How We Use Pullminder
on Pullminder
We built a verification layer for AI-assisted code. Then we became our own first customer.
The Challenge
AI writes fast. Review doesn't scale.
Pullminder's own codebase spans a Go API backend, a React dashboard, and an Astro marketing site. A significant portion of the code is written with AI coding tools — Cursor, Claude Code, and other assistants that let a small team ship at an outsized pace.
The problem: as a two-person team, we don't have the reviewer capacity to manually audit every AI-generated pull request. AI tools produce working code quickly, but they also introduce patterns that are easy to miss — leaked secrets in test fixtures, silently dropped error handling, test coverage regressions, and PRs that balloon to 800+ lines because the model "helpfully" refactored adjacent code.
We needed automated risk scoring and policy enforcement on our own repos — the same tool we were building for others.
The Solution
What we configured
25 rule packs
Running on every PR across all repositories. Each analyzer scores independently, contributing to a composite risk score.
Secrets detection policy
Block on any finding. No exceptions, no overrides. If a secret appears in a diff, the merge is blocked.
Test coverage policy
Warn if coverage drops relative to the base branch. Keeps AI-generated code honest about edge cases.
Large diff policy
Warn if a PR exceeds 500 lines changed. AI tools tend to over-generate — this flags scope creep early.
Slack alerts
High-risk PRs trigger an immediate Slack notification so neither of us misses something critical.
AI reviewer briefs
PRs scored above 50/100 get an AI-generated review brief summarizing what changed and why it scored high.
Results
What we measured
Every PR analyzed
Full coverage across all repos
Average analysis time
From PR open to risk score posted
Blocked merges
Secrets caught before production
Person team
Shipping with 10x confidence
Self-reported metrics from internal usage. Updated April 2026.
We built Pullminder because we needed it ourselves. Running it on our own repos isn't a marketing exercise — it's how we catch mistakes in code we wrote with AI assistance.
See it in action on your repos
Connect your GitHub repos and get automated risk scoring on every pull request. Setup takes under two minutes.
Start Free Trial