Build a Pull-Request Scanner for AI-Generated Code Risk

1. 1

   Position against the named threat

   Slopsquatting gives this product what most dev tools never get: a nameable, documented attack to defend against. The USENIX Security 2025 package hallucination study (about one in five package recommendations nonexistent across tested models, with much higher rates on some open models) is your anchor citation; use it accurately and link the research. This card is the product neighbor of the software supply chain security consulting service already in this library: they audit and advise, your check runs on every PR forever.

2. 2

   Ship the dependency reality check first

   Version one does one thing convincingly: for every dependency added in a PR, verify the package exists in the official registry, check its age, download history, maintainer signals, and name-similarity to popular packages (hallucinations cluster around plausible mashup names), and block or warn per the team's policy. This alone catches the scariest class of AI-introduced risk and takes minutes to demo.

3. 3

   Add the AI-pattern detectors that reviews miss

   Layer two flags what tired reviewers skim past: new network calls or filesystem access in code whose ticket was cosmetic, broadened permission scopes and IAM changes, disabled certificate checks or auth bypasses 'for testing,' secrets patterns, and diffs that touch far more surface than the stated intent. Score, do not sermonize: a ranked risk summary on the PR beats forty inline nags, and false-positive discipline decides whether teams keep you installed.

4. 4

   Be a required check, not a dashboard

   Distribution is through the code hosting platforms' marketplaces and CI ecosystems, installed per repo and made a required status check. The buying motion is bottom-up: one engineer installs it free on a side project, then brings it to work. Keep a genuinely useful free tier for public repos and small teams; the paid line sits at private repos, org policies, and audit reporting.

5. 5

   Price per repo or per developer, modestly

   This wants to be an easy yes on a team card: per-active-repo or per-seat pricing in single-digit dollars monthly, with an org tier that adds policy control and reporting. The comparison in the buyer's head is the cost of one supply chain incident and one security review cycle; do not price like an enterprise security suite before you have the enterprise features.

6. 6

   Let the research community be your marketing

   Publish what your scanner sees in the wild (anonymized, aggregate): hallucination-pattern frequencies, near-miss package names, permission-creep rates in AI-heavy repos. Security researchers and dev-tool newsletters amplify original data relentlessly, and every write-up plants your install link next to the scariest graphs. Momentum in this niche is real right now; the window rewards shipping over polishing.

Do not take our word for the money. Put your own numbers in and see what this idea has to earn before it works. Free, instant, no account needed.