contextburn: a run-efficiency meter for language-model coding agents

contextburn reads the local transcripts an LLM coding agent already writes and reports where the tokens actually went — by session, by machine and by reason. Beyond the usual counters it reports run efficiency: the share of paid tokens that became model output rather than a re-reading of context already sent. The share is given twice, by tokens and weighted by price, because the first is a property of agents and the second is a property of how the operator works. Nothing leaves the machine; the tool makes no network calls.