vibe-costlisted

# Vibe Cost Skill Tracks token usage and dollar cost at every granularity level. Auto-estimates when running automatically. Accepts real numbers on demand. Shows where the money is going and how to spend less of it. --- ## The principle You can't optimise what you don't measure. Most teams using Claude Code have no idea which tasks are burning the most tokens, whether costs are trending up or down, or whether a simple prompt change would cut their bill in half. This skill measures first, then recommends. Never recommends a change without data to back it. --- ## Two modes ### Mode A — Automatic (end of session) Triggered automatically by other vibe-* skills at session end. Uses estimates based on what the session did — files read, tasks completed, context size, CODEBASE.md length, agent patterns detected. Precision: ±20-30%. Good enough for trending. Not precise enough for billing. ### Mode B — Precise (manual invocation) User runs `cost:` and pastes the output of `/cost` from Claude Code. The skill reads real token counts, calculates exact dollar costs, compares against estimates, and produces a precise report. Also available: user can run `cost:` with no paste — skill uses latest estimates. --- ## Step 0 — Detect invocation mode and context ```bash # Check for existing cost data ls vibe/cost/ 2>/dev/null && echo "HISTORY EXISTS" || echo "FIRST RUN" # Read project context cat CLAUDE.md 2>/dev/null | grep -E "model|MODEL" || echo "model not specified" cat vibe/TAS