score, module rollups and trends, with the benchmark behind it.
Read9 min read
Engineering
Building a Deterministic PR Review Bot With Zero LLM Calls
How to build a deterministic PR review bot with no LLM calls: parse the diff, query a graph and git history, post one comment, and stay quiet when nothing fires.
Read9 min read
Engineering
How repowise Dead Code Detection Works: Graph Reachability and Git History
How repowise finds unused files, exports and packages: dependency-graph reachability, a git-age confidence ladder, and a single rule for safe_to_delete.
Read9 min read
Engineering
Architectural Decision Records: Capture Why Your Code Is the Way It Is
Write architectural decision records that stay true: a template, when to write one, how to recover the why from git history, and how to hand it to AI agents.
Read8 min read
Engineering
Building a Dependency Graph for Any Codebase
How a code dependency graph is built: parse imports into an AST, resolve them per language, then rank, cluster and cycle-check the result. Includes commands.
Read9 min read
Engineering
How We Built repowise: Codebase Intelligence Architecture
How repowise is built: one local index of graph, git history, docs, decisions and health, served to people and AI agents through ten MCP tools.
Read9 min read
Index your repo free.
See what repowise finds in your own code in minutes. Or explore a live repo first.