Simulated: the agent, the gateway, and every tool below are scripts. No real agent, service, or standard is used.
1. Write the task contract
agent-intent.json Proposal, not a standard
2. Chat with the simulated agent
Time left:
Press a button. The agent tries that tool call. The gateway decides.
3. What the gateway sees
This panel updates on each call and each contract change.
4. Decision timeline
No calls yet. Press a tool button above.
How it works
The diagram replays the last tool call, one step at a time. Before any call it shows .
Press Play or Step.
Pain point
A person can limit which tools an agent calls, but not what it may achieve.
Problem statement
Teams give an agent broad tool access or approve each step by hand. Neither lets a person state limits.
What sparked it
Mozilla's W3C WebExtensions group notes (issue 1041) say no browser-level primitive lets a user see, scope, or withdraw an agent's access.
Why this approach
The person writes a short task contract. A gateway checks every tool call against it.
Hypothesis tested
We believed that a short contract with limits and an expiry gives a person enough control. If it holds, the gateway allows, blocks, or asks for each call correctly.
How to try it
- Open the Prototype tab. Move a tool between lists in the contract form.
- Press the tool buttons next to the chat.
- Speed up the clock to see what happens after the contract expires.
What this does not do
- It does not connect to a real agent or tool.
- The format agent-intent.json is a proposal, not a standard.
Notes
- Who would pay: the platform or security team of a company that lets agents act.
- Weak spot: a contract is only as good as its author.
- Feedback: which rule would you add to the contract?
Traces
| Step | Prompt | Model | Tool | Date |
|---|
Cost
| Step | Model | Input tokens | Output tokens | Cost (USD) |
|---|---|---|---|---|
| Total | ||||
Measured: the author runs used 190,769 tokens in total. Estimated: the split into input and output tokens (90 percent input, 10 percent output) and the dollar cost, at $2 input and $10 output per million tokens for Sonnet 5.5 (about $0.53). The run reported one total only. The lead session's planning, research, and publishing are not included. The prompts in the Traces section are listed in the order they were given.