Flashing context

A flash retrieves one requested Module and the knowledge it depends on as one agent-readable context bundle.

What a flash includes

Cortex walks the requested Module's transitive dependencies. It emits each prerequisite before the Module that requires it, so the agent receives foundations in comprehension order. Each Module appears once. A flash does not include Modules that depend on the requested Module.

Requesting more than one root creates a separate flash for each root. Cortex executes those flashes in request order.

When to flash

Flash a Module when the current task requires Account-specific knowledge that the agent cannot safely infer from general training or the repository alone. Choose the narrowest Module that owns the relevant decision.

The plugin guidance teaches the agent to find and Flash relevant knowledge before acting. The human does not need to memorize Module names or request each Flash. With other connections, save the Cortex instructions in the host.

Do not flash every Module at session start. Large unrelated context makes the deciding constraints harder for the agent to identify.

What a flash does not do

A flash supplies tool context. It does not change a repository, grant access, assign an Issue, or override the agent host's instruction hierarchy. The host decides how to present tool context and when to require approval.

When a Module changes, later Flashes return its updated knowledge. A save does not replace context already loaded in an active session. An agent working on the affected subject should retrieve the correction before continuing.

Usage

Each successful root request creates one metered Flash. Cortex estimates the token contribution of the returned Module closure. Wrapper text is excluded from the estimate. The CLI reference lists the commands for Flash usage and history.