XFVE
XF Voice Enforcement

Every response your AI gives — plain language, actionable, never leading with jargon.

Claude is trained on a corpus that's majority developer-speak. Left alone, it defaults to file paths before meaning, function names before plain English, and neutral lists where a recommendation would actually help. XFVE enforces three invariants on every completed response: the answer must be actionable, it must lead with a recommendation when options exist, and it must open in plain language before going technical. It fires as a Stop hook on every response — not on code, not on data, but on every line the AI delivers to you.

Part of Quality Gates

Connect with GitHub — it's free

Install

XFVE activates automatically with the Xpansion Framework installation. No separate setup. Once XF is running, XFVE fires on every session.

To confirm XFVE is active: look for the ◈ XFVE stamp in your session or run /xfve-mode in Claude Code to inspect enforcement settings.

What it does

1

Actionable responses — no pure refusals.

A response that says 'I can't do that' and stops is a wall, not an answer. XFVE flags pure refusals and requires a path forward — an alternative, a clarification request, or a constrained version of what was asked. Every response leaves you with something to act on.

2

Lead with the recommendation — not a neutral list.

When multiple valid options exist, the default AI response is a numbered list ending with 'you decide.' XFVE enforces the recommendation pattern: one pick, one reason up front, alternatives listed below. You get a decision, not a delegation back to you.

3

Plain language first — meaning before mechanics.

File paths, function names, and stack traces belong second. XFVE flags responses that open with unexplained technical terms and requires context before jargon. Your AI speaks in plain English first, then backs it up with technical detail on request.

4

Fires on every response — not just on code edits.

XFBA and XSIA protect your code. XFVE protects the conversation. It runs as a Stop hook on every completed response — so the quality layer covers not just what the AI does to your files, but what it says to you.

How it works

1

Claude completes a response. The Stop hook fires.

2

XFVE runs three semantic checks against the response: actionable, directive, plain-language lead.

3

Compliant response: clean pass, no stamp, no friction.

4

Violation: XFVE writes a notice to the pending file, which blocks the next tool call with a plain-English explanation of what broke.

5

The violation notice names the rule and describes what to watch for — no jargon, no stack trace.

● ● ●
◈ XFVE ✗ neutral list — 4 options with no recommendation
rule: directive — lead with a pick, then alternatives
◈ XFVE ✗ jargon lead — response opened with function name before context
rule: companion — plain English first, technical detail second
◈ XFVE ✓ voice compliant
Other modules