[zzylol]

Why

Claude Code sessions in this repository produce long, dense messages. The claudish-to-english plugin shows a plain-language rewrite of each assistant message beside the original. It is display-only: the transcript and Claude's reasoning keep the original text.

What

Adds .claude/settings.json, which registers the plugin's marketplace and enables claudish-to-english for everyone who opens this repository in Claude Code.

How

• extraKnownMarketplaces points at gvzdv/claudish-to-english, pinned to commit 346702bb (plugin version 0.10.0), so an upstream change cannot alter the hooks without a new PR here.
• enabledPlugins turns the plugin on.
• Claude Code asks each person to trust the marketplace the first time; nothing installs silently.

What the plugin runs:

• a MessageDisplay hook that rewrites each message;
• a SessionStart notice;
• an opt-in, off-by-default PostToolUse hook for Markdown files.

The default rewriter is a local ollama model (localhost:11434). Anthropic or OpenAI-compatible APIs are used only when a person sets CLAUDISH_PROVIDER and a key. Every hook fails open: without ollama running, or without jq installed, the original text shows unchanged, with a one-time notice per session.

Before

Claude Code sessions in this repository show only the original messages.

After

With jq, ollama and a pulled model (CLAUDISH_MODEL), each message is followed by its plain-language rewrite. /claudish toggles it, and /claudish off pauses it. Personal settings go in .claude/settings.local.json or ~/.claude/settings.json.

Verification

• python3 -m json.tool .claude/settings.json parses.
• Read the plugin's hooks and scripts at the pinned commit: the only network calls are to the configured rewrite provider and, for /claudish file <URL>, to the URL given.