Mythic Agent：维京主题的多智能体终端 AI 编程助手

Welcome to Mythic Agent, an ultra-advanced, Viking-themed AI coding assistant right in your terminal. Designed with both intuition and rigorous architecture in mind, Mythic Agent is more than just a chatbot—it is a complete, multi-agent forge capable of managing entire repositories, running shell commands, coordinating sub-agents, and embracing the true spirit of the Mythic Engineering protocol.

🌟 Key Features

🛡️ The Viking TUI (Terminal User Interface)

Built on Textual, Mythic Agent features a gorgeous, fully interactive TUI styled with the Tokyo Night theme. Command your agent with a beautiful, heavy-bordered aesthetic, all from the comfort of your command line.

🧠 Mythic Engineering Mode

Enable the Mythic Engineering Mode directly from the sidebar to supercharge your AI. Following the MD Protocol, the agent switches to an Architecture-First mindset and delegates complex tasks to six highly specialized, built-in Sub-Agents:

Skald - The Visionary Poetess
Architect - The Dominant Designer
Forge Worker - The Fiery Builder
Auditor - The Merciless Verifier
Cartographer - The Sensual Wayfinder
Scribe - The Gentle Guardian of Memory

🐙 Seamless GitHub Integration

Forget switching contexts to authenticate or run raw git commands.

Easily configure your GitHub Repo and Personal Access Token (PAT) directly inside the TUI.
Instantly use high-level slash commands:
- /commit <message> - Automatically stages, commits, and pushes to remote.
- /issue <title> - Instantly opens a GitHub issue.
- /pr <title> - Instantly creates a Pull Request.
- /status - Checks your local Git status.
- /gh <command> - Runs raw GitHub CLI commands automatically authenticated via your saved PAT.

🛠️ Advanced Claude-Code Capabilities

Mythic Agent incorporates premium developer features:

Command Approvals: Any destructive or terminal command the AI wishes to run requires your explicit approval via a modal.
/doctor - Automatically catches terminal errors and fixes them.
/undo - Restores the last unchanged agent file edit from its workspace journal.
/add <file> - Explicitly pull specific files into the agent's context.

🌐 Multi-Provider Divine Sources

Mythic Agent supports multiple LLM endpoints out of the box. Easily configure:

OpenRouter (for Claude 3.5 Sonnet, Llama 3, etc.)
DeepSeek
OpenAI
OpenCode Go

🆕 Recent Features

mythic doctor — System health checks (config, providers, tools, sessions, disk, dependencies) with --json for machines, --fix for safe auto-repairs, and --live for opt-in live provider checks.
Agent lifecycle & orchestration — Typed task IDs with queued/running/completed/failed/cancelled states, delegation-cycle detection, cancellation propagation, and per-agent turn/tool budgets. Inspect via the MCP mythic_task_status / mythic_task_cancel tools.
mythic cache — Inspect or clear the LLM response cache (--stats, --clear) so you never pay twice for identical requests.
mythic costs — A model API cost ledger per workspace; unknown pricing is reported as unknown, never invented.
mythic metrics — Recorded timers, counters, gauges, and traces for the workspace.
mythic review — Static review of Python files with severity filtering and JSON output.
mythic theme — List and apply color themes for the interface.
mythic tutorial — Interactive first-run tutorial from any terminal.
Shell completions — Ready-made completions for bash, zsh, and fish in completions/.
MCP server (mythic-mcp) — External AI agents get the same workspace/policy/session/task services over stdio, with versioned tool contracts.
mythic sessions --search — Filter workspace sessions by id, status, model, or metadata.

See the User Guide and API reference for the full tour.

🚀 Installation & Setup

Clone the repository:

git clone https://github.com/hrabanazviking/Mythic_Agent_Coder_CLI.git
cd Mythic_Agent_Coder_CLI

Install the CLI globally:
```
 pip install -e '.[tui]'
```
Summon the Agent:
```
mythic
```

First-Time Configuration

Core-only installation uses pip install -e . and supports mythic chat, mythic run, mythic --help, and mythic --version without UI or speech packages. Use mythic run --format json "your task" for AI callers; it defaults to read-only. See the human/machine CLI guide for permissions, stdin, JSONL and durable workspace sessions (--resume, mythic sessions --export), recovery and exit codes. Optional integrations are installed with the mcp, knowledge, voice, and voice-cloning extras. Human TUI/terminal and machine callers share tool permissions. Ctrl+C in terminal or /stop in TUI cancels active owned requests, approvals and commands; command results include exit status and captured output. Direct/machine effects require an explicit policy; delegated agents inherit the caller's mode. See the development roadmap for active reliability work and the developer guide for test/build commands. This is an alpha; roadmap features are only available once their implementation is recorded.

When you run mythic for the first time, press F2 to open the Setup Wizard. Here, you can:

Select your Provider and enter your API Key.
Fetch and select the desired AI Model.
Edit the primary System Prompt (By default: A super smart, pretty, modern Viking female coder).
Add Global Rules that apply to all sub-agents.
Summon custom Warriors (Sub-Agents) or use the default Mythic Engineering team.

📜 How to Use

Simply launch mythic in your project's directory and type naturally! You can ask the agent to:

"Read my src/main.py and find the bug."
"Create a new React component for a login screen."
"Run the tests and fix any failing ones."

Runic Commands (Slash Commands)

You can type these at any time in the chat input:

/help - View all available commands.
/setup (or F2) - Open the Setup Wizard.
/clear - Clear the current conversation history.
/add <file> - Add a specific file to the context.
/commit <msg> - Auto-commit and push changes to GitHub.
/issue <title> - Create a GitHub issue.
/pr <title> - Create a GitHub Pull Request.
/status - Check Git status.
/gh <cmd> - Run native GitHub CLI commands.
/test - Run your project's test suite natively.
/undo - Roll back the last agent file edit.
/doctor - Auto-fix a command output.
/quit - Leave Valhalla (Exit the app).

⚖️ License

Mythic Agent Coder CLI is licensed under the Apache License, Version 2.0. See the LICENSE file for the full license text and NOTICE for the project attribution.

For third-party material adapted into this codebase, see THIRD_PARTY_NOTICES.md. Per the Apache-2.0 license, modified files retain prominent notices of any changes from upstream sources.

Unless required by applicable law or agreed to in writing, this project is distributed on an "AS IS" BASIS, without warranties or conditions of any kind, either express or implied.

Distribution and Privacy Position

Mythic Agent Coder CLI is published here as source code and project material.

The author does not require users to provide age, identity, government ID, biometric data, or similar personal information in order to access or use the source code in this repository.

The author may decline to provide official binaries, installers, hosted services, app-store releases, or other official distribution channels where doing so would require age verification, identity verification, or similar personal-data collection.

Any third party who forks, packages, redistributes, deploys, hosts, or otherwise makes this software available does so independently and is solely responsible for compliance with applicable law, platform policy, and distribution requirements in their own jurisdiction and context.

See LEGAL-NOTICE.md for details.

☕ Support the Project

If you enjoy my open-source projects and want to help support continued development, research, testing, and experimentation, you can leave a tip through PayPal:

Support my work on PayPal.Me

Support is always appreciated, but never required. Using, sharing, testing, contributing to, or starring the projects helps too. 🖤⚙️ᚱ