Network-AI
Repository
Project standards

Contributing

Contribution workflow, tests, security expectations, and review criteria.

Back to documentation hub Open source file on GitHub
Source file: CONTRIBUTING.md

Contributing to Network-AI

Network-AI is a solo-maintained project with high quality standards. Contributions are welcome but must meet strict requirements.

Before You Start

All contributions require prior discussion. Do not open a pull request without an approved issue first.

  1. Open an issue describing the problem or feature
  2. Wait for maintainer approval before writing code
  3. Only after approval, fork and implement

Unsolicited PRs without a linked, approved issue will be closed.

Requirements for All Contributions

Code Quality

  • All 4,588 existing tests must pass (npm run test:all)
  • Zero TypeScript compile errors (npm run typecheck)
  • New features must include tests with >90% branch coverage
  • Follow existing code style and patterns
  • No new runtime dependencies without prior approval

Security

  • No hardcoded secrets, keys, or credentials
  • No new network calls without explicit justification
  • Input validation required on all public API entry points
  • Path traversal and injection protections where applicable
  • New network listeners bind loopback by default and fail closed without a secret
  • Changes to lib/asi/ or lib/perpetual/ must keep the control invariants intact — configuration can only be tightened at runtime, the external stop works without agent cooperation, agents never modify their own limits, and spending stays inside operator-signed ceilings — and must add regression tests for each invariant they touch

Documentation

  • JSDoc on all exported functions and classes
  • Update README.md if adding user-facing features
  • Update CHANGELOG.md under [Unreleased]

Pull Request Process

  1. Fork the repository and create a branch from main
  2. Implement your change with tests
  3. Run the full test suite:
   npm run test:all   # 4,588 tests across 47 suites
   npm run typecheck
  1. Open a PR referencing the approved issue
  2. Fill out the PR template completely
  3. Wait for review -- the maintainer reviews all PRs personally

PR Review Criteria

  • Does it solve the approved issue?
  • Are tests comprehensive?
  • Is the code clean and idiomatic TypeScript?
  • Does it maintain backward compatibility?
  • Does it introduce any security concerns?

What We Accept

  • Bug fixes with reproduction steps and tests
  • Security improvements
  • Performance optimizations with benchmarks
  • New adapter implementations (following BaseAdapter pattern)
  • Documentation improvements

What We Do Not Accept

  • Breaking API changes without a migration path
  • Features that add external runtime dependencies
  • Options that let an agent loosen its own limits, renew its own lease, or bypass a guarded path, kill switch, or budget ceiling
  • Code that reduces test coverage
  • Cosmetic-only changes (formatting, renaming)
  • AI-generated code without human review and testing

Development Setup

git clone https://github.com/Jovancoding/Network-AI.git
cd Network-AI
npm install
npm run test:all  # Run all 4,588 tests (47 suites)
npm run test:phase4  # Phase 4 behavioral control plane tests only
npx ts-node test-perpetual.ts         # Perpetual Runtime: leases, egress, CLI
npx ts-node test-perpetual-safety.ts  # Perpetual Runtime: guarded paths, sandbox, sentries, pipeline, budget
npm run typecheck  # Type-check

License

By contributing, you agree that your contributions will be licensed under the MIT License.