Skip to content

Security: orin-dx/lumen

Security

SECURITY.md

Security Policy

Reporting a Vulnerability

Email security@orin-dx.com — don't open a public issue for anything that could be exploited before a fix ships.

Include:

  • Which crate is affected (lumen-model, lumen-session, lumen-analysis, lumen-pattern, lumen-store, lumen-cli)
  • The concrete failure scenario — what an attacker could do, and how
  • A sample transcript file or repro steps, if you have them

Expect an acknowledgment within 5 business days. We'll keep you posted as a fix moves through triage, and credit you in the release notes unless you'd rather stay anonymous.

Scope

Lumen parses AI coding agent transcript logs (Claude Code, Antigravity, Codex, OpenCode) and stores derived data locally. The relevant threat model:

  • Untrusted transcript parsing — lumen-session's SIMD streaming parser (simd-json + memmap2) reads transcript files that may come from an untrusted source (a shared repo, a downloaded session, a malicious agent run). A crafted transcript that crashes the parser, corrupts memory, or triggers unbounded resource use is a real finding here.
  • Local data at rest — lumen-store persists parsed session data (via rusqlite) on disk. If transcript content includes sensitive material (file paths, code snippets, credentials an agent handled), how that's stored and who can read it matters.
  • Supply chain — Cargo.toml's dependency tree, especially unsafe-adjacent crates like memmap2 and rusqlite's bundled SQLite.
  • CLI argument and env handling — lumen-cli's clap-based surface (--json output, file path arguments).

Narrower than it could be: the workspace enforces unsafe_code = "forbid" at the lint level (Cargo.toml's [workspace.lints.rust]), so memory-safety bugs from unsafe blocks are not in scope for the crates that ship with that lint — a report showing the forbid was bypassed or a memory-safety bug exists despite it is a legitimate, high-priority finding.

Out of scope: issues in Claude Code, Antigravity, Codex, or OpenCode themselves — report those to the respective platform, not here.

Supported Versions

Lumen is pre-1.0 (0.1.0). Security fixes land on main only — there is no version matrix to backport across yet.

There aren't any published security advisories