Skip to content

Security: mnemopay/mnemopay-sdk

Security

SECURITY.md

Security Policy

Supported versions

Version Supported
1.12.x Yes — current stable
1.11.x Yes — critical fixes
1.10.x Critical fixes only
< 1.10 No

Preview/experimental modules (x402/AP2/StripeMPP rails, Swarm) may change in minor releases — pin exact versions in production. See VERSIONING.md and README § Module stability.

Reporting a vulnerability

Do not open a public GitHub issue.

Email: info@getbizsuite.com with subject line [security] mnemopay-sdk: <short description>.

Please include:

  • Affected version(s) and the import path / subpath where the issue surfaces (e.g. @mnemopay/sdk/governance).
  • Reproduction steps or minimum failing example.
  • Whether the issue has been disclosed elsewhere.
  • Whether you would like to be credited in the fix release notes.

Response timeline

  • Acknowledgement within 48 hours.
  • Initial triage within 5 business days.
  • Coordinated disclosure: critical issues get a patched release within 7 days of triage; high within 14 days; medium within 30 days.

Out of scope

  • Issues that require a malicious agent runtime (e.g. "if the runtime lies about charter scope, the SDK accepts it"). The trust boundary is the runtime; we publish the audit chain so misbehavior is detectable, not preventable from inside the runtime.
  • Theoretical timing attacks on local-only code paths with no network surface.
  • Anything in examples/ or playground/ that is not imported by published code.

Secure deployment

When self-hosting agents with MnemoPay governance:

  • Compile policies once; enforce ctx.act() / evaluateAction on every side-effect
  • Swap InMemoryApprovalStore for a durable store in production
  • Never commit API keys, rail secrets, or .env files — use a secrets manager
  • Restrict agent spend with hard_cap_usd and approval_threshold_usd
  • Export and retain Merkle audit bundles for regulated workloads — see docs/AUDIT-BUNDLES.md

Hall of fame

Researchers who responsibly disclose accepted vulnerabilities will be credited here unless they prefer anonymity.

(none yet)

There aren't any published security advisories