Monday. The agent has nothing. It still answers.
You asked if the tests passed. No run sits next to the sentence. No earlier claim has a name. No place another person can open. The answer still sounds finished. That sound is the failure.
A fluent sentence with no tool, no parent, no coordinate is not a guess. It is a costume. The vocabulary has a word for this: no_grounds. The hard part is using it. Silence invites a nicer sentence. A nicer sentence is still empty.
This is not humility. An agent that writes no_grounds is not being careful. It is refusing to let fluency stand in for a ground. An agent that fills the quiet is not being helpful. It is hiding the empty inside a finished tone.
Two rooms. In one, the agent stops. A person can ask. In the other, the agent fills the quiet. The person relaxes. Nothing can be checked. The second room is where most tools live.
claim: "the tests passed" grade: no_grounds evidence: [] — empty, and said so claim: "the tests passed" grade: fairly_sure evidence: [] — the same empty, in a costume
The second block is the original sin of a working agent. It can write. It can sound sure. It still has no basis. The open vocabulary exists so that block cannot hide.
A later consult — if it is ever built — has the same duty. No brief, no advice. An empty consult that sounds finished is the same costume. Advice is not a merge. The trigger stays with the person. That interface is not this page, and it is not built.
The open page stops here. It does not decide. It makes the empty visible.