About This Skill

Agent workloads are getting expensive, and most agents plan with zero cost awareness — re-reading huge files, re-running searches, picking premium models for trivial steps. Token Budget Officer adds a cost-governance layer that pays for itself in the first week. It triggers at task start and before expensive operations. For the planned approach it estimates cost-vs-value per step, then proposes cheaper paths that preserve the outcome: cache and summarize large inputs once instead of re-ingesting, downshift mechanical steps to cheaper models, dedupe repeated searches, trim context, batch small calls, and cap open-ended loops. It tracks cumulative spend against a session budget and warns at 70%, 90%, and 100% so an autonomous run can't quietly balloon the bill. Output is a PROCEED / OPTIMIZE / DEFER verdict with a budget-status line, a per-step estimate table, concrete optimizations with projected savings, and a screenshot-ready headline like "Plan cut from ~$2.95 to ~$0.57 — 81% saved." It never trades away a required outcome for savings; if a cheaper path risks the goal, it says so. The package is a lean SKILL.md orchestrator plus a bundled REFERENCE.md with the cost-estimation heuristics, a model-downshift table, the full optimization catalog, a budget-tracking schema, a worked example, and a ready-to-fill report template. It reasons over the rates and token counts you provide — it does not read billing dashboards, move money, or change account settings. Built by a credentialed reviewer bringing regulated-industry rigor to everyday agent work.