stable · v1.7.0 · 10 analyzers · 70 rules · 49 playbooks

We prepare your codebase so AI writes 90% of the code.

RepoPrime audits 10 categories, scores your repo 0–100 for Agent-Readiness, and ships a fix playbook per finding that your AI agent can execute directly.

Scan your repoSee how it works →$npm i -g @evgeniynezivoy/repoprime
zsh — ~/acme-api — repoprime scan .
~/acme-api on ⎇ main ❯
↑ runs locally · never uploads source · writes only under .repoprime/
02the problem

Your AI agent is only as good as the context you give it.

◆ ETH Zürich, 2025
# AGENTS.md
This is a TypeScript project.
Run tests with npm test.
Follow best practices.
↳ score: 14/100 · worse than empty

Generic AGENTS.md makes your agent WORSE

A study comparing Claude Code output with generic AGENTS.md ("this is a TypeScript project") vs. no AGENTS.md at all found the generic instructions materially degraded task success. Boilerplate context is not neutral — it crowds out the specific intent the agent actually needs.

◆ what we see
✗ wildcard CORS in 3 handlers
✗ 14 undocumented env vars
▲ 7 deps pinned to ^
▲ process.env accessed 41×
✗ 0 migrations committed

Bugs agents silently reproduce

Undocumented env vars. Wildcard deps. Missing migrations. Unvalidated process.env. Wildcard CORS. Your agent has no way to see these are mistakes — so when it writes the next handler, it copies the pattern, and the bug propagates.

◆ tooling gap
ESLintcode for humans
SonarQubecode for humans
CodeClimatecode for humans
Copilotreads · does not score
Cursorreads · does not score
RepoPrimecontext for agents

ESLint and Sonar score code for humans

They catch unused variables, cognitive complexity, test coverage. None of them grade the context files your AI agent reads before writing a line. Until now, repo quality for agents was invisible.

03how it works

Three steps. One command. Zero mutations to your source.

One command. Ten analyzers. Seventy rules.

Runs locally on your machine. Reads your repo, never writes to source files. Produces report.json and updates your AGENTS.md in-place if allowed.

.repoprime/
report.json12.4 kBJSON · machine-readable findings
AGENTS.md8.2 kBregenerated · 267 lines
fixes/49 filesone markdown playbook per finding
history.ndjson1.1 kBdelta series for sparklines
04the ten analyzers

Ten categories. Seventy rules. Forty-nine auto-fixable.

Each analyzer runs AST-level checks against your repo and writes structured findings to report.json. Hover any card to see example rules.

01
documentation
README, quickstart, architecture.md, changelog — agent-readable source of intent.
8 rules75% fixable
02
ai-context
AGENTS.md, CLAUDE.md, .cursorrules, .github/copilot-instructions.md — scored for specificity.
7 rules100% fixable
03
testing
Framework presence, coverage config, CI test job, meaningful test structure.
8 rules63% fixable
04
security
CORS, helmet, gitignore, docker non-root, pinned versions, committed lockfile.
9 rules89% fixable
05
configuration
Env validation, .env.example sync, ports/URLs extracted to config layer.
7 rules71% fixable
06
errorHandling
Structured logger, base error class, request-id, process handlers, non-zero exits.
8 rules88% fixable
07
cicd
CI workflow, PR triggers, lint/test/build steps, commitlint, husky, PR template.
8 rules100% fixable
08
structure
Module boundaries, circular imports, dead code, layer coherence.
6 rules50% fixable
09
database
Migration scaffold, schema versioning, seed scripts, connection pooling.
5 rules20% fixable
10
apiContracts
API version prefix, validation pipe, rate limiter, contract tests.
4 rules75% fixable
05playbook demo

Every finding ships a playbook your agent can run.

Click a finding on the left to see its rendered .repoprime/fixes/<fixerId>.md. The Prompt for your agent block is exactly what goes to Claude Code or Cursor.

~/acme-api ❯ repoprime scan . --fixable
◇ 49 fixable findings across 10 analyzers · showing top 3
→ wrote .repoprime/fixes/security.fix-cors.md
~/acme-api ❯
.repoprime/fixes/security.fix-cors.md

Fix: security.fix-cors

severity: high · location: src/main.ts:14 · auto-fixable
## Context

CORS is currently `origin: true` — any origin passes preflight. Exposes browser clients to CSRF on authenticated endpoints. ETH Zurich research shows generic fixes (`cors(...)`) degrade agent output vs. no config at all.

## Files affected
  • →src/main.ts
  • →.env.example
  • →README.md
## Prompt for your agent
Restrict Nest CORS to an allow-list read from CORS_ORIGINS env.
Remove the wildcard `origin: true` and reject non-listed origins.
Add CORS_ORIGINS=<your allowed production origin> to .env.example.
Verify by running `curl -I` against the /health endpoint with a disallowed Origin header
and confirm the Access-Control-Allow-Origin header is NOT echoed back.
## Verification
$ pnpm build
$ pnpm test src/main.spec.ts
$ curl -I -H "Origin: <disallowed>" /health
06the dashboard

Visualize every repo's readiness across your organisation.

  • ·repo table with 30-day sparklines
  • ·per-finding drilldown with diff context
  • ·one-click copy of any playbook prompt
repoprime.com/dashboard
repoprime1.6.0
dashboardreposfindingsplaybooks
prod · onlineA
Overview
Live readout across 5 repositories · scans streaming from repoprime scan --push · historic series retained 180 days.
last scan 12m ago24 scans /24h
aggregate score · 5 repos
68.6/100+4.2
+4.2 vs. last week
agent-ready
1stable
holding at top band
foundation
1needs fix
priority attention
scans · 24h
24
last 24 hours
Repositories
allreadystrongdevfound
repositoryscorematuritytrend · 30dlast scansha
stripe/stripe-node
@stripe · node.js
94
Agent-Ready12m agoa8f3c21
vercel/next.js
@vercel · next.js
78
Strong1h agoe22d5b0
prisma/prisma
@prisma · typescript · monorepo
72
Strong3h ago7b19af4
anthropics/anthropic-sdk-python
@anthropics · python
58
Developing5h agoc9d0312
your-org/your-repo
@you · scanned yesterday
41
Foundationyesterday11f6ae8
07your agent stack, unchanged

RepoPrime writes the context. Your agent does the fix.

RepoPrime lives beside your existing agent setup — it does not replace it, and it never calls an LLM itself.

Claude Code
reads playbooks verbatim
Cursor
composer · paste prompt
GitHub Copilot
via .github/copilot-instructions.md
Paperclip
agent plugin
GitHub Actions
ship repoprime scan --ci
08get started

Scan your first repo in under a minute.

No account. No upload. Runs locally, writes only under .repoprime/.

Start scanningRead the docs· MIT · node ≥ 20