lingtai
May 30, 2026 · View on GitHub
Connect Claude Code to a LingTai agent network. Read mail from your agents, send instructions, check liveness, and manage the network — all through the shared human mailbox.
Other coding agents: Codex CLI users →
Lingtai-AI/codex-plugin. Anything else →Lingtai-AI/lingtai-skill(the canonical protocol skill — copy into your tool's skill dir).
Install
claude plugin add Lingtai-AI/claude-code-plugin
What you get
- SessionStart hook — detects
.lingtai/projects, reports inbox count and any mail awaiting pickup lingtaiskill — full reference for the filesystem mailbox protocol:- Send mail via the human's outbox (pseudo-agent model — agents poll and claim)
- Read incoming mail with threading via
in_reply_to - Agent discovery and orchestrator identification
- Liveness checks (heartbeat)
- Lifecycle management — sleep, suspend, CPR, refresh, clear
- Signal files — prompt injection, soul inquiry
- Delivery + reply monitoring with the Monitor tool
- Read tracking across sessions
- Per-host
via:tag for message attribution
Usage
Once installed, Claude Code auto-detects LingTai projects on session start. Ask it to:
- "Check my inbox"
- "Send a message to the orchestrator: research X"
- "Are any agents alive?"
- "CPR the orchestrator"
- "What's the network status?"
The skill activates on demand — it won't interrupt your coding unless you ask. In projects without .lingtai/, the plugin does nothing.
Related skills
The lingtai skill in this plugin is the integration protocol — how Claude talks to the mailbox. For deeper reference about the agent runtime itself, Claude should load the skills that ship with the LingTai TUI. When the TUI (lingtai-tui) is installed, these live at ~/.lingtai-tui/bundled-skills/:
| Skill | Location | What it covers |
|---|---|---|
| lingtai-anatomy | ~/.lingtai-tui/bundled-skills/lingtai-anatomy/SKILL.md | Memory system, filesystem layout, runtime anatomy (turn loop, state machine, signals, molt, mail atomicity) |
| lingtai-tutorial-guide | ~/.lingtai-tui/bundled-skills/lingtai-tutorial-guide/SKILL.md | How LingTai works — concepts, philosophy, lessons |
| lingtai-portal-guide | ~/.lingtai-tui/bundled-skills/lingtai-portal-guide/SKILL.md | Portal API endpoints, topology recording, replay |
| lingtai-recipe | ~/.lingtai-tui/bundled-skills/lingtai-recipe/SKILL.md | Behavioral recipes, network cloning, export/import |
| lingtai-mcp | ~/.lingtai-tui/bundled-skills/lingtai-mcp/SKILL.md | MCP server configuration for agents |
| lingtai-changelog | ~/.lingtai-tui/bundled-skills/lingtai-changelog/SKILL.md | Breaking changes, renames, migrations |
The plugin doesn't bundle these skills — they're maintained in the main LingTai repo under tui/internal/preset/skills/ and populated to ~/.lingtai-tui/bundled-skills/ on TUI install. If you use this plugin, the TUI is installed, so this path exists. When deeper information about LingTai internals is needed, Claude should Read the relevant file from that location — the plugin's own lingtai skill references this path as its authoritative fallback.
For plugin developers: the integration skill at skills/lingtai/SKILL.md is vendored from the canonical Lingtai-AI/lingtai-skill repo. Do not edit it here — edit it upstream and run scripts/sync-from-canonical.sh to pull the latest. The anatomy and other reference skills live in the main lingtai repo.
Requirements
- A running LingTai project (
.lingtai/directory with agents) - Python 3 (for UUID generation and liveness checks)
lingtai-tuiinstalled (populates~/.lingtai-tui/bundled-skills/with the reference skills above)
License
Apache-2.0