| name | cost |
| description | Run cc-cost on session transcripts: per-session cost, cache hit rate, tool-call distribution, top-expensive turns, optimization advice. Use for /cost or questions about token usage. |
| disallowed-tools | Write, Edit, NotebookEdit |
/cost — Local Claude Code Cost Diagnostics
Wraps ~/.claude/bin/cc-cost.py (from lob-labs/cc-cost, single-file Python, ~330 lines, no external deps beyond stdlib).
What it does
Parses session transcripts under ~/.claude/projects/<dir>/<id>.jsonl and reports:
- Per-session cost in USD + token counts
- Prompt-cache hit rate (load-bearing for cost optimization)
- Tool call distribution + per-tool token cost
- Top expensive turns
- Actionable cost-optimization recommendations (
--diagnose)
How to invoke
Default scan across all projects:
python3 ~/.claude/bin/cc-cost.py
Top N expensive sessions:
python3 ~/.claude/bin/cc-cost.py --top 10
Drill into one transcript:
python3 ~/.claude/bin/cc-cost.py /Users/chadsimon/.claude/projects/-Users-chadsimon/<session-id>.jsonl
Diagnose a specific session with optimization advice:
python3 ~/.claude/bin/cc-cost.py --diagnose <transcript.jsonl>
Machine-readable output for piping:
python3 ~/.claude/bin/cc-cost.py --json
When to suggest this
- User asks "how much did that cost" or "what's our token usage"
- A run blew through a dispatch budget — surface which turn was expensive
- Investigating low cache hit rate —
--diagnose highlights cache-busting patterns
- Planning a new agent — check whether the current similar agent type is cost-efficient before adding load
Out of scope
This skill reports on local transcript data only. Anthropic billing dashboard is the authoritative cost source; cc-cost is a fast local lens. If numbers diverge, the dashboard wins.
