cost-check.md

model	haiku
description	Check and report current session cost and usage patterns

Cost Check

Analyze the current session and report on cost efficiency. Be concise — this command should itself be cheap.

Instructions

1. Session summary: Count how many turns have occurred so far in this conversation. Note the approximate scope of work done.

2. Token usage estimate: Based on the conversation length and complexity, estimate:

   • Approximate input tokens consumed (including CLAUDE.md overhead per turn)
   • Approximate output tokens generated
   • Which model has been used (Haiku/Sonnet/Opus)

3. Cost patterns identified: Flag any of these wasteful patterns if they occurred:

   • Full file reads when partial reads would suffice
   • Multiple search passes for the same information
   • Verbose explanations where concise answers were appropriate
   • Unnecessary exploration before making targeted edits
   • Reading files that were not ultimately used

4. Recommendations: Suggest 1-3 specific actions to reduce cost for the remainder of this session. Examples:

   • "Switch to Haiku for the remaining simple edits"
   • "Use Grep with specific patterns instead of reading full files"
   • "Batch remaining edits into fewer turns"

Output Format

SESSION COST REPORT
-------------------
Turns so far:     {n}
Model used:       {model}
Est. input tokens: ~{n}k
Est. output tokens: ~{n}k
Est. session cost: ~${n.nn}

PATTERNS FOUND:
- {pattern 1}
- {pattern 2}

RECOMMENDATIONS:
1. {recommendation}
2. {recommendation}

Keep the report under 20 lines. Do not elaborate beyond what is shown above.