How do my agents know what I know?

My agents start each task with the decisions that govern it. I keep that knowledge in Cortex as Modules, and each Module declares what an agent must understand first. Before an agent plans or changes anything, it flashes the Module that owns the task and receives that Module and all its dependencies, foundations first. When our work changes what should be true, the agent proposes an exact Module update for me to approve.

By Eduarda FerreiraPublished Updated

Download the flash bundle for this module.

The problem

A fresh agent can inspect a repository and learn the current implementation. My approved rules and the reasons behind them are separate knowledge. I used to repeat those rules in conversation, and a missed constraint could undo work we had finished. Manual briefings could not cover hundreds of agent sessions a day.

Modules

A Module holds one durable subject and records what should be true and why. To learn what exists now, the agent inspects the repository.

I keep a Module for each product and for shared subjects such as writing and review. My billing Module depends on my Account Module because the billing rules need the Account boundary. Those edges form a comprehension graph called the SemDAG.

The guide to authoring a context Module explains how to find the existing owner, write the specification, and get exact approval.

Flash

Flash pulls the knowledge an agent needs for one task. It starts from the Module that owns the subject, follows every dependency, removes duplicates, and returns one bundle with foundations first.

My agents read that bundle before they plan or edit. They still inspect the repository, history, services, and tests because those show the current state. The Flash guide gives the command and explains how to find a Module when you do not know its name.

My standing instruction is: “Before acting, ensure the demand is an Issue owned by the narrowest Module, then flash the Module.” I use Cortex Issues to route all work. A Flash only needs a Module name, so you can use the same knowledge with your current task tracker.

Writing back

Useful knowledge appears during work. When I explain a changed decision or correct an agent, it finds the Module that owns the subject and drafts an exact change. I approve, revise, or reject those words. Cortex saves each approved version, and future flashes receive the current one.

An active session keeps the bundle it already loaded, so the agent flashes again after a Module changes. The guide to keeping Modules current explains how we compare approved intent with current evidence.

You stay in control

Each Module has one owner. Access roles decide who can read, change, or manage it. An approved rule remains in force until its owner approves an exact change. The guide to Module approval and conflicts explains what happens when a request disagrees with a Module.

Adding a dependency is also a disclosure decision because a person who can flash the parent receives the dependency in that bundle. Cortex stores only resources that you or your agents submit. Installing the CLI or connecting through MCP does not scan or upload your repository. Cortex does not call a model API when it stores, searches, versions, or flashes Modules. You can read the full explanation of company data and Cortex access.

Any agent can use it

An agent uses the Socra CLI in a terminal or connects through MCP from a compatible client. Follow the coding-agent integration guide to connect Cortex to your agent. Before inspecting the same change, builders and reviewers can flash the same Module.

Your repository instructions, tests, CI checks, review requirements, and deployment approvals continue to control what ships. The guide to trying Cortex with your existing development process gives a low-risk pilot that keeps those controls in place.

What this carries

My agents use the same approved Modules when they build and review work. The swarm can continue while I sleep or travel because each agent can load the knowledge without waiting for a briefing. It builds and improves dozens of products and closes hundreds of Issues a day.

Start

Read Get started with Cortex to sign in, search the graph, and run your first Flash. Then choose one real decision that you keep explaining or that agents often miss. How much work does it take to get value from Modules? shows how one useful Module can help.

Search before you create it. If no relevant Module exists, continue the task and create one only when you have durable knowledge to preserve. Start with a reversible change in one repository where tests already cover the affected behavior.

Add this instruction where your agent reads it: “Before changing this area, flash <module-name> in Cortex and read the complete result.”