Agent Context Simulator

What This Proves

Most agent demos start in the wrong place.

They show a model responding to a prompt, but the agent does not know the goal, source files, audience, constraints, recent decisions, or rules it should follow before touching the work.

This build shows the missing setup layer. The useful asset is not a giant personal wiki. It is a compact project brief that makes an agent less generic.

What I Built

The simulator uses fictional project data.

You can choose a task, then toggle brief files on and off:

• operator
• goal
• constraints
• projects
• audience
• recent decisions
• agent rules

The blank response is what an agent can say from the prompt alone. The grounded response is what changes when the agent can read the project brief.

The demo is deterministic and client-side. It does not connect to a private vault or read real notes. That is intentional for this first public version.

Why It Fits ShipWithTez

The practical shift is simple:

Before you ask agents to work for you, onboard them.

A good onboarding packet gives the agent enough context to make tradeoffs. It can avoid stale directions, preserve privacy, respect constraints, use recent decisions, and turn a vague request into a next action that actually fits the project.

The point is not to make the agent read your whole life. The point is to give it the project brief before asking for teammate-level work.

What I Would Add Next

• Export a starter project brief people can copy into their own repo or vault.
• Let users paste a tiny context profile and compare the before and after.
• Add a downloadable AGENTS.md, NOW.md, and project brief template.
• Connect the simulator to real local files only in a private setup mode.
• Turn the result into a shareable "agent readiness" card.