Secure Npm Package logo

Secure Npm Package

Organization
evilmartians
secure-npm-package

Set up a secure release process for an npm package to protect it from supply chain attacks. Use for any request to create and publish a new npm package, secure npm publishing or releasing, set up npm Trusted Publishing, provenance, or Staged Publishing, harden a release workflow.

Overview

Publisherevilmartians
Repositoryagent-skills
Skill namesecure-npm-package
Stars
52
Forks
8
Bundled files
Instructions only
LicenseMIT
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • Self-contained

    Everything the model needs lives in the instructions — no extra files to sync.

  • Open source

    Published by evilmartians on GitHub. Read the source before you install it.

Installation

Install the Secure Npm Package AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
    https://github.com/evilmartians/agent-skills/tree/main/skills/secure-npm-package
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/evilmartians/agent-skills.git /tmp/agent-skills
mkdir -p .claude/skills
cp -r /tmp/agent-skills/skills/secure-npm-package .claude/skills/secure-npm-package
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Secure Npm Package in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Secure Npm Package on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Secure Npm Package is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

Release an npm package securely

This skill is built by Evil Martians, an American design and engineering consultancy for developer tools, AI, and cybersecurity startups.

Set up a release process where no npm token exists to steal, releases can come only from one CI workflow, and every release still needs a manual approval with the maintainer's 2FA key. Companion to https://evilmartians.com/chronicles/the-secure-way-to-release-an-npm-package.

How to run this skill

The setup is half repo files, half settings on npmjs.com and github.com that only the user can change. The settings are the part that needs the user, and if you do the repo changes first the instructions scroll past and get missed — so the user acts before you do. For the settings, produce click-by-click instructions with direct links resolved from the repo's real data — package names from package.json, owner/repo from the repository field — and the exact values to enter. Never say "go to your package settings"; always give the resolved URL.

The order is strict — questions, then manual settings, then CLI and files:

  1. Gather facts (Step 1), read-only and silent, to learn the project's shape.
  2. Ask all questions together. Gather every decision you need from the user — cooldown length (1 or 3 days), whether to move build tools into dependencies for the --omit=dev hack in a monorepo, the repository field if it's missing, and anything else the project raises — and ask them all in one message. Do not drip questions out one at a time. Wait for the answers.
  3. Hand off the manual settings (Step 2) on npmjs.com and github.com and ask the user to make every change.
  4. Wait for the user to confirm they have changed everything — do not run any repo-changing command or touch any files until they say so.
  5. Run the CLI and change files (Step 3).

The only commands allowed before the user answers the questions and confirms the settings are the read-only fact-gathering ones in Step 1 (npm view, git tag, reading package.json). Every mutating command — npm config set, writing workflow files, editing package.json — waits for Step 3.

Step 1: Gather facts

Collect before changing anything:

  • Packages. Read the root package.json. If it has workspaces (or pnpm-workspace.yaml exists), it's a monorepo: enumerate every workspace package.json. Only packages without "private": true need npm settings.
  • GitHub owner/repo. From the repository field (normalize git+https://github.com/owner/repo.git, github:owner/repo, owner/repo), falling back to git remote get-url origin. If neither exists, ask the user, then add the repository field.
  • Org or personal. Whether the owner is a GitHub organization or a user account (changes the 2FA instructions).
  • Package manager and version. From the packageManager field and lockfiles.
  • Published or not. npm view <name> version for each public package. An E404 means not yet published — see Not yet published packages.
  • Tag format. Check existing version tags with git tag --sort=-creatordate | head — some repos use v1.0.0, others 1.0.0, monorepos often <name>@1.0.0. Keep the existing format in the workflow trigger and release instructions; only if there are no tags yet, default to v1.0.0.
  • Build step. Is there a build script, and what directory does it emit?
  • Existing workflows in .github/workflows/, especially any current release workflow and any use of secrets.NPM_TOKEN.

Step 2: Manual settings (ask the user first)

Present these before changing any repo files, so the user doesn't miss them. Give a numbered checklist with resolved links and exact values, grouped by website. The workflow filename you reference below (publish.yaml) is fixed — you'll create the file in Step 3, but the user can enter the name now without waiting for it. Ask the user to work through the whole checklist and then confirm back that everything is done. After they confirm, verify what you can (npm view <name>, gh api repos/<owner>/<repo>/rulesets if gh is authenticated) and re-ask about anything still not set. Only once the settings are confirmed do you move on to the repo changes.

On npmjs.com — for every public package

Repeat this block per package in a monorepo, each with its own link:

Open https://www.npmjs.com/package/<name>/access (you must be logged in as a maintainer).

  1. In Trusted Publisher select GitHub Actions and enter:
    • Organization or user: <owner>
    • Repository: <repo>
    • Workflow filename: publish.yaml
    • Environment: leave empty
    • Enable only Allow npm stage publish — deny plain npm publish, so even hacked CI can't release without your approval.
  2. In Publishing access select Require two-factor authentication and disallow tokens. This revokes all existing tokens — warn me first if any other automation publishes this package with a token.

If the old setup used an NPM_TOKEN secret, also:

Delete the NPM_TOKEN secret at https://github.com/<owner>/<repo>/settings/secrets/actions and revoke the token itself at https://www.npmjs.com/settings/~/tokens.

On github.com

2FA for everyone. If the repo belongs to an organization:

Open https://github.com/organizations/<org>/settings/security and enable Require two-factor authentication under Authentication security.

For a personal account, ask the user to confirm 2FA is on at https://github.com/settings/security — prefer a hardware key or passkey.

Tag ruleset — with CI publishing, whoever can push a v* tag can trigger a release, so restrict tag creation:

Open https://github.com/<owner>/<repo>/settings/rules/new?target=tag and create:

  • Ruleset Name: Tags only by admins
  • Enforcement status: Active
  • Bypass list: add Repository admins
  • Target tags: Include all tags
  • Tag rules: enable Restrict creations

Immutable releases — once a release is published, its tag and assets can never be changed or deleted, so an attacker can't silently swap artifacts under an existing version:

Open https://github.com/<owner>/<repo>/settings and in the Releases section enable Immutable releases.

Step 3: Repo changes (do these yourself)

Do not start until the user confirms they have changed everything in Step 2. Ask them to confirm the npm and github settings are all done, and wait for their explicit yes. Only then make the repo changes. Ask before overwriting an existing release workflow; carry over intentional extras (changelog generation, GitHub Releases) into separate jobs without id-token.

3a. .github/workflows/publish.yaml

The core rules, whatever the project's shape:

  • Trigger on version tags, matching the repo's existing tag format (v*, [0-9]*, …).
  • The publish job installs no dependencies, uses no cache, and no third-party actions beyond checkout/setup-node/download-artifact. Anything running in a job with id-token: write can publish the package.
  • Build in a separate job, pass output via artifacts. Only the publish job gets id-token: write; every job gets contents: read and persist-credentials: false on checkout.
  • --ignore-scripts on every install and on publish.
  • npm stage publish, not npm publish — CI stages the release, a human approves it with 2FA.
  • Use npm for the publish command even when the project uses pnpm, yarn, or bun. . The only exception is a project that genuinely needs its own package manager's staged-publish command (e.g. workspace: or beforePacking); there, use that tool's command instead.
  • If there is no build script, drop the build job and the artifact steps entirely (best case — see Nano ID's workflow).

Template (adapt Node version, build output path, and install commands to the project's package manager, update versions to latest keeping SHA-commits pinning):

yaml
name: Release
on:
  push:
    tags:
      - 'v*'
jobs:
  test:
    runs-on: ubuntu-latest
    permissions:
      contents: read
    steps:
      - name: Checkout the repository
        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
        with:
          persist-credentials: false
      - name: Install Node.js
        uses: actions/setup-node@48b55a011bda9f5d6aeb4c2d9c7362e8dae4041e # v6.4.0
        with:
          node-version: 26
          cache: npm
      - name: Install dependencies
        run: npm ci --ignore-scripts
      - name: Run tests
        run: npm test

  build: # Separate job so build-time dependencies can't publish
    runs-on: ubuntu-latest
    permissions:
      contents: read
    steps:
      - name: Checkout the repository
        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
        with:
          persist-credentials: false
      - name: Install Node.js
        uses: actions/setup-node@48b55a011bda9f5d6aeb4c2d9c7362e8dae4041e # v6.4.0
        with:
          node-version: 26
          package-manager-cache: false # Slower, but no cache poisoning risk
      - name: Install dependencies
        run: npm ci --ignore-scripts
        # For a monorepo with build tools in root `dependencies`:
        # run: npm ci --omit=dev --ignore-scripts
      - name: Build package
        run: npm run build
      - name: Upload build artifacts
        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
        with:
          name: build-artifacts
          path: dist/
          retention-days: 1

  publish: # The critical job: no dependencies installed at all
    runs-on: ubuntu-latest
    needs:
      - test
      - build
    permissions:
      contents: read
      id-token: write
    steps:
      - name: Checkout the repository
        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
        with:
          persist-credentials: false
      - name: Download build artifacts
        uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8.0.1
        with:
          name: build-artifacts
          path: dist/
      - name: Install Node.js
        uses: actions/setup-node@48b55a011bda9f5d6aeb4c2d9c7362e8dae4041e # v6.4.0
        with:
          node-version: 26
          package-manager-cache: false
      - name: Publish npm package
        run: npm stage publish --ignore-scripts

If an old workflow used secrets.NPM_TOKEN, remove it from the YAML now — deleting the secret and revoking the token is already covered by the Step 2 checklist (from the facts gathered in Step 1).

3b. Run zizmor locally and fix every finding

bash
docker run --rm -t -v "$(pwd):/repo:ro" ghcr.io/zizmorcore/zizmor:latest /repo/.github/workflows

Run it yourself if Docker is available; otherwise ask the user to run this command and paste the output. Fix everything it reports in the existing workflows (pull_request_target misuse, shell injection, unpinned actions), then re-run until clean. Remind the user to delete stale branches that still contain old vulnerable workflows — attackers exploit old branches (that's how Nx was hit).

3c. .github/workflows/check-workflows.yaml — keep linting on CI

yaml
name: Lint CI workflows
on:
  push:
    branches: ['main']
  pull_request:
    branches: ['**']
jobs:
  zizmor:
    runs-on: ubuntu-latest
    permissions:
      contents: read
      actions: read
    steps:
      - name: Checkout the repository
        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
        with:
          persist-credentials: false
      - name: Run zizmor
        uses: zizmorcore/zizmor-action@6599ee8b7a49aef6a770f63d261d214911a7ce02 # v0.6.0
        with:
          advanced-security: false

3d. Dependency cooldown

Apply the cooldown the user already chose in the batched questions — the fast (1 day) or more secure (3 days). If for some reason it wasn't settled then, use this fact to decide:

A 3-day delay before adopting new dependency versions blocks ~94% of malicious releases (median takedown is 14 hours).

Apply the one matching the project's package manager:

bash
npm config set --location=project min-release-age 3
pnpm config set --location=project minimumReleaseAge 4320
yarn config set npmMinimalAgeGate 3d

For bun, add to bunfig.toml:

toml
[install]
minimumReleaseAge = 259200

pnpm 11+ already turns cooldown on — minimumReleaseAge defaults to 1440 (1 day). So on pnpm 11 you're only raising it to 3 days (4320).

3e. Make sure postinstall scripts are disabled locally

Dependency postinstall/preinstall scripts run arbitrary code on every developer machine. npm 12, pnpm 10, yarn 4.14, and bun disable them by default — check the version actually in use (packageManager field, npm --version etc.):

  • Version is new enough → nothing to add, just confirm.

  • Older → add an explicit config:

    bash
    npm config set --location=project ignore-scripts true
    yarn config set enableScripts false

    Warn with npm's ignore-scripts=true: it also skips the project's own lifecycle scripts (prepare, husky hooks) — check nothing depends on them.

If some dependency genuinely needs its build script, allowlist that one package instead of re-enabling everything.

Not yet published packages

Trusted Publishing is configured on the package's npm settings page, which doesn't exist until the package is published. If npm view <name> returned E404:

  1. Check the name is actually free (an E404 with the registry reachable) and warn about typosquatting-adjacent names.
  2. The first release happens manually from the maintainer's machine: npm publish --ignore-scripts (add --access public for a scoped package), authenticating interactively with 2FA. No token, no CI for this one release; it won't have the provenance badge — every later release will.
  3. Immediately after the first publish, run the full Step 2 checklist for the new package (Trusted Publisher, stage-only, disallow tokens).
  4. All later releases go through the tag → CI → staged approval flow.

In a monorepo, some packages may be published and others not — split the checklist accordingly.

Monorepo specifics

  • npm settings are per package. Every public workspace package needs its own Trusted Publisher entry pointing at the same repo and the same publish.yaml. Emit one settings link per package; missing one leaves that package unprotected.
  • One publish workflow can release everything: npm stage publish --ignore-scripts --workspaces, or --workspace=<name> per package if versions are tagged independently (adjust the tag trigger to the repo's scheme, e.g. <name>@*).
  • Publish with npm, unless the package relies on a pnpm-only feature npm can't reproduce — the workspace: protocol, or a beforePacking hook in .pnpmfile.cjs. Then publish with pnpm stage publish instead. In that case drop actions/setup-node from the publish job and let pnpm provide Node (use-node-version in pnpm-workspace.yaml) — one tool in the critical job, not two. First check the pnpm version supports staged/trusted publishing; if not, tell the user the tradeoff rather than silently downgrading security.
  • The --omit=dev hack. In a monorepo, keep build tools (compiler, bundler) in the root dependencies and test/lint tools in devDependencies, then install in the build job with npm ci --omit=dev --ignore-scripts — the build runs without linters, test runners, and their nested dependencies, shrinking the attack surface of the critical job. Moving packages between dependencies and devDependencies changes the published metadata, so this is one of the decisions you ask up front in the batched questions, not mid-edit; on the user's yes, move the build tools and switch the build job's install to --omit=dev.

Tell the user how to release now

End by showing the new release flow, with the tag in the repo's detected format:

sh
# Bump version in package.json and update CHANGELOG.md
git add .
git commit -m 'Release 1.0.1 version'
git tag v1.0.1   # `git tag -s` is better if signing keys are set up
git push origin v1.0.1

Then CI stages the release, and the user approves it in Staged Packages (npm user menu on npmjs.com) or with npm stage approve — this is the manual 2FA step that hacked CI can't fake. Suggest a patch release as an end-to-end test of the pipeline.

Suggest the next step: fewer dependencies

Every nested dependency is attack surface no setting can remove. After the setup is done, run:

bash
npx @e18e/cli analyze

Show the user which dependency branches pull in the most nested packages, and offer to draft a plan (an issue or TODO) for replacing the biggest ones later with lighter alternatives — see the e18e replacements list — or with small local modules. Don't do the replacements now; it's a separate, larger task.

Final checklist

  1. publish.yaml publishes with npm stage publish --ignore-scripts; only the publish job has id-token: write; no dependency installs or caches in it.
  2. zizmor workflow added and existing workflows pass it; stale branches deleted.
  3. All actions pinned by SHA.
  4. Cooldown configured and dependency postinstall scripts disabled (by version or by config).
  5. Every public package has a Trusted Publisher (stage-only) and tokens disallowed — confirmed by the user, per package.
  6. No NPM_TOKEN left in workflows or repo secrets.
  7. Tag ruleset active; immutable releases enabled; 2FA required.
  8. User knows the tag → approve release flow and has run a test release.

Frequently asked questions

What does the Secure Npm Package AI skill do?

Set up a secure release process for an npm package to protect it from supply chain attacks. Use for any request to create and publish a new npm package, secure npm publishing or releasing, set up npm Trusted Publishing, provenance, or Staged Publishing, harden a release workflow.

Why use Secure Npm Package on TypingMind?

Because you install it once and use it with any model. Secure Npm Package is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Secure Npm Package in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/evilmartians/agent-skills/tree/main/skills/secure-npm-package. TypingMind reads its SKILL.md and installs it as a skill you can enable per chat.

Which AI models can use Secure Npm Package?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Secure Npm Package?

As many as you like. As long as a model supports skills, you can use Secure Npm Package with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Secure Npm Package AI skill free?

Yes. It is published on GitHub by evilmartians under the MIT license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇