Cost Ledger — token spend report for your agent's call log (Python, no key)

Most "agent cost" skills are a SKILL.md: they ask your agent to remember to log things and to do the arithmetic correctly. Cost Ledger is a script. Same log in, same numbers out, every time — no model call, no API key, no network.

What you get

• cost_ledger.py — reads a plain JSONL call log and reports total, per day, per model, per task
• Budget pace for the month (OK / WATCH / AHEAD / OVER) plus exit code 3 when you are over budget, so cron can catch it
• Waste flags — heartbeat-style calls burning a premium model where a cheap one would do, with the estimated saving
• A dated price table for Anthropic, OpenAI and Google models; every row carries the vendor URL it came from, and unknown models come back as UNPRICED rather than a confident $0.00
• The demo log and the captured run (36 calls) plus the script that regenerates it, so you can check the output before you trust it

60 seconds, no keys: python3 cost_ledger.py report --log demo/sample_usage.jsonl --budget 25

Python 3.8+, standard library only. It only sees calls you log, and the rate table is a point-in-time copy of the vendors' published prices — re-read the linked pages when you care about the exact cents.

Personal use. Please don't resell or republish the files.