More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Lockfile Diff Review — What Your npm Update Added

    1

    Compares two package-lock.json files and shows the new direct dependencies, install scripts, non-registry sources, integrity changes and name lookalikes.

    Free

    Security scanned
    Lockfile Diff Review — What Your npm Update Added

    Lockfile Diff Review — What Your npm Update Added

    Example session with this skill installed

    git show HEAD~1:package-lock.json > before.json
    cp package-lock.json after.json
    python3 scripts/lockfile_diff.py before.json after.json --fail-on high

    {"packages": {
      "": {"dependencies": {"express": "^5.0.0", "loadsh": "^4.17.21"}},
      "node_modules/loadsh": {"version": "4.17.21", "resolved": "https://registry.npmjs.org/loadsh/-/loadsh-4.17.21.tgz"}
    }}
    
    • Read your context and instructions
    • Compiled the lockfile diff review

    lockfile-diff-review 2 → 49 packages
    added 48 removed 1 version changes 1
    direct adds 3 install scripts 2
    score 37/100 grade F findings 7
    high 4 medium 3 low 0

    NEW DIRECT DEPENDENCIES

    • @acme/logger
    • chalk
    • loadsh

    HIGH L2 @acme/logger
    New package resolves outside the npm registry: git+https://github.com/acme/logger.git#abc123
    HIGH L4 loadsh
    Name is one character from 'lodash'.
    HIGH L1 chalk
    New package runs an install script (preinstall/install/postinstall). Direct dependency.
    MEDIUM L5 express
    Direct dependency major bump 4.18.2 → 5.0.0.

    Scope: npm lockfiles. A clean result is not a malware verdict.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    A dependency update can add fifty packages in a screen of lockfile JSON. The direct dependency is easy to see; the transitive tree, the install scripts and the source change under a pinned version are not. This compares the before and after trees and puts the reviewable changes in one place.

    It reports:

    • New direct and transitive packages, with the direct ones named separately
    • New packages that declare a preinstall, install or postinstall script
    • New packages resolved outside the npm registry, such as a git or file URL
    • New packages with no integrity hash
    • Package names one edit away from a popular package, including letter swaps
    • Direct dependency major bumps and downgrades
    • Source-host changes and same-version integrity changes on existing packages
    • Blast radius: 25+ new packages is a prompt, 60+ is a finding

    The shipped update goes from 2 to 49 packages. It adds 48, removes one, and names three direct additions. It catches loadsh one swap from lodash, a git-hosted @acme/logger, install scripts in chalk and a transitive package, and an express 4→5 major bump. Seven findings, four high, score 37/100, exit 1. The clean control is a single patch bump and scores 100.

    Good to know: this is static comparison, not a malware or vulnerability scan. It does not query OSV, GitHub Advisory or registry reputation. An integrity hash proves the bytes match the lockfile; it does not prove the package is trustworthy. Typosquat detection is one edit against a small popular-name list, so a legitimate package can be flagged and a two-edit lookalike can pass. Install-script findings mean the package declares a script; npm 12 blocks dependency scripts by default, but older npm and other managers may not. It reads npm lockfiles only and never installs or executes a package.

    Only supports npm lockfiles (v1, v2, v3). Static analysis only; does not query registry reputation or vulnerability databases. Typosquatting check limited to one-edit distance.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 1 day ago

    • Free with an Agensi account

    Requires

    Git

    Permissions

    Terminal / Shell
    Network Access
    Read Files
    Write Files
    Environment Variables
    Listed1 day ago

    What's inside

    Frequently Asked Questions

    This skill performs a deterministic static analysis of your npm lockfiles to identify security risks like malicious install scripts, typosquatting, and integrity hash mismatches that are easily missed during manual PR reviews.

    It supports npm lockfile versions 1, 2, and 3, and is designed to run in environments with Python 3.10 or higher. Note that it is currently specific to npm and does not support pnpm or Yarn.

    No, the skill operates entirely offline and does not perform `npm install` or execute any code from the lockfile, ensuring a secure analysis environment for sensitive projects.

    Yes, the skill provides specific exit codes and can output results in JSON or Markdown, making it easy to integrate into CI/CD pipelines as a quality gate to block high-risk PRs.

    Unlike generic LLM prompts, this skill uses a specialized comparison engine to ensure 100% accuracy in parsing deeply nested JSON structures without the risk of hallucinated security threats.

    The skill focuses on structural and integrity risks (like source changes or install scripts) rather than known CVEs; it is intended to be used alongside vulnerability scanners like `npm audit` for comprehensive coverage.

    New in APIs & Backend