Anthropic Claude Code HERMES.md billing flaw: Difference between revisions

'''Anthropic Claude Code HERMES.md billing flaw''' was a technical defect in Anthropic's Claude Code product that bypassed flat-rate subscription plans to charge users direct API fees. In April 2026, a flaw triggered by the exact string "HERMES.md" in a user's Git commit history routed subscribers directly to pay-as-you-go billing, charging one account over $200.<ref name="github_hermes">{{Cite web |url=https://github.com/anthropics/claude-code/issues/53262 |title=HERMES.md in git commit messages causes requests to route to extra usage billing instead of plan quota #53262 |author=sasha-id |website=GitHub |date=2026-04-25 |access-date=2026-04-26}}</ref>

The flaw emerged in the context of a wider crackdown by Anthropic on third-party agent harnesses. On April 4, 2026, Anthropic began blocking Claude Pro and Max subscribers from routing their flat-rate plan usage through third-party agentic tools, requiring those tools to be billed under a separate pay-as-you-go "extra usage" tier instead.<ref name="techcrunch_openclaw">{{Cite web |url=https://techcrunch.com/2026/04/04/anthropic-says-claude-code-subscribers-will-need-to-pay-extra-for-openclaw-support/ |title=Anthropic says Claude Code subscribers will need to pay extra for OpenClaw usage |website=TechCrunch |date=2026-04-04 |access-date=2026-04-26}}</ref> Boris Cherny, Anthropic's Head of Claude Code, stated on X that Claude's "subscriptions weren't built for the usage patterns of these third-party tools" and that the policy "applies to all third-party harnesses and will be rolled out to more shortly."<ref name="techcrunch_openclaw" /><ref name="venturebeat_openclaw">{{Cite web |url=https://venturebeat.com/technology/anthropic-cuts-off-the-ability-to-use-claude-subscriptions-with-openclaw-and |title=Anthropic cuts off the ability to use Claude subscriptions with OpenClaw and third-party AI agents |website=VentureBeat |date=2026-04-04 |access-date=2026-04-26}}</ref>

[https://hermes-agent.nousresearch.com/ Hermes Agent], an open-source self-improving AI agent built by Nous Research, is among the third-party harnesses affected by the rollout. Hermes can route requests to Anthropic models either through pay-per-token API keys or by reading Claude Code's credential store on the same machine, the latter of which became the target of Anthropic's enforcement actions.<ref name="hermes_providers">{{Cite web |url=https://hermes-agent.nousresearch.com/docs/integrations/providers |title=AI Providers — Hermes Agent |publisher=Nous Research |access-date=2026-04-26}}</ref>

On April 25, 2026, a user reported that the exact string "HERMES.md" in their local git commit history caused Claude Code to route queries outside of their fixed-rate subscription plan.<ref name="github_hermes" /> The system bypassed the $200 per month Max plan and charged the user direct, pay-as-you-go API rates.<ref name="github_hermes" /> The user lost over $200 in extra charges while 86 percent of their prepaid plan capacity remained available.<ref name="github_hermes" />

The string "HERMES.md" appears to have been treated by Claude Code's billing classifier as a signal of third-party harness usage — likely as part of the broader detection logic deployed to enforce the April 4 policy — even when the string appeared only in incidental git history rather than as evidence of an active Hermes session.

When the affected user contacted Anthropic support, the company refused to issue a refund for the $200 lost to the glitch, categorizing the overcharge as an un-refundable technical error.<ref name="reddit_psa">{{Cite web |url=https://www.reddit.com/r/ClaudeAI/comments/1svdm1w/psa_the_string_hermesmd_in_your_git_commit/ |title=PSA: The string 'HERMES.md' in your git commit history silently... |website=Reddit |date=2026-04-26 |access-date=2026-04-26}}</ref>

{{reflist}}

[[Category:Anthropic]]

[[Category:Incident]]