Skip to content

Security: Keyoku-ai/keyoku

Security

SECURITY.md

Security Policy

Reporting a vulnerability

Use GitHub's private vulnerability reporting on this repository, or email support@keyoku.ai. Please do not open public issues for security reports. We aim to acknowledge reports within 72 hours.

Supported versions

Only the latest published 0.x release receives security fixes.

Execution model — read this before installing

Keyoku is a local automation tool. Be aware of what it does by design:

  • Approved workflow templates execute shell commands on your machine with your privileges via workflow_execute. The approval step (workflow_approve) is the trust boundary: review every step of a template before approving it, exactly as you would review a shell script before running it. Steps time out (30s, SIGTERM→SIGKILL) and output is captured.
  • The activity log (~/.keyoku/activity.jsonl) records summaries of your tool usage (commands, file paths). It stays on your machine. There is no telemetry and no network calls unless you configure an SLM key (GEMINI_API_KEY / ANTHROPIC_API_KEY), in which case pattern summaries are sent to that provider for refinement.
  • State files under ~/.keyoku are written with mode 0600 (dir 0700), the same posture as ~/.aws. Connector configs may contain credentials — treat the directory accordingly.
  • Connector autonomy: external MCP connectors default to gated execution; write-capable calls can be routed through an approvals queue (keyoku approvals).

There aren't any published security advisories