"outputs": ["dist/**", ".env.production"] copies your production secrets into every cache artifact, and a deploy task without "cache": false can be replayed from cache instead of run: Turborepo prints the old logs and reports success while nothing was deployed. This skill finds those and the other monorepo cache mistakes below before your next build, and tells you how to fix each one.

Your agent runs a scanner on turbo.json and turbo.jsonc, nx.json, project.json, package.json files that use turbo, nx or lerna (scripts, dependencies and nx targets), CI workflows and Dockerfiles that run turbo or nx, and the JavaScript and TypeScript code of a Turborepo workspace (build output skipped). It reads JSON with comments, knows the current defaults (Turborepo caches every task and runs in strict env mode; Nx caches only targets marked "cache": true), merges package turbo.json files with the root one above them, knows the variables Turborepo passes to every task and the framework prefixes it infers, tells read-only tasks such as lint:migrations or release:notes from real deploys, and treats CI secret references and commented placeholders as what they are. It never runs turbo, nx or your code. It returns BLOCK, REVIEW, or NOT FLAGGED, with the file, line, reason, and a safer approach. Secrets are never printed.

What it checks

BLOCK (fix before shipping)

• M01: .env files and other secret files (.pem, .key, .npmrc, SSH keys) in task outputs, where every cache copy and remote cache upload then holds them
• M02: deploy, publish, release and migrate tasks that can be served from cache (no "cache": false in Turborepo, "cache": true or cacheableOperations in Nx); read-only tasks such as lint:migrations, migrations:check or release:notes are not reported
• M03: TURBO_TOKEN, Nx Cloud tokens, the remote cache signature key or the Nx S3 cache secret key written in active lines of CI files, Dockerfiles, package.json scripts or nx.json

REVIEW (check before release)

• M11: build tasks in the root turbo.json without outputs
• M12: environment variables the code reads (also inside template literals) that no turbo.json declares, one finding per variable
• M13: loose environment mode
• M14: a self-hosted Turborepo remote cache (remoteCache.apiUrl, TURBO_API) without signature verification; the Vercel Remote Cache set up as the Turborepo docs show is not reported
• M15: the Turborepo 1.x pipeline key
• M16: task inputs without $TURBO_DEFAULT$
• M17: an Nx Cloud access token in nx.json
• M18: the deprecated Nx S3, GCS, Azure and shared-fs caches (CVE-2025-36852)
• M19: a real-looking cache token in a commented-out CI or Dockerfile line (placeholders such as your-token are fine)

What's included

• The scanner (Python 3.9+, standard library only)
• A rules guide with the reason, safer approach and official source for each rule
• Risky and safe example repositories (a Turborepo 2 shop, a Turborepo 1.x app and an Nx platform), and the real report the scanner produces on the risky ones

Limits

It reads each file as written: tasks that Nx plugins infer, settings kept in the Nx Cloud or Vercel dashboards (token scopes, workspace access levels) and the remote cache Vercel turns on for its own builds are not seen. Environment variables are matched across all turbo.json files of the scan, not per task, env access through helper functions is not followed, build output in folders that no outputs glob, main field or known name points to is read like source, and a .env file caught only by a broad output glob such as ** is not reported. NOT FLAGGED is not a guarantee.

Contents

One ZIP file with 29 files: SKILL.md, the scanner in scripts/ (6 Python files), references/rules.md (every rule with its reason and safer approach), and examples/ (20 example files plus report.md, the real scanner output on the risky example).

How to use: This is an agent skill: a folder with a SKILL.md file and Python scripts that an AI coding agent such as Claude Code, Cursor, or Codex CLI can run. Unzip it into your agent's skills folder (for Claude Code: ~/.claude/skills/), then ask your agent to check your project. The scripts also run on their own with Python 3.9 or newer, with nothing to install. They read your files only: no network access, and nothing is sent anywhere.