Skip to content

Security: coldlogicAI/3notch

Security

SECURITY.md

Security Policy

Supported Versions

3Notch is pre-1.0. Security fixes target the main branch. Once a 1.0 release cuts, this section will pin a support window.

Reporting A Vulnerability

If you believe you've found a security issue:

  1. Do not open a public issue or PR that describes the vulnerability.
  2. Open a private security advisory via GitHub's "Report a vulnerability" flow on the repository, or contact the maintainer privately before public disclosure.
  3. Include a reproduction, the observed impact, and (if possible) a suggested fix or mitigation.

Please give the maintainer a reasonable window to respond before disclosing publicly. Vague reports without reproductions ("the scanner missed a pattern that might be a secret") and unverified scanner output are not actionable — please verify before reporting.

Out Of Scope

  • Local filesystem privilege escalation by an already-authenticated user.
  • The secret scanner missing benign content. The scanner is a guardrail, not a proof.
  • Bundle authorship spoofing. 3Notch does not sign bundles; SHA-256 hashing only proves bytes did not change in transit.

For the full operational security story — what 3Notch scans, audits, and refuses to overwrite — see docs/reference/security-story.md.

Security Invariants 3Notch Maintains

  • Records under .notch/ are stored locally. No data leaves the machine without an explicit notch packet pack + user-initiated transfer.
  • No telemetry, analytics, cloud sync, or hosted dependencies.
  • No arbitrary shell execution through MCP.
  • Path-safety checks reject absolute paths, parent traversal, symlinks inside .notch/, and source links resolving outside the project root.
  • Imported packet folders are immutable: packet.md, manifest.json, and any file under artifacts/ must continue to match the recorded SHA-256 values.
  • Private records under .notch/private/ are hidden from MCP unless the server starts with --include-private.

A break of any invariant above is in scope for a security report.

There aren't any published security advisories