Claude Code appears to detect “OpenClaw” references in git history or chat and triggers session disconnects or usage-limit burns.
Key Takeaways
The behavior surfaces in Claude Code when commit messages, file contents, or linked URLs reference OpenClaw, a competing AI tool.
Affected users report immediate session termination and usage counters jumping to 100% or hitting 5-hour limits after OpenClaw mentions.
No official Anthropic statement explains whether this is intentional policy, a keyword filter, or an accidental pattern match.
The trigger appears broad: git commit messages containing OpenClaw schema strings, blog post text, and direct links to openclaw.ai all reproduce it.
Hacker News Comment Review
Multiple commenters independently reproduced the disconnect with minimal test cases, including a bare git repo with an OpenClaw JSON schema in the commit message, strengthening the case this is systematic not random.
Commenters debate whether this is intentional competitive blocking or sloppy keyword fingerprinting, but converge on the view that Claude Code’s deep filesystem access makes any such scanning especially invasive and unacceptable.
Several users cited this alongside other trust incidents (Google banning an agent user, quality regressions) as the final push to cancel paid subscriptions and move to self-hosted models like Qwen 3.
Notable Comments
@abdullin: Reproduced with a four-command git repo; commit message containing openclaw.inbound_meta.v1 caused immediate disconnect and 100% usage burn.
@data-ottawa: Points out the specific harm to legitimate use: running OpenClaw locally in a NixOS QEMU sandbox with Claude-assisted setup should not be flagged.
@jamescontrol: “these things where they ‘accidentally’ whoops match on specific keywords and trigger extra usage charges is giving a evil-microsoft-vibe”
