Act 1 · Prepare
Steps 1–3
Frame the work
Bound the goal, draft a plan, load only the context worth trusting. Misframing here scales bad output faster than later steps can recover.
What an agent run actually looks like — 9 steps, 3 acts.
Click any step to see what the harness is doing, what evidence proves it ran, and what fails when this step is missing. Three acts: Prepare (frame the work), Execute (do the work inside policy), Govern (prove it, escalate, and close the loop).
Act 2 · Execute
Steps 4–6
Do the work inside policy
Use sandboxed tools to inspect the system, update state, and verify each step proved itself instead of just narrating progress.
Act 3 · Govern
Steps 7–9
Prove it, escalate, close the loop
Handle exceptions explicitly, package decision-ready evidence, and return final authority to a named human owner. Memory carries the lesson forward.
Act 1 · Prepare · Step 1
Bounded goal is accepted
Click any step above to open its detail.