Send to Claude - Self-Aware Task Completion for Claude Code

English | 中文

Claude can now send messages to its own input to wake itself up.

This plugin enables autonomous multi-step workflows where Claude continues working without waiting for you to return.

Requirements

Your Claude Code session must run in one of these terminals:

• tmux
• iTerm2

The plugin auto-detects your terminal environment.

---

The Problem

You: Train the model, then analyze results and generate a report
Claude: Starting training...
[10 minutes later training completes]
Claude: (silence)
You: (come back) Is training done?
Claude: Yes! Starting analysis...

Claude doesn't know when background tasks complete. Multi-step workflows break here.

---

The Solution

Let Claude send messages to itself:

# Claude's training script
import os

train_model()  # Takes 10 minutes

# Wake itself up when done
os.system('send-to-claude "Training complete! Accuracy: 94%, starting analysis..."')

When the script finishes, Claude's input receives this message automatically, Claude wakes up and continues.

Also enables periodic polling: Check emails hourly, trigger sleep mechanism every 4 hours, periodic memory consolidation, etc.

---

How It Works

1. Claude runs in tmux/iTerm2 with environment variables $TMUX or $ITERM_SESSION_ID
2. Scripts created by Claude inherit these environment variables
3. Script calls send-to-claude → finds Claude's session via inherited variables
4. Sends message to Claude's terminal input

Technical implementation:

# tmux
tmux send-keys -t $session:$window.$pane "message"
sleep 0.5
tmux send-keys -t $session:$window.$pane "Enter"

# iTerm2
tell session to write text "message"
delay 0.5
tell session to write text ""

---

Installation

In Claude Code:

/plugin marketplace add https://github.com/shanjiaming/send-to-claude-plugin
/plugin install send-to-claude

Then run this command to complete installation:

bash ~/.claude/plugins/cache/send-to-claude/send-to-claude/0.1.0/install.sh

Make sure ~/.local/bin is in your PATH.

To uninstall:

bash ~/.claude/plugins/cache/send-to-claude/send-to-claude/0.1.0/uninstall.sh

---

Usage

Just give Claude multi-step tasks:

"Train model, then analyze results"
"Process 1000 files, report progress every 100"
"Download data, clean it, train, deploy"

Claude automatically adds callbacks to scripts and completes workflows autonomously.

Test it: Tell Claude "Test the callback mechanism"

---

Comparison with Background Agents

Feature	send-to-claude	Background Agents
Task completion callback	✅	✅
Any-time callbacks	✅	❌ Only on completion
Claude continues autonomously	✅	❌ Needs user input
Multi-step automation	✅	❌ User needed per step
Periodic polling	✅	❌

---

Technical Details

Why not polling?

Polling occupies session, wastes tokens, blocks other tasks.

Callbacks: Claude releases resources, wakes up precisely when needed, zero consumption while waiting.

How do environment variables work?

Unix child processes inherit parent environment variables:

Claude (tmux, has $TMUX)
  └─ python train.py (inherits $TMUX)
       └─ os.system('send-to-claude') (inherits $TMUX)
            └─ Reads $TMUX, finds Claude's session

---

Summary

This plugin transforms Claude from "passive responder" to "autonomous worker."

After installation:

• No need to watch progress
• No need to come back and say "continue"
• No need to manually chain steps

Claude can truly work independently.

---

License

MIT License - see LICENSE