Agenthud — AI skill for Claude Code

AgentHUD

[](https://www.npmjs.com/package/agenthud) [](https://github.com/neochoon/agenthud/actions/workflows/ci.yml) [](https://codecov.io/gh/neochoon/agenthud)

A **heads-up display** for your AI coding agents — **Claude Code**, **OpenAI Codex**, **AWS Kiro**, and **opencode**. AgentHUD reads each agent's on-disk sessions (JSONL files, or opencode's SQLite store) and merges them into one tree, so a project you've touched from several agents shows as a single row with combined sessions and sub-agents.


It's organized as **three layers, by who's reading** — a human at a glance, a human at the end of the day, and a machine:

• agenthud — the live HUD. A real-time TUI showing which session is [working] versus [waiting] on you, right now, across every agent. Glance at it in a side terminal and know what all your agents are doing. This is the hero — the name is HUD for a reason.
• agenthud summary — the daily digest. Fold a day (or a week) into an LLM engineering summary — a standup-in-one-command habit, via whichever agent CLI you have installed (claude, codex, or kiro-cli, auto-detected).
• agenthud report — the machine layer. Structured Markdown / JSON. summary is literally report piped into an agent CLI, so the daily digest isn't a black box — and you can pipe report into anything else, too.
• agenthud follow — the live machine feed. A chronologically-merged stream of every event (activity + state + lifecycle) across all sessions and sub-agents, as human lines or --json NDJSON — the read-only substrate a higher-level supervisor can consume. See FEATURES.md.

→ **See [FEATURES.md](./FEATURES.md) for the full surface** — every flag, keybinding, config k