Vmem

Cross-platform process memory manipulation library written in V. Target feature surface: the union of Cheat Engine (desktop) and GameGuardian (Android) core memory tooling — scanning, editing, freezing, pointer tools, code patching, breakpoints, speedhack — as an importable library with CLI and GUI front ends.

Status: v0.1.0 implemented — P0–P8 core landed with tests (see CHANGELOG.md).

Targets

Target	Tier	Notes
Windows x64	1	ReadProcessMemory/VirtualQueryEx family
Linux x64/arm64	1	/proc + process_vm_* + ptrace
Android arm64/arm/x86_64	1	Frida-style: host drives an on-device agent over ADB/TCP (root daemon, gadget, no-root paths); scans run agent-side
macOS arm64/x64	2	task_for_pid gated by SIP/entitlement; best effort
FreeBSD/OpenBSD	3	falls out of the Linux /proc-less backend where feasible

Repository layout

Path	Contents
vmem/	the library module (import vmem)
vmem/sys/	local OS shims, the only place OS APIs are called
vmem/sys/remote/	device-agent transport client (frida-style, over ADB/TCP)
agent/	on-device agent daemon (static Linux ELF, runs on Android like frida-server)
cli/	terminal front end (scan/freeze/maps/read/write)
gui/	desktop GUI (V UI widgets: buttons, dropdowns, listbox)
gui-radar/	gg-based radar visualizer (donut, sparklines)
tests/	integration tests + deterministic victim target process
examples/	runnable recipes
docs/	plan, architecture, platform API, feature parity, GUI design
mcp/	zero-dependency docs/conventions MCP server (server.mjs) + registry.json + sync-skills.mjs
.omp/ .claude/ .codex/	per-agent MCP registration and skill copies (generated)

Agent integration (OMP / Claude Code / Codex)

• Skill: canonical SKILL.md (repo root) is synced into .omp/skills/vmem/, .claude/skills/vmem/, .codex/skills/vmem/ (and ~/.claude/skills/vmem/, ~/.codex/skills/vmem/ for use from any project) via node mcp/sync-skills.mjs. After editing SKILL.md, run the sync (--check verifies drift).
• MCP: mcp/server.mjs (Node, zero deps) serves the docs as query tools (repo_context, search_docs, read_doc, get_rules, get_recipe, get_vref, get_phase, get_platform, get_parity, check_structure, plan_files) plus doc resources and a feature-scaffold prompt. Registered per agent: .omp/mcp.json (OMP), .mcp.json (Claude Code / portable), .codex/config.toml (Codex). Server script paths are machine-absolute — adjust if the repo moves.
• Context: AGENTS.md is the normative rules file (Codex + OMP + agents-md read it directly); CLAUDE.md points Claude Code to it.

Documentation

• docs/PLAN.md — scope, workstreams, phases, risks
• docs/ARCHITECTURE.md — module map, file rules, threading
• docs/PLATFORM-API.md — per-OS capabilities and permission models
• docs/FEATURE-PARITY.md — Cheat Engine / GameGuardian matrix
• docs/GUI.md — GUI concept and layout
• AGENTS.md — contributor/agent operating rules
• SKILL.md — V language + memory-tooling domain reference
• CHANGELOG.md — memory changelog (every memory-affecting change)

Intended use

Memory inspection and modification of processes you own or are explicitly authorized to test (your own games in single-player/offline contexts, CTF binaries, your own software, lab VMs). Not aimed at multiplayer cheating or evading third-party anti-cheat; kernel drivers and anti-anti-cheat work are out of scope by design.