rynfar

Meridian Plugin Openclaw Scrub — Development skill for Claude Code

Development community

Meridian plugin: stop OpenClaw traffic being metered as a third-party app by removing the prompt sections that trigger it.

How to install Meridian Plugin Openclaw Scrub

This entry records only its repository, not the path inside it, so there is no exact command to give. Open rynfar/meridian-plugin-openclaw-scrub and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Meridian Plugin Openclaw Scrub does

Meridian plugin: stop OpenClaw traffic being metered as a third-party app by removing the prompt sections that trigger it.

Alternatives in Development

Third-party Notices — THE FOLLOWING SETS FORTH ATTRIBUTION NOTICES FOR THIRD PARTY SOFTWARE THAT MAY BE CONTAINED IN PORTIONS OF THI 94.7k ★
Third Party Notices — Third-Party Notices 1.7k ★
Clawgod — This is NOT a third-party Claude Code client 837 ★

README

@rynfar/meridian-plugin-openclaw-scrub

A [Meridian](https://github.com/rynfar/meridian) plugin that stops [OpenClaw](https://github.com/openclaw/openclaw) traffic from being metered as a third-party app.

Why

Point OpenClaw at Meridian to use a Claude Max subscription and every request fails:

API Error: 400 You're out of extra usage. Add more at claude.ai/settings/usage and keep going.

The request is being billed as a **third-party app** — drawing from Extra Usage rather than the Max plan — and once Extra Usage is spent, nothing works. The trigger is in OpenClaw's system prompt.

What actually trips it

Measured on **2026-08-12**, against a Max account, by splitting a captured 30KB OpenClaw prompt into sections and replaying each one looking for `400 You're out of extra usage`. Three sections failed **on their own**:

Section	Source
the output-directive block (`## Reply Tags` in 2026.4.x, `## Assistant Output Directives` from 2026.7)	OpenClaw's system prompt
`## 💓 Heartbeats - Be Proactive!`	scaffolded `AGENTS.md`
`## Heartbeats`	scaffolded `BOOTSTRAP.md`

Two of the three are heartbeats — instructions for acting on a schedule with nobody watching. The third describes wiring replies into a chat surface. Together they read as an autonomous bot rather than an assistant answering a person.

**Two obvious suspects were not the trigger.** The `## Tooling` block listing `read`/`write`/`edit`/`exec`, and the 8KB skills index, passed cleanly when sent together — as did the `tools` array itself with the same tool names. A coding tool surface is not what gets flagged.

Read this before trusting the section list

**The classifier is not stable, and these findings are a snapshot.**

Later the same day, with `extraUsage` unchanged (`isEnabled: false`, `usedCredits: 0` — the same state that produced the 400), the *verbatim* directive block passed, and so did the full unscrubbed prompt on two consecutive attempts, on both a