System Rules

System Rules are the standing instructions that shape how the product's in-app chat behaves. They are written in plain language and apply to every conversation your team has with the ContextCache assistant.

What they are

A system rule is a short, natural-language statement that tells the assistant how to behave in the dashboard's own chat experience. Typical rules cover tone ("respond concisely"), priorities ("always check the Decisions database before proposing a change"), and guardrails ("never send external messages without confirmation").

Rules live at Dashboard → System Rules and apply across every cache when you use the in-app assistant.

Workspace rules vs cache instructions

There are two places to shape AI behavior, and they serve different purposes:

Workspace System Rules (/dashboard/rules) apply to the product's own chat on web and mobile. They are global across every cache.
Cache system instructions (under each cache's settings) shape AI behavior when that specific cache is the active context. Use them for facts, tone, and priorities unique to one body of work.

External MCP clients such as Claude Desktop or Cursor do not inherit Workspace System Rules. They use their own configuration.

Writing effective rules

The goal is context, not rigid scripting. A good rule describes a preference or a fact in one short sentence; an overloaded rule list pulls the assistant in too many directions.

Be specific. Prefer "summaries should be under 150 words" to "keep it short".
State facts once. If the assistant should always treat Q2 targets as confidential, say so in a rule instead of adding the reminder to every prompt.
Keep the list small. Aim for five to ten rules. More than that and the assistant starts to compromise on the ones that matter most.
Revise over time. Rules are editable. When you notice the assistant repeatedly getting something wrong, add or refine a rule instead of correcting the response each time.

How rules compose at runtime

When the assistant responds, the product composes its prompt in a fixed order:

Workspace System Rules — global behavioral grounding.
Cache system instructions — context specific to the active cache.
Blueprint instructions — if a blueprint is running, its workflow-specific prompt.
Resolved inputs — the typed parameters the run supplies.
Data bindings — live cache content or integration data pulled at runtime.

Rules set the tone; blueprints and bindings provide the specifics. This order is intentional: rules give the assistant ground truth before any task-specific framing arrives.

Managing rules

Rules are added, edited, reordered, and deleted from Dashboard → System Rules. Order matters when two rules conflict, so place the ones that should take precedence higher in the list. Changes take effect on the next conversation turn.