Skip to content

Security: Autovara/AgentSync

Security

SECURITY.md

Security Policy

Reporting a Vulnerability

If you believe you've found a security issue in AgentSync, please report it privately rather than opening a public GitHub issue.

We aim to:

  • Acknowledge your report within 48 hours
  • Provide a fix or mitigation timeline within 7 days
  • Resolve critical issues within 30 days

You will receive public credit (unless you request anonymity) in the security advisory and release notes.

Scope

In scope:

  • Code execution from .agentsync/ content
  • Path traversal / file system escape
  • Secret exposure via generated output
  • Supply-chain risks in dependencies
  • Crashes or data corruption from malformed input

Out of scope:

  • Vulnerabilities in third-party AI agents themselves
  • Issues requiring physical access to the user's machine
  • Social engineering / phishing

Our Security Commitments

AgentSync follows the Privacy & Security pillar of the project:

  • Zero network calls in MVP — verified by CI: the network-free job in .github/workflows/ci.yml re-runs the test suite and a real agentsync init / build / check round-trip inside a Linux network namespace (sudo unshare -n). Any outbound DNS or socket call from the CLI would fail there with EAI_AGAIN / ENETUNREACH and break the build.
  • Zero telemetry — no data collection, ever
  • Strict file system allowlist — refuses to read/write outside the project
  • Restricted templating — Nunjucks runs without a filesystem loader and without exposed globals, so templates cannot include other files or reach env/fs/process. We treat .agentsync/ content as trusted (it lives in your repo); a malicious template author can still reach JS internals via expressions like ({}).constructor, so do not paste untrusted templates into .agentsync/.
  • Built-in secret detection — blocks builds that contain API keys
  • Atomic file writes — temp file is fsynced before rename
  • Reproducible builds — same input always produces byte-identical output

Supply Chain

  • Dependencies are minimal and audited (pnpm audit in CI)
  • Lockfile committed
  • Releases signed via npm provenance (sigstore)
  • SBOM published per release

Disclosure Policy

We follow coordinated disclosure. We will not publicly disclose details of a vulnerability until a fix is available and users have had reasonable time to update.

Past Advisories

None yet.

There aren't any published security advisories