AI Agent Workflows

imux hosts AI work on a terminal-first command center. Prefer real agent CLIs in the terminal for heavy coding loops; use Agent Chat for bounded in-app turns; use Supervisor only after a charter exists. Keep the full loop inside imux.

Which surface should I use?

Pick one primary surface per task:

Terminal agent CLI — Claude Code, Codex, Aider, Cursor Agent CLI, shell tools. Best for long runs and tools you already trust.
Agent Chat — streaming conversation, @file/@symbol, slash commands, apply-with-review diffs. Best for focused, short cycles.
Supervisor — charters, handoffs, loops, evidence. Best for multi-panel orchestration after the goal is clear.

Replacing VS Code / Cursor

Run agents, tests, remotes, browser verification, and Git evidence in imux as the primary workbench. Use agent CLIs and Agent Chat for generation loops; keep files and SCM visible on the same surface.

Slash Commands

Type a slash command in Agent Chat to invoke specific AI actions. Each command targets the current workspace context:

/review — AI-powered code review of staged changes, pull requests, or selected files. The agent reads the diff and provides inline feedback on logic, style, and potential issues.
/fix — AI-assisted bug fixing. Describe the issue or point the agent at failing output, and it proposes targeted changes with diffs you can review and apply.
/explain — AI explanation of code, error messages, stack traces, or architectural decisions. Helps you understand unfamiliar code or debug complex failures.
/test — AI-generated tests for selected code or functions. Produces test cases with appropriate coverage and edge-case handling.

Diff Application

Agents often produce code changes as diffs. imux makes it easy to review and apply those changes:

Agent output containing code blocks with diff syntax is automatically detected and presented with a preview.
Review each change inline before applying it to your working files.
Apply accepted diffs directly into the workspace file tree, keeping Git history clean and auditable.

Context Mentions

Share workspace context with the agent using @mentions. Type @ in Agent Chat to see available context sources:

@file — Mention a file from the workspace to include its contents as agent context. Useful for targeting specific modules or configurations.
@symbol — Mention a symbol (function, class, type) to focus the agent on a specific code unit. Resolved from the workspace source tree.
@pr — Mention a pull request to bring its diff, description, and discussion into the agent context for review or follow-up work.

Agent Chat

Open Agent Chat to start a streaming, multi-turn conversation with your AI agent directly in the workspace:

Launch with Cmd+Shift+L. The chat panel appears inline within the workspace, keeping your terminal and files visible.
Streaming replies: watch the agent's response build in real time, so you can interrupt or redirect mid-generation.
Multi-turn conversations: ask follow-up questions, refine the output, and build context across multiple exchanges.

Skills System

Extend Agent Chat with custom slash commands defined in your repository or home directory:

Create a .clavue/commands/ directory in your project or home folder to define custom slash commands.
Each command is a markdown file with frontmatter that describes what the command does and how the agent should respond.
Custom commands appear alongside built-in slash commands in Agent Chat, scoped to the active workspace.

Use one primary surface at a time. Agent Chat and terminal agents can share a workspace, but stacking chat + supervisor loops + three CLIs without a goal creates noise. You remain in control of what gets applied.