Per-tenant "Directives to Activate" — choose, vary and protect agent behaviours

What it is

A settings pane where each tenant chooses which standing agent behaviours ("directives") are active. More than a hundred behaviours that used to live scattered across config files and memory are pulled into one governed catalog.

Why

• Nobody should have to retype standing orders every session.
• A tenant can use the owner's version of a behaviour or their own variation. For example, "be concise" can become a dropdown of styles, or a custom one saved to their profile.
• The directive corpus is intellectual property. A tenant must not be able to bulk-export it, paste it somewhere else and walk away with all of it.

The pieces

1. Catalog — the canonical list, grouped by category.
2. Activation and override — a checkbox (behaviours are toggled, never deleted), a variation dropdown, and "Add…" for a custom one.
3. Per-directive access — reveal, hide, or scope a directive to specific tenants. Tenants see only what they're allowed to, with no bulk view or raw export.
4. ASSESS — when someone adds a custom directive, it's classified as fits, duplicate, conflict, vague or out of scope, with a suggested rephrase.
5. A full-corpus auditor — checks the whole active set for conflicts, redundancy and incoherent combinations, and proposes a precedence order.
6. A master inspector — the owner sees every active behaviour and the existence of internal controls. Tenants see some controls only as "exists, here's what it does", never their internals.
7. Anti-clog guardrails — tied to token burn-watch, so the directive set can't bloat until the agent stops working well.