Agent Panel

The Agent Panel is a collapsible right-side panel that runs a full terminal session — Claude Code, Codex, OpenCode, Cursor, Qoder, WorkBuddy, or a plain shell — side-by-side with your spec documents and code. It is the primary interface for handing work to AI during Phase 07 (Code Generation) and beyond.

Opening the panel

Click the Agent button in the top-right toolbar, or press Cmd/Ctrl + Shift + A. The panel slides in from the right. Drag the divider to resize it.

Supported agents

Pick a type card when you create a new agent. Since RunLogos 1.0.3, seven types are supported:

Type	Command launched	Extra flag when Auto-run is checked	Platforms
Claude Code	`claude`	`--dangerously-skip-permissions`	macOS / Windows
Codex	`codex`	`--dangerously-bypass-approvals-and-sandbox`	macOS / Windows
OpenCode	`opencode`	none	macOS / Windows
Cursor	`cursor-agent`	`-f`	macOS / Windows
WorkBuddy	`codebuddy` (bundled with the WorkBuddy desktop app)	`-y`	macOS
Qoder	`qodercli` (`npm install -g @qoder-ai/qodercli`)	`--dangerously-skip-permissions`	macOS
Plain terminal	your default shell	n/a	macOS / Windows

RunLogos looks up the command before creating the session. If it is missing, no session is created and you get an install hint.
Auto-run skips per-action confirmations. Only enable it in project folders you trust.
Trust prompt: Qoder and WorkBuddy ask whether to trust a folder the first time they open it. This is expected, not a failure — answer it yourself in the terminal; RunLogos never answers it for you. Qoder's Auto-run only takes effect in trusted folders.
To let these agents read OpenLogos rules and skills, make sure aiTool in logos/logos.config.json includes them (for example "qoder", "workbuddy", "cursor"), then run openlogos sync and start a new agent session.

RunLogos does not manage agent credentials — configure them in your shell profile or through each agent's own login flow.

Dragging file paths

The file tree on the left supports drag-and-drop into the Agent Panel terminal. Drag any file to insert its absolute path at the cursor. This is the fastest way to give the agent context:

> claude "read @/path/to/requirements.md and implement the login endpoint"

You can drag multiple files in sequence to build up a context list.

Working with Comment-to-Prompt

The most effective workflow combines the Markdown editor and the Agent Panel:

Open a spec document in the center panel.
Highlight a section that needs work.
Use Comment-to-Prompt to generate a precise AI instruction.
The generated prompt appears in a floating card — click Copy or drag it directly into the Agent Panel.
The agent has the exact context it needs without any copy-pasting.

Terminal panel (bottom)

The bottom terminal panel is separate from the Agent Panel and supports multiple tabs. Use it for:

Running tests (npm test, pytest, etc.)
Starting dev servers
Running openlogos verify
Any shell command that isn't an AI agent session

Clickable error lines: When a test or build fails, error output in the terminal includes file paths and line numbers. Click them to jump directly to that file and line in the center editor.

Multi-tab sessions

Both the Agent Panel and the bottom terminal support multiple tabs. Open a new tab with Cmd/Ctrl + T inside the panel. Each tab is an independent PTY session — run the agent in one tab and a dev server in another.

Typical workflow during Phase 07

Open the Spec Chain overview in the center panel.
Open the Agent Panel on the right.
Start Claude Code: claude
Drag requirements.md, openapi.yaml, and the test case docs into the agent.
Prompt: "Implement the endpoints defined in openapi.yaml. Use the test cases in logos/resources/test/ as your acceptance criteria."
Watch the agent write code in the center panel's file tree.
Run tests in the bottom terminal. Click any failing line to jump to the source.

Next step

→ Orchestration Runner