orc-budget

What problem does it solve? Before running a multi-task AI orchestration plan, you cannot know what it will cost in tokens, dollars, or subscription quota — and account-level dashboards only report spend after the fact. This Skill forecasts the cost of a specific plan forward, per scoring band, so you can choose a cheaper lane or split a risky task before any tokens are burned. ## Core Features & Use Cases - Plan-based cost forecasting: Takes a plan file (never a prose sentence) and produces a four-component token vector (input, cache write, cache read, output) per scoring band, always as a range with a sample count. - Four rendering views: Renders the same vector as raw tokens, dollars from a dated price table, percent of a 5-hour/weekly subscription quota window, and per-task context-window pressure. - Transcript-to-trace join: Joins Claude Code JSONL session transcripts with ORC dispatch traces to attribute real measured costs to tasks, with unattributed tokens always reported. - Use Case: Given a 14-task, 4-wave plan, ask what it will burn — get "1.23M raw / 564k weighted tokens, 18% of your 5-hour window" plus a warning that task T12 forecasts 189k of a 200k context window and should be split. ## Quick Start Ask the assistant to forecast what the current ORC plan will cost in tokens, dollars, and quota before running it.