v2026.2.3

Previous release

v2026.2.3 - "Deep Context" Release

Released: February 9, 2026

This release adds the RLM addon for recursive context decomposition, a daemon subsystem for background automation, bidirectional messaging integration, and first-class CLI addon support. Together these features extend AIWG's reach from interactive coding sessions into continuous background operation and multi-platform chat interfaces.

Highlights

What ChangedWhy You Care
RLM addonProcess 10M+ tokens through recursive sub-agent decomposition
Daemon modeBackground file watching, cron scheduling, IPC, tmux management
Messaging subsystemBidirectional Slack, Discord, and Telegram bot integration
CLI addon support`aiwg use rlm` — addons are first-class CLI targets
Copilot RLM artifactsRLM agents, skills, and rules deploy to GitHub Copilot

RLM Addon — Recursive Language Model Processing

The RLM addon implements recursive context decomposition based on REF-089 (Zhang et al., 2026). When a task exceeds comfortable context window limits — scanning hundreds of files, analyzing entire codebases, or batch-processing large directories — RLM decomposes the work into manageable chunks processed by focused sub-agents, then aggregates results.

Architecture

User Request (e.g., "find all TODO comments across the codebase")
    │
    ├── rlm-orchestrator: decomposes into chunks
    │   ├── rlm-chunk-processor: processes src/auth/**
    │   ├── rlm-chunk-processor: processes src/api/**
    │   ├── rlm-chunk-processor: processes src/models/**
    │   └── rlm-chunk-processor: processes src/utils/**
    │
    ├── rlm-aggregator: combines chunk results
    └── rlm-quality-validator: validates completeness

Usage

# Deploy the addon
aiwg use rlm

# Query across large file sets
/rlm-query "src/**/*.ts" "Extract all exported interfaces" --model haiku

# Batch process files in parallel
/rlm-batch "src/components/*.tsx" "Add TypeScript types" --max-parallel 4

# Check processing status
/rlm-status

What's Included

TypeCountExamples
Agents4rlm-orchestrator, rlm-chunk-processor, rlm-aggregator, rlm-quality-validator
Commands3/rlm-query, /rlm-batch, /rlm-status
Skills1rlm-mode (detects large-scale operations)
Rules2rlm-context-management, rlm-subagent-scoping
Schemas5rlm-config, rlm-chunk, rlm-result, rlm-cost, rlm-manifest
Docs2README, rlm-patterns

GitHub Copilot Integration

RLM artifacts automatically deploy to GitHub Copilot:

  • `.github/agents/rlm-agent.yaml`, `rlm-batch.yaml`, `rlm-query.yaml`, `rlm-status.yaml`
  • `.github/skills/rlm-mode/SKILL.md`
  • `.github/copilot-rules/rlm-context-management.md`

Daemon Mode

The daemon subsystem enables AIWG to run as a background process, watching files for changes and executing scheduled tasks without user interaction.

Components

ComponentPurpose
`daemon-main`Core daemon lifecycle and process management
`ipc-server` / `ipc-client`Inter-process communication between daemon and CLI
`agent-supervisor`Manages long-running agent processes
`task-store`Persistent task queue with priority scheduling
`repl-chat`Interactive REPL for daemon sessions
`tmux-manager`Terminal multiplexing for parallel sessions
`automation-engine`Event-driven workflow triggers

Documentation

Full guide: docs/daemon-guide.md

Messaging Subsystem

Bidirectional chat integration enables AIWG agents to communicate through Slack, Discord, and Telegram.

Components

ComponentPurpose
`chat-handler`Bidirectional message routing between adapters and agents
`base adapter`Unified interface for all messaging platforms
`telegram adapter`Telegram Bot API integration
Slack adapterSlack Events API integration
Discord adapterDiscord.js bot integration
`types`Structured message type system
Hub wiringRoutes messages between adapters and agents

Documentation

Full guide: docs/messaging-guide.md

CLI Addon Support

Addons are now first-class targets in the AIWG CLI, alongside frameworks.

# Before (only frameworks)
aiwg use sdlc
aiwg use marketing

# Now (frameworks AND addons)
aiwg use sdlc
aiwg use marketing
aiwg use rlm              # Deploy RLM addon

The `use` handler auto-detects whether a target is a framework or addon and deploys accordingly. Error messages have been updated to reflect addon support.

Install / Upgrade

npm install -g [email protected]