AIWG Framework

AI coding that survives multi-week projects.

Persistent semantic memory, multi-agent ensemble review, and autonomous Agent Loop — deployed to the AI tools you already use.

Type help below for commands. MIT licensed.

cognitive architecture — agents, skills, rules, semantic memory, multi-agent ensembles, traceability

orchestration runtime — Agent Loop, daemon, Mission Control, aiwg serve with HITL drawer

isolation runtime — optional VM-grade per-agent isolation (/sandbox)

Write once. Deploy to Claude Code, Cursor, Copilot, Factory AI, Codex, Hermes, OpenCode, OpenClaw, Warp, or Windsurf with the same artifacts.

Identical-form portability — same agents, same memory, same gates, across every AI tool you use.

Claude Code — full orchestration with autonomous workflows

Factory AI — Tier 1 headless droid exec, native MCP/tasks/missions

Cursor — inline agent assistance and rule enforcement

GitHub Copilot — contextual agents with workspace instructions

OpenAI Codex — autonomous agent orchestration

Hermes — first-class deployment target, token-optimized templates

OpenCode — extensible agents with tool and hook integration

OpenClaw — native behaviors support, 10th platform

Warp Terminal — terminal-native context and command workflows

Windsurf — cascade agents with rule-driven guidance

sdlc-complete — full lifecycle: requirements, architecture, implementation, testing, deployment, hypercare

media-marketing-kit — campaign, content, brand, analytics

voice — SOUL.md identity, brand voice consistency, content blending

ops-complete — sysops, devops, itops, stream extensions; 9 templates from real ops patterns

forensics-complete — digital forensics, IR, IOC extraction, chain of custody

research-complete — FAIR/PROV provenance, GRADE quality, citation guard

Plus 21 addons covering daemon, Mission Control, RLM, behaviors, agentic-installer, prose-integration, and more.

188 agents · 128 skills · 50 CLI commands · 100+ artifact templates

Agent Loop / Ralph — closed-loop self-correction; external Ralph survives crashes for 6-8h+ autonomous runs

Daemon mode — persistent supervisor with web UI, scheduled tasks, multi-room messaging, cross-session memory

Mission Control — parallel agent loops as background missions with live dashboard

aiwg serve — PTY orchestrator + HITL drawer; supervise and approve agent actions from a browser

RLM — recursive context decomposition for repos that don't fit in any one window

Path A — Project-local: drop a bundle into .aiwg/extensions/, addons/, frameworks/, or plugins/. Auto-discovered, no fork.

Path B — Fork: customize AIWG core, dev mode, contribute back upstream.

Path C — Corpus: cross-project sharing without going public.

Identical-form portability — bundle layout is byte-identical across project-local, fork, and upstream. aiwg promote is a hash-verified copy with zero rewrite.

Run Claude Code in its own VM, for hours, on your hardware.

QEMU/KVM or Docker per-agent isolation, virtiofs shared storage, telemetry, reconciliation. For regulated industries, on-prem requirements, and 6–8h+ autonomous coding runs.

> /sandbox

# CLI + multi-platform

Requires Node.js 18+. Works with all providers.

# Claude Code plugin

No npm required. Also: marketing@aiwg, voice@aiwg, ops@aiwg, utils@aiwg

# Install framework to your project

# Start an interactive workflow in Claude Code

# Provide guidance to tailor the output

+ aiwg regenerate CLI command for fast context-only doc updates

+ aiwg-issue and aiwg-pr kernel skills automate high-quality filing

+ aiwg steward capabilities now work end-to-end for all users

+ Config-driven release flow via .aiwg/release.config executable spec

+ Fixed path resolution and schema mismatch in aiwg steward handler

+ aiwg-doctor gate catches discovery-surface degradation early

Type changelog or releases below for full history.