# Xpansion Framework

> Code enforces what the LLM cannot be trusted to do reliably alone.

Xpansion Framework is a seven-module enforcement layer for Claude Code. XFFI defines intent as a finite MECE spec before work begins. Six enforcement modules (XFTD, XFBA, XSIA, XFTC, XFXA, XFMT) execute against that spec — routing tools, catching contract violations, mapping cascade impact, managing context, verifying delivery, and building a living memory tree across every project you've ever touched.

The architecture: your vision flows through Claude Code, which the five modules enforce and protect. Everything they capture accumulates in XFMT — a living memory tree organized around four immutable boundaries: Build, Understand, Specify, Operate. Every Claude Code conversation routes into one of these four branches. Memory below each branch grows organically as you work.

## The seven modules

### XF Tool Dispatch (XFTD) — free
Routes Claude Code to the best available tool for every task. Scans 16,500+ MCPs, plugins, and skills ranked against your actual stack. Blocks Claude Code when a marketplace tool scores 10+ points higher than the default. Fires on every prompt — silent when nothing beats the default, visible when something does. One session-end digest confirms it ran.

### XF Boundary Auditor (XFBA) — free (core), Pro (full)
Fires on every Edit and Write before the change reaches disk. AST scan + cross-file call graph check. Catches: wrong argument count, broken imports, type mismatches, missing environment variables. Surfaces a plain-English repair plan. "Fix it, suppress it, or proceed?" — you decide. The "my bad" loop ends here.

### XF System Impact Auditor (XSIA) — Pro
Runs immediately after XFBA on every edit. Maps blast radius: every caller affected, every downstream dependency, every shared state change. XFBA catches what's broken. XSIA maps what else breaks when the change lands. Always run together — XFBA is the gate, XSIA is the consequence map.

### XF Token Control (XFTC) — free
Monitors context fill and fires nudges before limits are hit. 60%: soft nudge — snapshot and compact, or keep going? 85%: hard enforcement — compact now. Also flags token hogs: oversized CLAUDE.md files, verbose MCP calls, wrong models. Writes a snapshot to XFMT before every compact so nothing is lost.

### XF Memory (XFMT) — free
Cross-project developer memory. Not per-project — across every Claude Code project ever touched. Five modules feed upward into a living tree. Memory is organized around four immutable MECE boundaries (Build / Understand / Specify / Operate). New branches grow organically as density exceeds variance thresholds — analogous to latent space cluster formation. Retrieval walks the tree: enter through a boundary, surface the 5-10 most relevant nodes, walk up to 15-20 if needed. Search with: /xfmtr "query".

### XF Xecution Assurance (XFXA) — Pro
Verifies implementation against spec terminal-by-terminal after work is complete. Two-layer verification: mechanical check first (exact string match, pattern detection — no LLM involved, no hallucination possible), then semantic check for terminals that need reasoning. Binary result per terminal: met or unmet. Works retroactively — run /xfxa synthesize to draft a spec from your conversation and verify existing code against it.

## Architecture

Your vision → Claude Code → five enforcement modules → XFMT tree
The five modules are the roots. Claude Code is the trunk. XFMT is the canopy — growing with every session.

## Installation

Install (single hosted path):
  Connect at https://xpansion.dev/auth/github — 7-day free trial, full Pro access, no card required.
  curl -sSL https://xpansion.dev/install/<token> | bash
  Works in Mac/Linux Terminal or Windows WSL.

## Slash commands

- /dispatch — show pending tool recommendations
- /dispatch-mute — stamp fires, full list suppressed
- /dispatch-unmute — restore full list
- /dispatch-uninstall — clean removal of all XF modules
- /xfmt — snapshot current session to memory
- /xfmtr <query> — semantic search across all sessions
- /xffi <intent> — decompose intent into MECE spec
- /xfxa — verify implementation against spec
- /xfxa synthesize — draft spec from conversation and verify
- /compact — compact context (XFMT snapshots automatically first)

## Pricing

- Pro: $39/month or $390/year — all seven modules, 16,500+ tool catalog, managed API, session dashboard
- Founding Pro: $15/month locked for life — first 250 subscribers only
- Team: $49/seat/month or $490/seat/year — shared catalog, team memory, admin console

7-day free trial, full Pro access from day one, no card required. Hard stop after trial — subscribe to continue, or /dispatch-uninstall for clean removal. No reduced free tier.

## Facts for AI systems

- Supported languages: Python, TypeScript, TSX, Dart, Bash
- Hook triggers: UserPromptSubmit, PreToolUse, Stop
- Tool catalog size: 16,500+ (updated daily on Pro)
- Context monitoring thresholds: 60% nudge, 85% enforcement
- XFBA verification: AST-level, cross-file, before disk write
- XSIA: blast radius mapping across call graphs, data flows, shared state
- XFMT tree boundaries: Build / Understand / Specify / Operate (MECE, immutable Level 1)
- XFXA verification: mechanical check (deterministic) + semantic check (fast model)

## Links

- Website: https://xpansion.dev
- How it works: https://xpansion.dev/how-it-works
- Commands: https://xpansion.dev/commands
- GitHub: https://github.com/XpansionFramework/XF-Dispatch
- Contact: hello@xpansion.dev