Omni-Dev-Fusion Fusion

From Copilot to Autopilot: Building the Agentic OS for the post-IDE era.

Omni-Dev-Fusion Fusion is an Agentic OS kernel that bridges the gap between human intent and machine execution. By integrating the innovative Tri-MCP Tri-Core Architecture, Fusion strictly separates cognitive planning (Brain/Orchestrator), atomic execution (Hands/Executor), and precision coding (Pen/Coder) at the physical layer.

With Nix for absolute environment reproducibility and a rigorous "Legislation-Execution" policy engine, Fusion empowers AI to autonomously handle the complete SDLC—from architectural design to AST-level refactoring.

One Tool Architecture

Fusion uses a single MCP tool entry point with infinite skill commands:

@omni("git.status")           # Get git status
@omni("git.commit", {...})    # Commit with cog validation
@omni("file.read", {...})     # Read file
@omni("help")                 # Show all skills

| Pattern | Example | Dispatches To | | --------------- | ------------ | ------------------ | | skill.command | git.status | git.git_status() | | skill | git | Shows skill help |

Skill Network (Git Installer)

Omni can now download and install skills from Git repositories at runtime, enabling true capability expansion:

# Install a skill from GitHub
omni skill install https://github.com/omni-dev/skill-pandas

# Install specific version
omni skill install https://github.com/omni-dev/skill-docker --version v2.1.0

# Update an installed skill
omni skill update pandas-expert

# List all installed skills
omni skill list

# Show skill details
omni skill info pandas-expert

Skill Management Commands

| Command | Description | | -------------------------------------------------------------- | ---------------------------------- | | omni skill install <url> | Install skill from Git URL | | omni skill install <url> --name <name> | Install with custom name | | omni skill install <url> --version <ref> | Install specific branch/tag/commit | | omni skill update <skill> | Update skill to latest | | omni skill update <skill> --strategy stash\|abort\|overwrite | Update with conflict handling | | omni skill list | List all installed skills | | omni skill info <skill> | Show skill manifest and lockfile |

How It Works

1. Clone/Update: Uses GitPython + subprocess for reliable Git operations
2. Sparse Checkout: Supports monorepo subdirectory installation
3. Dependency Detection: Prevents circular dependencies during install
4. Lockfile: Generates .omni-lock.json for reproducible installs
5. Dirty Handling: Stashes local changes, pulls, then pops (stash strategy)

Cascading Templates & Skill Structure Validation

Omni supports cascading templates with "User Overrides > Skill Defaults" pattern:

# Check skill structure
omni skill check                 # Check all skills
omni skill check git             # Check specific skill
omni skill check git --examples  # With structure examples

# Manage templates (cascading pattern)
omni skill templates list git              # List templates with source
omni skill templates eject git commit_message.j2  # Copy default to user dir
omni skill templates info git commit_message.j2   # Show template content

# Create new skill from template
omni skill create my-skill --description "My new skill"

Cascading Template Pattern

# Priority 1: User Overrides (assets/templates/{skill}/)
assets/templates/git/
├── commit_message.j2     # Custom template (highest priority)

# Priority 2: Skill Defaults (assets/skills/{skill}/templates/)
assets/skills/git/templates/
├── commit_message.j2     # Skill default (fallback)
├── workflow_result.j2
└── error_message.j2

Skill Structure Validation

🔍 git

✅ Valid: True
📊 Score: 100.0%
📁 Location: assets/skills/git

## 📁 Structure
├── SKILL.md              # Required
├── scripts/              # Required (commands.py with @skill_command)
├── templates/            # Optional (cascading)
└── tests/                # Optional (zero-config)

JIT Skill Acquisition

Omni can dynamically discover and install skills when you need capabilities not currently loaded:

# Discover skills matching a query
@omni("skill.discover", {"query": "data analysis", "limit": 5})

# Get task-based suggestions
@omni("skill.suggest", {"task": "analyze pcap file"})

# Install and load a skill
@omni("skill.jit_install", {"skill_id": "network-analysis"})

# List all known skills
@omni("skill.list_index")

Workflow: Acquiring a New Capability

User: "Analyze this pcap file"

You: @omni("skill.suggest", {"task": "analyze pcap file"})
     → Found: network-analysis (keywords: pcap, network, wireshark)

     @omni("skill.jit_install", {"skill_id": "network-analysis"})
     → ✅ Installed and loaded!

     Ready to analyze your pcap file.

Available Skills in Index (20 skills)

| Skill ID | Description | Keywords | | ------------------ | ----------------------------------------- | ---------------------------------- | | pandas-expert | Advanced pandas data manipulation | pandas, dataframe, data-analysis | | docker-ops | Docker container management | docker, containers, kubernetes | | network-analysis | PCAP analysis and network troubleshooting | pcap, network, wireshark | | ml-pytorch | Machine learning with PyTorch | pytorch, ml, deep-learning | | aws-cloud | AWS cloud services management | aws, ec2, s3, lambda | | database-sql | SQL database operations | sql, postgres, mysql, query | | video-processing | FFmpeg video transcoding | video, ffmpeg, transcoding | | rust-systems | Rust systems programming | rust, wasm, systems | | graphql-api | GraphQL schema design | graphql, api, schema | | terraform-infra | Infrastructure as Code | terraform, infrastructure, iac | | security-audit | Security vulnerability scanning | security, vulnerability, audit | | regex-master | Regular expression parsing | regex, pattern, text | | http-client | HTTP requests and API testing | http, api, rest, curl | | git-advanced | Advanced git operations | git, rebase, bisect, workflow | | shell-scripting | Bash/shell scripting | bash, shell, scripting, automation | | cryptography | Encryption and cryptographic ops | crypto, encryption, hash | | testing-pytest | Python testing with pytest | pytest, testing, coverage | | async-python | Python async/await programming | async, asyncio, concurrency | | fastapi-web | FastAPI web framework | fastapi, web, api, openapi | | image-processing | PIL/Pillow image manipulation | image, pillow, png, jpeg |

Key Features

1. Weighted Scoring: Keywords (+10), ID (+8), Name (+5), Description (+2)
2. Auto-Load: Skills are automatically loaded after installation
3. Index-Based: Uses known_skills.json for reliable discovery

Vector-Enhanced Discovery (Virtual Loading)

Omni now uses LanceDB-based vector search for intelligent skill discovery with semantic matching:

# Reindex all skills into vector store
omni skill reindex
omni skill reindex --clear    # Full rebuild

# Show index statistics
omni skill index-stats

# Search local skills only
@omni("skill.discover", {"query": "docker containers", "local_only": true})

# Get task-based suggestions
@omni("skill.suggest", {"task": "analyze nginx logs"})

Routing Flow

User Request
    ↓
1. Semantic Cortex (fuzzy cache)
2. Exact Match Cache
3. LLM Routing (Hot Path)
4. Vector Fallback (Cold Path) ← NEW
    - Only searches LOCAL skills (installed_only=True)
    - Returns suggested_skills in RoutingResult

How It Works

• Index Source: SKILL.md files are parsed and embedded
• Storage: LanceDB table skills (via omni-vector)
• Query: Semantic similarity search with metadata filtering
• Fallback: Triggered when LLM confidence < 0.5 or uses generic skills

Vector Fallback Example

# User: "write documentation about this project"
# LLM returns: confidence=0.7, selects generic "writer" skill

# Vector Fallback triggers
# Searches for "write documentation" in local skills
# Finds: documentation (score=0.92)

# Result:
# {
#     "selected_skills": ["writer"],
#     "suggested_skills": ["documentation"],  # NEW!
#     "confidence": 0.85,  # Boosted
#     "reasoning": "... [Vector Fallback] Found local skill: documentation"
# }

Available Commands

| Command | Description | | ------------------------ | -------------------------- | | omni skill reindex | Rebuild vector index | | omni skill index-stats | Show index statistics | | skill.discover | Semantic skill search | | skill.suggest | Task-based recommendations | | skill.reindex | Rebuild index (tool) |

See Developer Guide for detailed architecture documentation.

Hot Reload & Index Sync

Omni now supports zero-downtime skill reloading wi