Skip to content

Security: zafrirron/Agent-Platform

Security

SECURITY.md

Security Declaration

What this platform is — and what it is not

Agent Platform Bootstrap installs markdown files that instruct AI agents how to behave during development. It is not an application, not a runtime library, and not a dependency. It has no code that runs inside your project.


What the platform does

Action What happens
Install Writes .agent/, .claude/, .cursor/, .agents/, .codex/ — plain text files only
Upgrade Updates the <!-- PLATFORM:START --> section of expert files — your content is never changed
Session start An AI agent reads markdown files and follows the instructions in them
Update check One read-only HTTPS call to api.github.com/repos/zafrirron/Agent-Platform/releases/latest
Uninstall Deletes its own files and restores your original AI configs from backup

What the platform never does

Concern Reality
Injects executable code into your project ❌ Never — only markdown, YAML, and JSON files are installed
Modifies your source code, tests, or configuration ❌ Never — installer creates new files only, existing files are skipped
Makes network calls from inside your project code ❌ Never — no runtime dependencies are added
Collects telemetry or sends data anywhere ❌ Never — no analytics, no tracking, no callbacks
Commits anything to your repository ❌ Never — all platform files are gitignored on install
Runs without your knowledge ❌ Never — every action requires you to explicitly invoke it
Overrides your security decisions ❌ Never — your PROJECT sections are never overwritten

How to verify everything yourself

The installer is open source. Every line of apply.js is visible at: https://github.com/zafrirron/Agent-Platform/blob/main/AGENT-PLATFORM-TEMPLATES/.agent/bootstrap/apply.js

Every expert rule is a readable markdown file. After install, open any file in .agent/agents/ — you can read every rule the platform gives your AI. Nothing is hidden or obfuscated.

Every rule traces to its source. MAINTAINER/platform-improvements.md records which real-world failure each rule was added to prevent.

The unit tests are public. tests/apply-utils.test.mjs — 40 tests verifying the installer behaves exactly as documented.

The gitignore block is visible. After install, open .gitignore — you can see exactly which files the platform added and remove any entry you disagree with.


The expert rules model

Expert rules and playbooks are instructions to an AI agent. They tell the agent things like:

  • "Use parameterised queries — no string-concatenated SQL"
  • "Every API endpoint must check auth"
  • "Grep for secrets before committing"
  • "Run the test suite before marking done"

These are quality and security improvements. They make your agents write more secure, better-tested code. None instruct agents to exfiltrate data, add backdoors, weaken security, bypass tests, or violate any principle.

You can read, edit, or delete any rule at any time. The <!-- PROJECT:START --> section of every expert file is yours — the platform never touches it. You can add your own rules, remove platform rules you disagree with, or delete expert files entirely.


Supply chain

  • No npm registry — installed directly from GitHub, not from the npm package registry
  • Version pinnablenpx github:zafrirron/Agent-Platform#v2.17.0 installs exactly that commit
  • No dependenciesapply.js uses only Node.js built-in modules (fs, path) — no third-party packages that could be compromised
  • Auditable — the full source is public and the tag history is immutable

What your AI agents do with the rules

The platform tells your AI agents to follow security best practices. The agents then write code. The code they write is yours to review before committing. The platform does not auto-commit, does not bypass code review, and does not remove your ability to inspect and reject changes.

The platform makes your AI agents more secure — it does not make your project less secure.


Reporting a security issue

If you find a security concern in the platform rules, the installer, or any platform file:

  1. Open an issue at https://github.com/zafrirron/Agent-Platform/issues
  2. Or email the repository owner directly

Security improvements found through responsible disclosure will be credited in MAINTAINER/platform-improvements.md and shipped to all users via the next upgrade.

There aren't any published security advisories