N-gram autocomplete silently ingests other tools' prompt history (Codex and Claude) by default

[BonzoIPNMadrid]

I just started trying out OMP recently and noticed the word autocomplete feature in the composer, neat!

But, I then noticed that some unique strings were being predicted that were only used in my Codex prompt history. I looked into it and nowhere in the docs is it noted that OMP reads prompt history from other tools like Codex, Claude, etc when word autocomplete is enabled.

See packages/coding-agent/src/predict/foreign-history.ts, it reads:

const files = [
		path.join(resolveClaudePaths().configDir, "history.jsonl"),
		path.join(os.homedir(), ".codex", "history.jsonl"),
	];

No gates or checks to enabledProviders or otherwise before ingesting prompt history from Claude and Codex.

This feels inconsistent with the behavior of other features related to providers, per docs/settings.md:

enabledProviders defaults to empty
user roots from Cursor, Codex, Claude… are normally excluded.

"Normally" being the key word, this behavior is technically consistent with that statement although it does seem to be the only relevant exception that I can find.

The N-gram option in the Word Autocomplete options states "Learns your vocabulary from prompt history" but that doesn't indicate how it includes prompt history from Codex and Claude.

Suggestion - Either:

1. Update the foreign-history.ts default behavior to gate prompt ingestion behind enabledProviders or via a setting like "Include Codex/Claude prompt history" adjacent to the Word Autocomplete settings. Also, update docs and descriptions.

   or

2. Clearly state in documentation that prompt history from other providers is ingested by default when N-gram or SmolLM are used for Word Autocomplete.