AgentMeasure
Check Codex / Claude Code logs for duplicate records, retries and token anomalies. Verify agent metering with reproducible evidence.
Website · 中文 · Healthcheck guide · Release notes
Try it locally
pip install agentmeasure
agentmeasure demo # synthetic example; no account or upload
agentmeasure check # inspect your local Codex / Claude Code sessionsThe demo shows findings from a bundled synthetic session. Your own check reports coverage, failures, retry chains and available token evidence. Missing evidence stays unknown; a retry is not automatically a second logical operation.
Python 3.9+. CLI 0.7.1 on PyPI; Provider SDK 0.2.0 on npm, verified 2026-10-15. These packages have separate version numbers.
Open the HTML report at the path printed by the command. On macOS/Linux use
python3 -m pip if needed; on Windows use py -m pip. For permission or
“externally managed” errors, use the virtual-environment instructions.
With both runtimes present, auto-discovery prefers Codex; use
agentmeasure check --runtime claude for Claude Code. No logs in the default
seven-day window? Try --all or an explicit --dir and --runtime. A missing
input is not a passing check. More recovery steps.
Get help with one question
| Option | Price | What you get | |---|---|---| | Run the open-source tools | Free | Local Healthcheck, conformance checks, SDK and published rules | | Healthcheck Diagnosis | $10 | One anomaly on one machine: interpretation of your sanitized summary, evidence limits and next checks | | Metering QA | $20 | One event sample, one ruleset and one metering question: supported checks, findings and a rerun command |
See scope, input requirements and apply by email. No subscription or sales call. First response within 24 hours. Delivery within 24 hours after payment, fit approval and complete input; queue changes are agreed before payment. A confirmation reserves a place; it does not start paid work. Card checkout is being set up; applications and price acceptance are separate from payment.
$10: counts alone support interpretation and general triage. Specific troubleshooting needs an optional manually redacted error excerpt; a root cause is not guaranteed. $20: fit is checked free before payment. Core FMT-002 checks operation/attempt accounting and sample cost conservation; it does not verify token/cache semantics, vendor invoices or pricing accuracy.
To prepare a summary locally:
agentmeasure check --json report.json
agentmeasure share report.json --out summary.json
agentmeasure validate summary.json--out writes immediately on your machine. Inspect the file before you choose to
send it. Nothing is automatically uploaded. Refund terms and evidence limits are
on the service page; “no finding” and “not enough evidence” are valid outcomes.
No local mail app? Download an application template or copy the email. Attachments are manual. Do not send the full report.json, HTML report or raw logs when only a summary is requested. Tool bugs can be reported free in Issues, with a redacted reproduction.
For tool maintainers and providers
| Resource | Use it for |
|---|---|
| Conformance pack | Check fixtures and reported metrics; PASS / FAIL / UNPROVABLE |
| Provider SDK | Capture canonical observations from MCP tool handlers: npm install @agentmeasure/mcp |
| Core specification and AMS-1 | Define measurement grains and the evidence needed for claims |
| Claims index | Inspect dated findings, reproductions and upstream responses |
| Lab | Existing experimental tooling; no hosted product commitment |
Buyer reconciliation tools remain available as an experimental reference. They are not one of the two current fixed-price services. Current development priorities are in the roadmap.
Evidence and limits
The September audit report documents the 110-tool review. Individual findings and subsequent corrections are linked in the claims index. Engineering evidence does not establish paid demand, customer savings or a right to a refund. Customer value and repeat purchase remain under validation.
479 Healthcheck unit tests passed on 2026-10-15. See CI for current results and v0.7.1 for the release scope. Example fixtures are not production customer data.
Open-source commitments
- MIT-licensed tools and public rules remain free. Paying does not change a verdict.
- Analysis runs locally. Sharing a sanitized sample is your explicit choice.
- No paid rankings or recommendation slots; no payment custody.