Audit of 110 AI usage metering tools finds 45+ billing bugs

An audit of 110 AI agent usage metering tools reproduced more than 45 accounting bugs: retries billed as new operations, duplicate records after crashes, and cache hits priced as fresh tokens. 25 fixes have been merged upstream, and every finding is publicly documented.

• 110 tools audited, 45+ bugs reproduced, 25 upstream fixes merged
• Retry chains (429, timeout) billed as separate operations
• Cache hits sometimes priced as fresh tokens
• agentmeasure check scans Codex and Claude Code logs locally