What problem does it solve? AI-agent bills grow fast and it is hard to see why. This Skill audits local agent session logs to quantify wasted read tokens (identical re-reads, subset re-reads, write read-backs, line-number scaffolding) and to check whether prompt caching is actually working, so you can find the recoverable share of your agent spend. ## Core Features & Use Cases - Read-waste audit: Runs agents_advisor_run over the last 30 days (adjustable 1-365 via window_days) and reports recoverable_usd against total_cost_usd with per-mechanism line items ranked by share of read bytes. - Cache-health diagnostics: Surfaces an alignment_score that flags volatile content (UUIDs, timestamps, JWT-shaped strings, hex digests) in prompt-prefix files, plus gaps_over_5m cache-death windows and residency_median context-residency stats. - History comparison: Uses agents_advisor_history to compare runs over time instead of recomputing. - Use Case: Your Claude Code bill doubled this month. Ask for an agent waste report; the Skill shows that identical file re-reads account for 40% of read bytes and that timestamps in your CLAUDE.md are churning the prompt cache. ## Quick Start Ask the agent to run an agent cost audit with yeaboi and show the recoverable spend and cache health for the last 30 days.
