Agent sessions

An agent session is Claude Code or Codex working on one task. PacedMind starts the agent in a terminal or in the agent's desktop app on your computer, in the task's folder, or sends the task to the agent's cloud. The agent reads the task over PacedMind's MCP server, works, and reports back when the work is ready for review.

What PacedMind does, and what it doesn't

PacedMind tracks state. It:

• starts the session where the task runs, in the right folder, with the right agent and the connection to PacedMind;
• records when the agent picks the task up, the plan, problems and questions it reports while it works, and when it hands the task back, with its report: a summary, its answer to each Done when item, screenshots, how to check the result and its questions;
• shows running and finished sessions in every view, and notifies when one finishes or its agent waits for you in its terminal.

PacedMind doesn't start sessions by itself, chat with the agent, show its output or review its code. The conversation happens in the agent's terminal or app, where you answer its questions, approve its actions and see what it changed.

Requirements

• Claude Code, Codex, or both, installed and signed in: the command-line tool, the desktop app, or both. PacedMind runs the command-line tools with the commands claude and codex; other commands or full paths can be set in Settings → Sessions. Settings → General shows what PacedMind found on the computer, under Agents found here.
• On Windows, Windows Terminal, for sessions in tabs of one window. Without it, or with Command Prompt chosen in Settings → Sessions, each session opens in its own Command Prompt window. On macOS, each session opens in its own window of the built-in Terminal, or of iTerm when it's chosen in Settings → Sessions.
• For code projects, a folder set on the project, or on the task. See Start, resume and close sessions.
• For the cloud, a git repository in that folder, and cloud sessions set up with the agent's provider.

Sessions PacedMind starts in a terminal are connected to its MCP server automatically. The desktop apps, and sessions you start yourself, need the connection once: select Connect for the agent in Settings → MCP server, under Connect your agents, or see MCP server.

A session at a glance

1. Start: select Start with Claude Code on a task. The agent starts where the task runs, and the task goes to In progress.
2. Pick up: the agent calls start_task and reads the task, its description and sub-tasks.
3. Work: the agent works in the task's folder. It ticks off sub-tasks, records follow-up tasks and reports its plan and questions as it goes. When it waits for you in its terminal, PacedMind shows it.
4. Hand back: the agent calls finish_task with its report, or you select Mark finished for a session that can't. The session is finished, waiting for you, the task is In review, and PacedMind notifies.
5. Review: read the report and check the result, then select Mark done, Request changes to send the work back with what should change, or Answer its questions when the agent asked you something.
6. Next: start the next task. Sessions names a task that waits for this one, with Start session. See Dependencies.

Pages in this section

• How a session runs: every state, what the agent is told, and what PacedMind records.
• Start, resume and close sessions: where sessions start from, where they run, the folder and terminal, permissions, resuming, finishing and closing.
• Review finished work: where finished sessions show up, the report, Request changes, partly done and blocked hand-backs, and what Mark done does.