CTOP: Terminal Monitor for Claude Code and AI Agent Sessions

Terminal Dashboard for Agent Telemetry

ctop solves the visibility problem of running background AI agents. When running multiple iterative sessions of Claude Code, Codex CLI, OpenCode, or Devin, keeping track of active context windows and API spend is challenging. The tool reads process information from ps (or PowerShell on Windows) and matches active processes with local metadata and database entries from Claude (~/.claude/projects/), Codex (~/.codex/sessions/), OpenCode (~/.local/share/opencode/), or Devin (~/.local/share/devin/cli/).

The dashboard provides a visual breakdown of the context window, highlighting free, input, output, and cached segments. It also estimates per-session and aggregate API costs, preventing surprise bills from runaway agent loops.

Autonomous Agent Self-Introspection

You can install a native skill in Claude Code to let your agent query its own metrics. By copying the ctop skill into your projects, the agent can use the CLI mode under the hood.

mkdir -p ~/.claude/skills && cp -r skills/ctop ~/.claude/skills/

Once installed, you can ask Claude Code: "how much has this session cost?" or "is my context about to compact?". The agent will execute CLI commands like ctop whoami or ctop stats to analyze its state and optimize its prompt strategy.

Session Control and CLI Automation

The tool offers list and card-grid views, live log streaming, and process controls. Developers can search and filter sessions by directory or branch, pin active sessions, and safely terminate hanging agents via SIGTERM. For scripting, ctop ls --json exports machine-readable metrics, enabling you to build automated scripts that detect when an agent's context window is running low (e.g. less than 15% free capacity) to prompt a /compact command.