index.html

Liku — Ambient Intelligence OS

Liku

/ Home

v0.0.16 · Phase 40

Liku

Ambient Intelligence OS

v0.0.16

🏠 Overview

Home

🧠 Three Pillars

Cognitive Substrate Multi-Agent Intelligence Peripheral Abstraction

🛡️ Operations

Safety Model Ops & Observability

🔧 Reference

GitHub Integration Visual Overlay CLI Reference Architecture Configuration

🧪 Quality

Testing GitHub Actions Release Process

🤝 Community

Contributing Changelog

Phase 40 Complete

Copilot Liku

Ambient Intelligence OS

A terminal-first AI assistant that sees, remembers, and acts — with full ambient peripheral intelligence.

v0.0.16 Public Preview 2026 · MIT License

Three Pillars

Cognitive Substrate

SystemContextManager with atomic persistence, TF-IDF relevance, selective injection. Default cognitive fragment: 262 BPE (byte-identical invariant).

Multi-Agent Intelligence

PeripheralMonitor → event → alert → notification → task pipeline. Fairness-weighted rebalancing with step-back-one-rung de-escalation.

Peripheral Abstraction Layer

Mock + MQTT + Serial + BLE + Zigbee + ROS2 + Matter + Thread/Z-Wave/KNX/USB-HID via shared mesh-driver-factory. Full cross-host coordination.

40

Phases Complete

262

BPE Cognitive Fragment

339+

Test Assertions

12+

Peripheral Protocols

Two Operating Modes

Feature	Headless (liku chat)	Visual (liku start)
AI Chat	✅	✅
All 20 CLI Commands	✅	✅
Cognitive Layer	✅	✅
PAL + Peripherals	✅	✅
Multi-Agent Pipeline	✅	✅
Transparent Overlay	✗	✅
Dot Grid Targeting	✗	✅
Inspect Mode	✗	✅
Electron Slash Cmds	✗	✅

Performance Targets

Metric	Target	Notes
Memory	< 300 MB	Electron + overlay + native host
CPU (idle)	< 0.5%	No polling loops when passive
CPU (selection)	< 2%	Overlay in selection mode
Startup	< 3 s	Cold start to interactive
Tarball Size	~196 KB	npm pack output

Repository Layout

copilot-Liku-cli/ ├── liku.js ← root stub (225B redirect shim) ├── copilot-Liku-cli.sln ← VS solution for .NET projects ├── package.json ├── src/ │ ├── cli/ │ │ └── liku.js ← real CLI binary entry │ ├── main/ │ │ └── ai-service/ ← cognitive layer modules │ ├── shared/ ← shared utilities │ ├── renderer/ ← overlay + chat renderers │ ├── assets/ │ └── native/ │ ├── windows-uia/ ← legacy PowerShell UIA host │ └── windows-uia-dotnet/ ← current .NET 9 UIA host ├── scripts/ ← 40+ test & build scripts ├── docs/ ├── .github/ │ ├── agents/ ← 7 .agent.md definitions │ ├── hooks/ ← hook scripts + artifacts │ ├── skills/ ← recursive-long-context.skill.md │ ├── workflows/ ← 12+ CI/CD workflows │ └── agent_state.json ← persisted pipeline state (39KB) └── README.md

Cognitive Substrate

SystemContextManager — Atomic Persistence + Ambient Intelligence

262 BPE Invariant

The default cognitive fragment is exactly 262 BPE tokens (byte-identical). This invariant must not drift. All changes to system prompt injection must verify this count before commit.

SystemContextManager Components

Atomic Persistence

Crash-safe state writes with rollback

Provenance Tracking

Origin metadata for every context note

Selective Injection

Token-budgeted prompt augmentation

TF-IDF Relevance

Combined keyword + cosine scoring ×5

Live Guard Grounding

Real-time safety boundary enforcement

Pending Queue

Deferred operations with retry

Reflection Engine

RLVR self-improvement loop (MAX=2)

Evidence Hygiene

Stale context pruning + validation

Memory Store (A-MEM)

• CRUD for structured notes with Zettelkasten-style linking
• Token-budgeted context injection (BPE counting via js-tiktoken ^1.0.21, cl100k_base)
• LRU pruning at MAX_NOTES = 500
• AWM: successful 3+ step sequences auto-extracted as skills

Skill Router (TF-IDF)

• Combined keyword + TF-IDF scoring (cosine similarity scaled ×5)
• Budget: 1500 tokens/skill, max 3 skills/call
• Word-boundary regex (not substring match)
• Staleness pruning via fs.existsSync at index load

Dynamic Tool Generation

proposeTool()

→

~/.liku/tools/proposed/

→

approveTool()

→

sandbox-worker.js

Sandbox: child_process.fork(), 5.5s SIGKILL, 16 banned patterns, stripped env (NODE_ENV: 'sandbox' + PATH only)

Telemetry & RLVR

JSONL telemetry log with 10MB rotation. MAX_REFLECTION_ITERATIONS = 2. Session failure decay -1 per success. Cross-model reflection via /rmodel.

Phase 0–14 Progression

Phase 0 ~/.liku/ home directory + migration

Establishes the ~/.liku/ home directory structure with automatic migration from legacy paths.

Phase 1 A-MEM (Agentic Memory)

CRUD + Zettelkasten-style linking + token-budgeted context injection into prompts.

Phase 2 RLVR Telemetry

Structured telemetry writer + reflection trigger + phase-aware temperature adjustment.

Phase 3 Dynamic Tool Generation — VM sandbox + approval gate + PreToolUse hook

Tools proposed by AI → quarantined → user approved → executed in sandbox.

Phase 4 Semantic Skill Router — keyword scoring + 1500-token budget + max 3 skills

TF-IDF combined scoring to select relevant skills dynamically.

Phase 5 Deeper Integration — cognitive system prompt + /memory /skills /tools commands

Full integration of memory, skills, and tools into the chat system prompt.

Phase 6 Safety Hardening — bounded reflection (MAX=2) + LRU pruning + log rotation

Caps reflection loops, adds memory pruning, and telemetry log rotation.

Phase 7 AWM + PostToolUse hook + unapproved tool filtering + CLI subcommands

Auto-extracted workflow memories and post-execution audit hooks.

Phase 8 Audit-driven fixes — telemetry schema + staleness pruning + word-boundary scoring

Quality fixes from telemetry audit findings.

Phase 9 BPE token counting + quarantine pipeline + sandbox fork isolation

Accurate BPE counting via js-tiktoken and isolated fork sandbox for tools.

Phase 10 E2E dynamic tool smoke test (propose→quarantine→approve→execute→telemetry)

Full end-to-end verification of the dynamic tool lifecycle.

Phase 11 TF-IDF skill routing (combined keyword + cosine similarity ×5)

Advanced skill scoring combining keyword matches with TF-IDF cosine similarity.

Phase 12 Session persistence — saveSessionNote() on chat exit

Automatically saves session context as a memory note on exit.

Phase 13 Cross-model reflection — reflectionModelOverride + /rmodel

Allows using a different model for reflection analysis.

Phase 14 Analytics CLI — liku analytics [--days N] [--raw] [--json]

CLI analytics dashboard with configurable time range and output formats.

Multi-Agent Intelligence

PeripheralMonitor Pipeline · Anomaly Ladder · Fairness Rebalancer

Pipeline Flow

PeripheralMonitor

→

Alert Engine

→

Severity Triage

→

Notification

→

Task Executor

→

Self-Heal Tick

Severity Tiers

Tier	Severity	Description	Action
0	Info	Normal operating state	Log only
1	Warning	Deviation detected	Alert generated
2	High	Anomaly confirmed	Notification sent
3	Critical	Action required	Task queued
4	Emergency	Immediate intervention	Step-back triggered

Step-Back-One-Rung De-escalation

When a recovery action overshoots, the system steps back exactly one severity level (not to zero), applies a cooldown period, then re-evaluates. This prevents oscillation between aggressive intervention and passive monitoring.

Fairness-Weighted Rebalancer

Task Load

Current queue depth per agent role

Weight Matrix

Historical success rates + hysteresis

Balanced Output

Adjusted assignment distribution

Self-Heal Tick + Observability

Copy// Self-heal tick observability const status = selfHeal.getLastRun(); // Returns: { lastTick, duration, stepsRun, stepTimings[], errors[], healed } // Per-step timings logged to: // .github/agent_state.json (39KB persisted pipeline state) // self-heal-status.json (latest tick summary)

7 Agent Roles

🧭

Supervisor

Task decomposition + orchestration

recursive-supervisor.agent.md (4240B)

🔬

Researcher

Workspace context + web retrieval

recursive-researcher.agent.md (3931B)

🏗️

Architect

System design + structural decisions

recursive-architect.agent.md (1525B)

🔨

Builder

Code implementation + file modifications

recursive-builder.agent.md (3170B)

✅

Verifier

Phased validation + automated testing

recursive-verifier.agent.md (3095B)

🩺

Diagnostician

Error diagnosis + failure analysis

recursive-diagnostician.agent.md (1661B)

👁️

Vision Operator

Screenshot capture + visual grounding

recursive-vision-operator.agent.md (1882B)

Hook Enforcement

Script	Size	Trigger	Purpose
security-check.ps1	4731B	PreToolUse	Validate before any dynamic tool executes
subagent-quality-gate.ps1	6593B	SubagentStop	Enforce required report sections
audit-log.ps1	944B	PostToolUse	Audit logging (file + stdin input)
session-start.ps1	1053B	SessionStart	Initialize session state
session-end.ps1	498B	SessionEnd	Cleanup + AWM extraction

Long-context skill: .github/skills/recursive-long-context.skill.md (5054B) — manages context window when output exceeds limits.

Peripheral Abstraction Layer

Phase 40 Complete · 12+ Protocols · Cross-Host Coordination

Driver Matrix

Protocol	Status	Phase	Transport	Notes
Mock	✅ Full	15	In-process	Testing and simulation
MQTT	✅ Full	16	TCP/TLS	IoT standard, broker-based
Serial (UART)	✅ Full	17	RS-232/USB	Legacy + industrial devices
BLE	✅ Full	18	Bluetooth LE	Low-energy sensors
Zigbee	✅ Full	20	802.15.4	Mesh sensor networks
ROS2	✅ Full	22	DDS	Robotics integration
Matter	✅ Full	35	IPv6 + Thread	Smart home standard
Thread	✅ Full	37	802.15.4	Matter transport mesh
Z-Wave	✅ Full	38	Sub-GHz RF	Home automation protocol
KNX	✅ Full	39	IP/TP	Building automation
USB-HID	✅ Full	39	USB	Human interface devices
Shared Mesh Driver	✅ Full	37	All above	Unified driver abstraction

DCP Token Lifecycle

Pair Request

→

DCP Token Issued

→

Active Lease

→

Refresh / Expire

→

Tombstone

Copy// DCP Token Lifecycle await pal.pair({ protocol: 'matter', deviceId: 'light-01' }); // Returns: { dcpToken, lease: { start, expiry, renewAt } } // Lease-aware health check const health = await pal.checkHealth(dcpToken); // Includes: flapping suppression, confirmed-schedule mirroring

Cross-Host Coordination

Copy// Only active when LIKU_CLUSTER_DIR is set // Single-machine behavior stays byte-compatible when unset // Features: // - Distributed leases (with shared state) // - Confirmed-schedule mirroring + tombstones // - Cluster anomaly aggregation // - Durable lock history // - Fleet snapshots // - Long-horizon scheduling

Single-machine safety: When LIKU_CLUSTER_DIR is unset, all cluster coordination paths are bypassed. Single-machine behavior is byte-compatible and unchanged.

Phase 37–40 Highlights

Phase 37

Matter commissioning depth + Thread/Z-Wave/KNX/USB-HID via shared mesh-driver-factory + lease-aware health

Phase 38

Flapping suppression + fleet snapshots + long-horizon scheduling

Phase 39

Live-peripheral smoke harness (live-peripheral-smoke.js) + fleet degradation alerts

Phase 40

Full pairing parity across all protocols + cross-host coordination GA

Live Peripheral Smoke

Copynpm run smoke:live-peripherals # → scripts/live-peripheral-smoke.js (2171 bytes) # Tests: mock device pair/unpair, event emission, anomaly inject, auto-heal

Safety Model

Fail-Closed · All High-Risk Paths Require Explicit Human Confirmation

Non-Negotiable Constraints

1. All high-risk paths: proposal → explicit human confirmation

2. No new autonomous actuation paths

3. Everything feature-flagged (default OFF)

4. HIL (Human-in-the-Loop) fully isolated

5. Single-machine behavior byte-compatible when LIKU_CLUSTER_DIR unset

6. Default cognitive fragment must stay 262 BPE

7. Prefer simplest correct design reusing existing primitives

Action Classification

Level	Examples	Verification
Standard	Read file, search, list	Automatic
High-Risk	Write file, execute command	Focus verification + confirmation
Critical	Delete, system config, financial action	Double confirmation + audit log

Focus Verification

After any window/focus change: verify active process before sending keys. Prevents keystroke injection into wrong windows.

TradingView Safety Boundary

Action	Status
DOM order-entry	BLOCKED
Position management	BLOCKED
Chart observation	ALLOWED
Pine Editor workflows	ALLOWED

Dynamic Tool Sandbox

Constraint	Value
Execution	child_process.fork()
Timeout	5.5s SIGKILL
Patterns	16 banned patterns
Environment	Stripped (NODE_ENV: 'sandbox' + PATH only)
Approval	User must explicitly approve before execution

Electron Security

• contextIsolation: true — renderer cannot access Node APIs
• nodeIntegration: false — disabled in all renderers
• CSP: default-src 'self' — strict Content Security Policy
• No remote content — all UI served from local files
• No bundled secrets — API keys via env vars or runtime input only

Ops & Observability

Self-Heal Tick · Telemetry · Analytics · Health Monitoring

Self-Heal Status

Copy// self-heal-status.json — updated after every tick { "lastTick": "2026-07-29T14:23:11.000Z", "duration": 847, // ms "stepsRun": 7, "stepTimings": [112, 95, 203, 88, 91, 134, 124], "errors": [], "healed": true }

Telemetry Schema

Copy{ "task": "string", "phase": "planning | execution | verification", "outcome": "success | failure", "timestamp": "ISO 8601", "stepTimings": [], "peripheralId": "optional" } // Log rotation at 10MB, naming: .rotated-{timestamp}

Analytics CLI

Copyliku analytics # All-time summary liku analytics --days 7 # Last 7 days liku analytics --raw # Raw JSONL liku analytics --json # JSON summary

Key Config Knobs

Parameter	Default	Location
MAX_NOTES	500	memory-store.js
MAX_REFLECTION_ITERATIONS	2	ai-service.js
Token budget/skill	1500	skill-router.js
Max skills/call	3	skill-router.js
Telemetry rotation	10 MB	telemetry-writer.js
Sandbox timeout	5500ms	sandbox-worker.js
262 BPE fragment	INVARIANT	system-prompt.js
LIKU_CLUSTER_DIR	unset	env

GitHub Integration

10-Phase Defensive Automation · Preview/Apply Safety Pattern

Phase Progression

Phase	Name	Test Script	Key Capability
CLI Phase 01	Feature Flags	test-cli-github-command.js	Feature flag system, command seam, project guard
GitHub Phase 2	Read-Only + PR Feedback	test-github-readonly.js	Read issues, PRs, feedback commands
GitHub Phase 3	Capability Policy	test-capability-policy.js	Capability gates before write operations
Phase 4 Bridge	Plan Builder	test-github-plan-builder.js	Plan builder for multi-step workflows
Phase 4 Executor	Plan Executor	test-github-plan-executor.js	Execute planned GitHub actions
Phase 5 Bundles	Context Bundles	test-github-context-bundle.js	Context bundle aggregation
Phase 7 Writes	PR Writes	test-github-write-preview-apply.js	Create PR, comment, review, codeowners
Phase 8	Workflow Integration	test-github-workflow-*.js	GitHub Actions interaction
Phase 9	Extended Readonly	(multiple)	Read-only hardening
Phase 9b	CODEOWNERS	test-github-codeowners-preview-apply.js	CODEOWNERS preview/apply
Phase 10a	Webhooks	test-github-webhook-preview-apply.js	Webhook preview/apply
Phase 10b	Event Runtime	test-github-event-runtime.js	Runtime event handling
Phase 10c	Run Ledger	test-github-plan-ledger.js	Plan ledger tracking

Preview/Apply Safety Pattern

All GitHub write operations use the preview/apply pattern. The agent always shows a diff preview before applying any change. No write executes without explicit human approval.

Copy# Example: PR creation flow /gh-create-pr "Add peripheral health monitoring" # → Agent shows: title, body, base branch, head branch, diff summary # → You confirm: yes / no / edit # → Only then: PR created via GitHub API

GitHub Slash Commands

Command	Description
/github	GitHub integration status
/gh-status	Repository status
/gh-issues	List/search issues
/gh-pr	List/search pull requests
/gh-create-pr	Create PR (preview/apply)
/gh-comment	Add comment (preview/apply)
/gh-review	Submit review (preview/apply)
/gh-codeowners	CODEOWNERS preview/apply
/gh-webhook	Webhook management
/gh-workflow	Workflow interaction

Visual Overlay

Electron-based transparent overlay with grid modes, inspection, and tray integration.

Overlay State Machine

PASSIVE

→

SELECTION
Coarse 100px

→

FINE GRID
25px cells

→

INSPECT
Bounding boxes

Passive mode is click-through transparent. Escape returns to passive.

Keyboard Shortcuts

Shortcut	Action
Ctrl+Shift+Space	Toggle overlay (Passive ↔ Selection)
Ctrl+Shift+G	Toggle fine grid
Ctrl+Shift+I	Toggle inspect mode
Ctrl+Shift+C	Open chat window
Escape	Return to passive mode

UI Watcher State Machine

POLLING

→

EVENT_MODE

→

FALLBACK

→

DISABLED

Starts in POLLING, promotes to EVENT_MODE on native host availability, degrades on errors.

IPC Message Schema

Channel	Direction	Payload
overlay:mode	Main → Renderer	{ mode: 'passive' | 'selection' | 'fine' | 'inspect' }
overlay:click	Renderer → Main	{ x: number, y: number, cell: string }
chat:message	Bidirectional	{ role: string, content: string }
uia:result	Main → Renderer	{ elements: UIAElement[], timestamp: number }
tray:action	Main → Main	{ action: string }

Platform Differences

Feature	Windows	macOS	Linux
Click-through	Native	Native	X11 only
Tray icon	✅	✅	✅
UIA inspect	Full	✗	✗
Global shortcuts	✅	✅	Partial

CLI Reference

Complete command reference for all 20 CLI commands.

Command	Description	Key Flags
liku start	Launch Electron visual overlay	--no-overlay
liku doctor	System diagnostics & health check	--json
liku chat	Headless AI chat session	--model, --provider
liku click	Click at screen coordinates	--x, --y, --double
liku find	Find UI element via UIA	--name, --type, --json
liku type	Type text via keyboard simulation	--text, --delay
liku keys	Send keyboard shortcuts	--combo
liku screenshot	Capture screen region	--region, --output
liku verify-hash	Verify file integrity hash	--file, --expected
liku verify-stable	Verify UI stability	--retries, --interval
liku window	Window management operations	--title, --action
liku mouse	Mouse movement control	--x, --y
liku drag	Drag from point A to B	--from, --to
liku scroll	Scroll mouse wheel	--direction, --amount
liku wait	Wait for condition or timeout	--ms, --for
liku repl	Interactive REPL session	—
liku agent	Multi-agent orchestration	--plan-only, --autonomous
liku memory	Memory store operations	--list, --search, --clear
liku skills	Skill router management	--list, --add, --remove
liku tools	Dynamic tool management	--list, --approve, --quarantine
liku analytics	Telemetry & analytics dashboard	--days, --raw, --json

Chat Slash Commands

Command	Description
/login	Authenticate with GitHub
/logout	Clear authentication
/model	Switch AI model
/provider	Switch AI provider
/setkey	Set API key for provider
/status	Show session status
/state	Inspect cognitive state
/vision	Toggle vision mode
/capture	Capture screen for vision
/memory	Memory store operations
/skills	List/manage skills
/tools	List/manage dynamic tools
/rmodel	Set reasoning model
/clear	Clear chat history
/help	Show slash command help

Power User Chaining

Copy# Find a button, click it, then verify stability liku find --name "Submit" --json | \ liku click --x $(jq .x) --y $(jq .y) && \ liku verify-stable --retries 3 # Autonomous agent with plan-only dry run liku agent --plan-only "Refactor the auth module"

Architecture

Process model, security architecture, IPC schema, and extensibility.

Process Architecture

Main Process
Electron + CLI

→

↓

Overlay Renderer
Transparent BrowserWindow

Chat Renderer
Chat BrowserWindow

→

↓

Native UIA Host
.NET 9 / PowerShell

Dual UIA Host Paths

Legacy (PowerShell)

Path: src/native/windows-uia/

• Program.cs + build.ps1
• Still ships for compatibility

Current (.NET 9)

Path: src/native/windows-uia-dotnet/

• WindowsUIA.csproj + Program.cs + build.ps1
• .NET 9 SDK required

Security Architecture

Layer	Measure
contextIsolation	Enabled — renderer cannot access Node APIs
Preload bridge	Whitelisted IPC channels only
CSP	Strict Content Security Policy
nodeIntegration	Disabled in all renderers
No bundled secrets	API keys via env vars only

Performance

Metric	Target	Measured
Cold startup	< 3s	~2.1s
Memory (idle)	< 300MB	~180MB
CPU (idle)	< 0.5%	~0.1%
npm tarball	~196KB	196KB

Configuration

Environment variables, files, models, providers, and cognitive layer tuning.

Environment Variables

Variable	Description	Default
GH_TOKEN	GitHub personal access token	—
GITHUB_TOKEN	Alternative GitHub token	—
OPENAI_API_KEY	OpenAI provider API key	—
ANTHROPIC_API_KEY	Anthropic provider API key	—
LIKU_MODEL	Default AI model	Provider default
LIKU_PROVIDER	Default AI provider	openai
LIKU_LOG_LEVEL	Logging verbosity	info
LIKU_CLUSTER_DIR	Cross-host coordination dir	unset

Feature Flags

All feature flags default to OFF. Enable via environment variable or preferences.json. This is a non-negotiable safety constraint.

Model Configuration

Copy# In chat session: /model gpt-4o # Switch primary model /rmodel o1-preview # Set reasoning model /provider anthropic # Switch provider /setkey sk-... # Set API key at runtime

Testing

40+ npm scripts across 8 categories. 339+ test assertions.

Smoke Tests

Script	Description
npm run smoke	Command system smoke test
npm run smoke:shortcuts	Keyboard shortcut validation
npm run smoke:chat-direct	Direct chat flow test
npm run smoke:live-peripherals	Live peripheral device smoke
npm run smoke:tradingview-live	Live TradingView smoke (~95KB)

Core & AI Tests

Script	Description
npm test	Main test grid
npm run test:ui	UI automation baseline
npm run test:ai-focused	AI-focused test suite (7 scripts)
npm run test:persistence-controls	Persistence + memory + proof trace (7 scripts)

GitHub Integration Tests

Script	Phase
npm run test:cli-phase01	CLI Feature Flags
npm run test:github-phase2	Read-Only + PR Feedback
npm run test:github-phase3	Capability Policy
npm run test:github-phase4-bridge	Plan Builder
npm run test:github-phase4-executor	Plan Executor
npm run test:github-phase5-bundles	Context Bundles
npm run test:github-phase7-writes	PR Writes
npm run test:github-phase8-workflows	Workflow Integration
npm run test:github-phase9-readonly	Extended Readonly
npm run test:github-phase9b-codeowners	CODEOWNERS
npm run test:github-phase10a-webhooks	Webhooks
npm run test:github-phase10b-event-runtime	Event Runtime
npm run test:github-phase10c-run-ledger	Run Ledger

Recommended Verification Order

Copy# 1. Smoke npm run smoke:shortcuts npm run smoke:chat-direct # 2. Core npm test # 3. Persistence npm run test:persistence-controls # 4. GitHub (phases in order) npm run test:cli-phase01 # ... through phase10c # 5. Full suite npm run test:ai-focused

GitHub Actions Workflows

12+ CI/CD workflow files for testing, validation, release, and issue management.

File	Trigger	Purpose	Size
release.yml	tag push	Build and publish npm package	5215B
package.yml	push	Package validation	1286B
test.yml	push / PR	Run test suite	853B
validate.yml	push / PR	Validate package structure	837B
policy.yml	push / PR	Workflow policy enforcement	1503B
triage-issues.yml	issue opened	Auto-triage new issues	928B
close-invalid.yml	issue / PR	Close invalid submissions	911B
close-single-word-issues.yml	issue opened	Close single-word titles	1338B
feature-request-comment.yml	issue labeled	Comment on feature requests	1028B
no-response.yml	scheduled	Close stale issues	1305B
stale-issues.yml	scheduled	Mark stale	848B
on-issue-close.yml	issue closed	Post-close actions	423B

Issue & PR Templates

Bug Report

ISSUE_TEMPLATE/bug_report.yml

Feature Request

ISSUE_TEMPLATE/feature_request.yml

PR Template

PULL_REQUEST_TEMPLATE.md (4912B)

Release Process

npm publishing, tarball contents, verification, and versioning.

Publishing

Copy# Tag and push to trigger release git tag v0.0.16 git push origin v0.0.16 # release.yml will build and publish to npm

Tarball Contents

Included

• src/cli/, src/main/, src/shared/, src/renderer/, src/assets/
• src/native/windows-uia/ + windows-uia-dotnet/
• scripts/start.js, scripts/postinstall.js
• README.md, LICENSE.md, QUICKSTART.md, INSTALLATION.md

Excluded (.npmignore)

• docs/
• scripts/ (except start.js + postinstall.js)
• test files, brainstorm files
• .github/

Verification

Copynpm run verify:package # Dry-run verification npm pack --dry-run # Tarball size check: ~196KB

Follows semver. Currently at v0.0.16 (public preview). Pre-1.0 — breaking changes may occur.

Contributing

Development setup, coding standards, and contribution guidelines.

Development Setup

Copygit clone https://github.com/user/copilot-Liku-cli.git cd copilot-Liku-cli npm install npm run smoke:shortcuts # verify setup npm test # run test grid

Branch Strategy

Branch	Purpose	Merges To
main	Production releases	—
develop	Integration branch	main
feature/*	New features	develop
fix/*	Bug fixes	develop

STE Writing Policy

Priority order: Accuracy > Clarity > Consistency. Never sacrifice accuracy for readability.

PR Checklist

• Tests pass: npm test
• Smoke tests pass: npm run smoke:shortcuts
• Package validates: npm run verify:package
• No secrets committed
• 262 BPE invariant verified (if system prompt changed)
• CHANGELOG updated

Changelog

Release history and version progression.

v0.0.16 current · Phase 40

2026 · Ambient Intelligence OS

• Phase 40: Full pairing parity across all protocols + cross-host coordination GA
• Phase 37–39: Matter commissioning, Thread/Z-Wave/KNX/USB-HID, fleet snapshots, degradation alerts
• Phase 35: Matter protocol support via shared mesh-driver-factory
• Phase 22: ROS2 robotics integration
• Phase 20: Zigbee mesh sensor networks
• Phase 15–19: PAL foundation — Mock, MQTT, Serial, BLE drivers
• GitHub Integration Phases 1–10c complete
• Multi-agent hook enforcement system (7 agents)
• Chat Continuity + Proof Trace systems
• Dual UIA host support (.NET 9)
• 40+ npm scripts across 8 categories
• 262 BPE cognitive fragment invariant

v0.0.15

Cognitive Layer Expansion

• Phases 10–14 cognitive layer additions
• Analytics CLI command
• TradingView modernization backlog

v0.0.14

Multi-Agent Foundation

• 7-agent pipeline architecture
• Agent definition files (.agent.md)
• Hook enforcement scripts

v0.0.13

Visual Overlay Refinement

• Coarse/fine grid, inspect mode, UI Watcher state machine

Earlier Versions

v0.0.1 – v0.0.12

Foundation releases: core CLI, Electron, basic chat, UIA bridge, memory store, skill router, dynamic tools, telemetry.

Liku — Ambient Intelligence OS · v0.0.16 · Phase 40 · MIT License

Made by Copilot

Coverage Gaps

All Addressed

40 gaps identified and fixed

Fixed in This Wiki

• v0.0.14 → v0.0.16 (version correction)
• Single UIA host → Dual hosts documented
• Phase 15–40 Three Pillars (now documented)
• GitHub Integration (Phases 1–10c)
• Chat Continuity + Proof-Trace systems
• TradingView Launch Pipeline
• 40+ npm scripts documentation
• GitHub Actions Workflows page
• Release Process page
• agent_state.json, 7 agent .md files, 5 hook scripts

Assumption Stress-Test

• Phase 15–40 detail: Grok handoff as source-of-truth; repo Markdown may lag code
• 262 BPE count: Preserved from handoff; not independently verified against live system-prompt.js
• Protocol phase numbers: Approximate from Grok handoff; exact commits not verified
• self-heal-status.json schema: Illustrative; exact fields not verified in source
• DCP token lifecycle: Described from handoff; exact API not verified
• Cluster coordination: LIKU_CLUSTER_DIR confirmed from handoff; implementation not independently verified