Oh-My-Kimi: A Verification-Focused Agent Runtime for Kimi Code

Oh-My-Kimi (OMK) is a verification-focused agent runtime designed specifically for Kimi Code CLI, aiming to address pain points of AI programming assistants such as "premature completion", parallel conflicts, and long-session amnesia. Through mechanisms like evidence gating, DAG task execution, isolated workspaces, and local graph memory, it transforms Kimi Code into a verification-capable bounded coding team, helping developers deliver code in a verifiable manner.

As the code generation capabilities of large language models improve, developers widely use AI programming assistants, but there are the following pain points:

• Premature "completion": Agents claim tasks are done without sufficient verification
• Parallel work conflicts: Multi-threaded operations easily lead to context contamination and file conflicts
• Long-session amnesia: Forgets previous decisions and goals as conversations grow longer
• Lack of visibility: Difficult to understand AI's working status, progress, and bottlenecks
• Disorganized workflow: Lack of unified management for MCP servers, skills, etc. OMK was created to solve these problems.

OMK is a native Kimi control plane with the core philosophy: "Kimi writes the code; OMK coordinates, verifies, remembers, and guards". Its core design includes:

1. Kimi First: Built specifically for Kimi Code, ensuring Kimi is the primary writer and merger
2. Bounded Coding Team: Transforms Kimi into a coding team with clear division of labor, processes, and quality gates
3. Verifiability: Tasks must pass verification via evidence such as file existence, command execution, and code differences before completion; claims of completion based solely on narrative are not accepted.

OMK's workflow is based on DAG (Directed Acyclic Graph): User Prompt → OMK Ingestion → Prompt Shaping + Project Rules → Goal/DAG Compiler → Scheduler → Agent/Skill/MCP → Kimi Writing → Evidence Gating → Graph Memory + Running State → HUD/Cockpit → User. Key links include:

• Prompt Ingestion and Shaping: Generate structured prompts by combining project rules and Git status
• Goal and DAG Compilation: Convert requests into task graphs, supporting dependency management and parallelism
• Evidence Gating: Tasks must pass verification via evidence like files, commands, and differences before completion
• Local Graph Memory: Persistently store entities such as goals, decisions, and risks, providing a secure context.

Key Features:

• Parallel Execution and Workspace Isolation: Multiple agents run in independent Git workspaces to avoid conflicts
• Skill and MCP Management: Manage skills and configurations via omk skill/omk mcp, with multiple pre-installed skill packages
• Visualization and Observability: Provide tools like HUD, cockpit, and graph viewing
• Decision Tracking: Record all strategic decisions to ensure reproducible runs Use Cases:
• Deliver New Features: omk init→plan→parallel→verify→summary-show
• Code Review: omk review + cockpit
• Design Collaboration: omk design open-design
• Project Memory Query: omk graph view + index

Technical Implementation: OMK is based on TypeScript and Node.js (Node.js 20+), installed via npm: npm install -g @oh-my-kimi/cli, including Rust-native secure loading paths and CI support. It deeply integrates Kimi Code's capabilities like code understanding and multi-file editing. Comparison with Similar Tools:

Feature	OMK	Other AI Programming Frameworks
Target Model	Optimized specifically for Kimi	Multi-model general purpose
Evidence Verification	Core feature	Usually missing
Workspace Isolation	Git worktree	Rarely supported
Graph Memory	Local persistence	Usually none
HUD/Cockpit	Built-in	Rarely built-in
DAG Execution	Natively supported	Rarely natively supported

OMK represents an important evolution direction for AI programming assistants—from "letting AI write code" to "letting AI deliver code reliably". It solves practical problems in AI programming, providing a predictable experience for individual developers and standardized collaboration processes for teams. Control plane tools like OMK will make AI capabilities more reliable, controllable, and auditable, driving the maturity of AI programming tools.