OpenCLI

Executes CLI commands in the context of Chrome's existing authenticated browser sessions via a Browser Bridge Chrome Extension and micro-daemon, eliminating credential storage. The architecture intercepts Chrome's session cookies and authentication state through Chrome DevTools Protocol (CDP) connections, allowing commands to piggyback on user-authenticated web sessions without ever exposing passwords or tokens to the CLI runtime.

Transforms websites into CLI commands using declarative YAML pipelines that define data extraction, transformation, and output steps without code. The pipeline executor (src/pipeline/executor.ts) chains together steps like HTTP requests, DOM parsing, template rendering, and data filtering using a template expression syntax that supports variable interpolation and conditional logic, enabling rapid adapter creation for simple-to-moderate use cases.

Defines a composable set of pipeline steps (download, parse, filter, tap, intercept) that can be chained together to build complex data extraction and transformation workflows. Each step type performs a specific operation (HTTP fetch, DOM parsing, data filtering, side effects, network interception) and passes results to the next step, enabling declarative definition of multi-step workflows without imperative code.

Enables developers to extend OpenCLI with custom adapters, commands, and pipeline steps through a plugin architecture. Plugins can register new adapters, define custom pipeline steps, and hook into the command execution lifecycle, allowing third-party developers to add functionality without modifying core OpenCLI code.

Defines a standardized AGENT.md format that describes OpenCLI adapters and commands in a machine-readable way, enabling AI agents to discover, understand, and execute tools through a unified interface. The format includes command descriptions, parameters, examples, and execution patterns, allowing LLM-based agents to reason about available tools and construct appropriate commands.

Enables developers to write robust adapters in TypeScript that execute custom code within the browser context via CDP injection, allowing full access to DOM APIs, JavaScript execution, and complex state management. Adapters are compiled and executed as injected scripts within Chrome's runtime, providing programmatic control over browser interactions beyond what declarative YAML pipelines support.

Implements a hierarchical strategy system (src/cascade.ts) that automatically detects and applies appropriate authentication methods across different website types. The cascade evaluates strategies in order (cookie-based, token-based, OAuth, form-based, custom) and selects the first applicable method based on site characteristics, enabling adapters to work with authenticated sessions without explicit credential configuration.

Generates YAML or TypeScript adapters automatically from website URLs using an AI-driven AutoResearch engine that explores site structure, identifies API endpoints, and synthesizes adapter definitions. The engine combines deep exploration (API discovery), strategy cascade (authentication detection), and synthesis (YAML generation) to create working adapters from minimal user input, enabling rapid CLI wrapper creation without manual adapter writing.

Maintains a centralized registry (src/registry.ts) of 79+ pre-built adapters covering hundreds of commands across global platforms (Bilibili, Zhihu, Twitter/X, Reddit, etc.). The registry provides a unified interface to discover, list, and execute commands from diverse sources, abstracting away differences in underlying authentication, API patterns, and output formats into a consistent CLI experience.

Implements a template engine (referenced in Architecture Deep Dive) that supports variable interpolation, conditional logic, and filter functions within YAML pipelines and adapter definitions. The engine evaluates expressions like `{{ variable | filter }}` at runtime, enabling dynamic data transformation and output formatting without requiring full programming language capabilities.

Supports multiple output formats (JSON, plain text, formatted tables) for command results, with automatic format selection based on context or explicit user specification. The output system (src/output.ts) handles serialization of complex data structures into human-readable or machine-parseable formats, enabling CLI results to integrate with downstream tools and scripts.

Implements a persistent micro-daemon that maintains CDP connections to Chrome instances, enabling long-lived browser sessions for command execution without spawning new browser processes per command. The daemon (referenced in Browser Integration Implementation) manages session lifecycle, handles CDP message routing, and provides a stable interface for adapter execution within browser contexts.

Provides mechanisms to discover available adapters, list their commands, and inspect command parameters and descriptions without executing them. The discovery system enables users and AI agents to understand what tools are available, what commands they support, and what parameters are required before execution, supporting both interactive exploration and programmatic tool discovery.