Operator Directives
The Directives page (/governance/directives) lets operators author natural-language business guidance that is injected into agent system prompts at runtime.
Current readiness
| Area | Status | Notes |
|---|---|---|
| Directive CRUD (create, edit, pause, delete) | Live | Full lifecycle management in the dashboard. |
| Scope-based filtering (fleet, team, agent) | Live | Directives apply only to agents matching their scope. |
| Priority-based ordering | Live | Higher-priority directives appear first in the injected prompt section. |
| Impact measurement / A/B testing | Planned | Before/after behavioral comparison not yet available. |
What the page is for
Use Directives to:
- Inject business rules into agent behavior without modifying code
- Apply compliance guidance across the fleet or to specific teams
- Enforce tone and style guidelines
- Pause or activate directives instantly
Typical workflow
- Click Create Directive in the upper right.
- Write the directive content in plain language (up to 2000 characters).
- Choose a scope: fleet-wide, a specific team, or a single agent.
- Set priority (1-10) and category (safety, business, compliance, tone).
- Save as draft or activate immediately.
- Monitor directive status in the card-based list view.
- Pause a directive to immediately stop its injection without deleting it.
How directives reach the agent
Directives are fetched via the policy sync infrastructure and injected into the system prompt by _apply_education() in the SDK plugin base. They appear after education templates and custom rules. All 7 framework plugins support directives.