Home

agent-watch

CLI tool for monitoring and browsing Claude Code sub-agent and session transcripts.

Why

Claude Code stores all session and sub-agent activity as JSONL files in ~/.claude/projects/, but provides no built-in way to browse, search, or monitor them. When you're running multiple sub-agents in parallel, you have no visibility into what they're doing, how much context they've consumed, or whether they've finished.

agent-watch solves this by giving you:

• Visibility -- see what every agent is doing right now, what model it's using, how many tokens it's burned, and whether it's still running
• Session history -- browse and search past sessions across all projects, filter by project name, paginate through large transcripts
• Agent orchestration -- wait blocks until agents complete, enabling reliable parallel workflows where you launch N agents and continue only after all finish
• Agent-readable output -- non-interactive commands are designed for Claude Code agents themselves to use, so your main agent can read sub-agent transcripts and act on results

Without agent-watch, monitoring sub-agents means manually finding JSONL files, piping them through jq, and guessing whether an agent is still running. With it, you get agent-watch list to see all active agents, agent-watch <id> to read any transcript, and agent-watch wait <id> to block until completion.

Pages

• UI Overview -- screenshots of interactive session browsing, agent history, transcript view, and pagination
• Token Log & PID Tracking Setup -- configure Claude Code hooks for faster wait detection, token usage logging, and process-level agent tracking