Skip to content

Security: andrepontesmelo/deep-horizon

Security

SECURITY.md

Security Policy

Scope

deep-horizon is a CLI + adapter set: it maintains a deterministic horizon store (.horizon/) and wires harness hooks that inject it at session start. It holds no credentials — API keys live in each harness's own profile/config, not here — and nothing leaves the machine: injection is local text composition through the horizon-inject bin, not a network call.

horizon install writes to global harness configs (~/.zcode/cli/config.json, ~/.hermes/config.yaml plus the plugins dir). Every write is parse → merge → validate → timestamped backup → atomic; a config that does not parse, or a shape install does not recognize, is refused loudly rather than blind-written.

Supported versions

Fixes target the main branch HEAD and the latest published npm version; release tags lag behind main, so name the commit or npm version in reports.

Reporting a vulnerability

Email the owner via the contact on the GitHub profile (andrepontesmelo) rather than opening a public issue. Include: affected version/commit, the harness and config involved (redact anything private), and expected vs actual behavior. You will get an acknowledgement within 7 days and a fix or a documented mitigation for anything confirmed.

What is NOT a vulnerability

  • A harness running the installed hooks and injecting the horizon into sessions — that is the product's stated purpose. Audit unfamiliar hook configs before installing them.
  • horizon install refusing to write an unrecognized config shape, or horizon doctor reporting FAIL lines — those are the designed safety surface, not a denial of service.
  • Hooks failing open (a missing bin or an error injects nothing) — by design: injection must never block a session, so the failure mode is silence.

There aren't any published security advisories