Enji Guard
Browse documentation

Automation

Automation overview

Understand which Guard work can run manually, on a schedule, after commits, or from provider events.

Documentation · Updated July 22, 2026

What automation covers

Guard automates repeated health work without turning every event into a repository write. Audits can be requested manually or become eligible on a schedule. Improvement Jobs can run once or recur with a configured write boundary. Auto-pentest adds explicit target consent. Code review reacts to supported provider events and remains read-only.

Each automation belongs to an exact project, repository, audit or runbook kind, and—where relevant—website target. One schedule does not grant permission to work across neighboring repositories or hosts.

Manual and recurring work

Manual actions are useful for setup, verification after a merge, or an immediate current result. Recurring schedules maintain health over time, but a schedule is an eligibility rule rather than a promise of duplicate work at an exact second. Current access, active runs, repository changes, cooldown, account limits, backlog state, and inactivity pauses can affect whether work starts.

Run-now controls are independent from recurring switches where the product exposes both. Turning a schedule off preserves completed history.

Provider-triggered review

Code review can start when a non-draft pull or merge request is opened and on supported comment or note mentions. It resolves the exact provider target, understands scope, performs read-only checks, and posts a compact review reply when authorized. It does not create branches, commits, issues, releases, or a second review request, and it never merges or closes the original request.

Mentions are routed through a safety and scope check before any repository action. A safe project-specific question can receive a bounded answer; a request for secrets, destructive action, or unrelated external work is refused.

Pauses and backpressure

Automation can pause when a repository has been quiet or when Autofix issues and review requests accumulate beyond the safe backlog. Manual audit runs remain available during an inactivity pause. Guard surfaces the reason and requires a person to resume schedules after reviewing current state.

These pauses keep the improvement engine aligned with human review capacity instead of optimizing for artifact volume.

Choose the next page