Build Mechanic Runs
Build Mechanic Runs shows what your AI agent runs cost. Every /build-mechanic and /update-docs run of the agent workflow is recorded with its cost, time and tokens, split per agent and, for debug runs, per step.
Open it from Tools → Power → Build Mechanic Runs. It needs Unity 2022.2 or newer; on older versions the window isn't available.
The log is .claude/build-mechanic/runs.json in your project's root folder, outside Assets. It is empty until your first run, and it stays on your machine.
When to use it
- What did that mechanic cost? Each run shows its total, and how much went to building the mechanic and how much to improving the docs.
- Is it getting cheaper? The chart shows the trend across runs, for example after a docs change or a model switch.
- Where does the money go? In a debug run, see which step of which agent was the most expensive.
Start the request with --debug (/build-mechanic --debug <request>) to get a cost report per step at the end of the run, in the chat. The run log then has the steps too.
Runs
Each row is one run: Date, Label, Kind (build-mechanic or update-docs), Outcome, Cost and its Build / Docs $ split, Active and Wall time, API and Tools calls, New ctx (new context tokens), Gaps (rounds where the builder reported a docs gap), Cold tests, Fixes (your corrections), Debug and the Docs version the run used.
| Control | Does |
|---|---|
| Kind filter | Shows all runs, or only build-mechanic or update-docs runs. |
| Chart | The metric the chart plots per run: cost, active minutes, API calls, new context or cache reads. Bars are split into build and docs work. |
| Search | Filters runs by label, request and notes. |
| Reload | Reads the log again. The window also reloads when you focus it after a new run. |
| Show JSON | Shows runs.json in your file browser. |
A run's details
Select a run to see its request, outcome and notes, and its Agents: role, model, cost and share of the run, active time, API calls, cost per call and context sizes. Select an agent to see its Steps (debug runs only): the step name, calls, time, tokens, cost, share of the agent's cost and the power.py commands it ran.
Ctrl/Cmd-click two runs to compare them: the cost, the builder's cost, active time and more are shown side by side with the change in percent, above the details.
Costs are estimates from the agents' transcripts, priced per model from AiFlow/build-mechanic/prices.json. The output cost is a lower bound, because thinking isn't logged.