FB. SKETCHBOOKMOTION STUDY / 09

A SKETCH TO EXPLAIN THE SYSTEM

Inside the agent harness.

The harness connects context, tools and checks around the model.

A goal. Actions.
Results to verify.

FIG. 01 CUTAWAY VIEW
The harness around the model A pencil-drawn workshop: a human defines a goal at his desk, context is prepared, the model proposes an action, and the harness checks permissions and calls a tool. Observations return to the model even after successful actions. A failed test calls for diagnosis: fix within the agreed limits or hand off. Checks are tied to V1, then V2. The result is submitted for review; the human stays in control. 01 / THE BRIEFAn outcome and acceptance criteria. 02 / CONTEXTThe right context, when needed. 03 / THE MODELPropose the next action. 04 / TOOLSAuthorized action → execution ACTION COMPLETED ✓The result returns to the model. V1 / TEST FAILEDInspect the failure, then decide.Progress + budget → fixOtherwise → hand off V2 / CHECKING AGAINCheck the revised version. V2 / CHECKS PASSED ✓Ready for review. YOU STAY IN CONTROL.Goal · limits · decisions
Preparing the sketchbook…
DRAWING THE WORKSHOPBLUE WORK PACKAGE / PENCIL WORKSHOP
01

Define the outcome.

You set the goal, the expected result and the criteria used to check it.

Read the full explanation

    If execution is blocked, the human handoff includes previous attempts, the blocker and the decision needed. A test that did not run has an unknown result; it is not a failed test.

    This cutaway is a teaching metaphor. The Crew represent roles: a harness can run with a single agent. The pipe represents returning observations, not retraining the model. Tools, checks and human approvals depend on the system. Observations inform the next decision; a passing test alone does not establish that the whole request has been fulfilled.

    Further reading: From code to merge · The agent loop · What a check establishes · Following work through to use