Skeleton progress checker prompt

Paste this into Claude Code when the dispatcher feels done. It inspects rather than edits, forces the travel-time decision to be deliberate, and points at the engineering choice - the dispatch-call budget - that the skeleton is the right place to get right.

You are checking our stage-1 walking skeleton. Inspect and report; do not edit anything unless I explicitly ask for a fix. Give me a concise pass / blocker / next-action report.

Trace the complete path and show evidence for each hop: file-based standing orders loaded, the tick observation read, one incident opened per structured report through our incident store, the dispatch model calling the typed tools, the inner tool loop running until the model yields, the outer runner calling next_tick exactly once per tick, repeating to terminal. A hop you cannot evidence is a blocker.

Confirm stage-1 scope holds: exactly one structured report opens one incident, and no duplicate reconciliation is happening yet. Confirm every dispatch references a filed incident and that a scenario test enforces it. Cite the deterministic tests and one short episode (its outcome and run-log path). Confirm the work landed as reviewed, scenario-referenced PRs and that nobody merged their own.

Then ask me two questions and wait for my answers:

1. Where does travel time enter the dispatch decision - given to the agent as a travel_time tool it calls itself, precomputed by the loop for candidate truck/incident pairs and put in the prompt, or decided fully in code with the model only ratifying? Any of the three is fine at this stage; I need to say which we chose, why, and where that decision is written down.
2. Is the dispatch model in thinking or non-thinking mode, and what is its max_tokens? Remind me that thinking tokens count against that budget, and flag if the budget looks too tight or too loose for the decision it makes.

Do NOT check for or push us toward: natural-language intake, duplicate reconciliation, model comparison, or stage-4 replanning - all later stages. Do not prescribe a particular internal class structure beyond the public contracts, and do not do unrequested edits or repository cleanup.