scoutme

Milk — Productivity skill for Claude Code

Productivity community

milk — a CLI/TUI that routes prompts between agents: a primary agent and a configurable escalation agent, with session-aware state management and real-time streaming.

How to install Milk

This entry records only its repository, not the path inside it, so there is no exact command to give. Open scoutme/milk and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Milk does

milk — a CLI/TUI that routes prompts between agents: a primary agent and a configurable escalation agent, with session-aware state management and real-time streaming. Escalation happens automatically (via routing rules) or manually, preserving context across the handoff between agents.

Alternatives in Productivity

Linear - Ticket Management — You are tasked with managing Linear tickets, including creating tickets from thoughts documents, updating exis 10k ★
Product Team Skills — Product management team skills 5.3k ★
Claudekit Meta Tools — Meta-tools for skill management 1.9k ★

README

milk

![milk](docs/images/milk.png)

Switch models, not context.

milk is a terminal AI assistant that routes each prompt between a fast primary agent and a deep escalation agent — keeping the full conversation in sync across both. Start cheap. Go deep when you need it. Switch mid-workflow.

Install

Pre-built binary (macOS, Linux, Windows via WSL2):

curl -fsSL https://raw.githubusercontent.com/scoutme/milk/main/install.sh | sh

Installs to `~/.local/bin/milk`, verifying the release checksum. Pin a specific version with `MILK_VERSION=v0.2.0`.

Build from source instead (`git clone` + `task build`, or in one line):

curl -fsSL https://raw.githubusercontent.com/scoutme/milk/main/install-from-source.sh | sh

Requires Go 1.21+. See [docs/getting-started.md](docs/getting-started.md) for what to do next.

What it does

Automatic routing — each prompt is classified and sent to the right agent without you changing tools
Context handoff — when escalation fires, the primary conversation is reformatted as context; the escalation agent orients itself without a separate setup step
Persistent memory — a Percept store survives across sessions; key facts are reinforced, decay, and promote to long-term memory over time (NREM consolidation)
Built-in tools — the primary agent has bash, file read/write/edit, grep, find, HTTP GET, session access, and memory tools without any extra configuration
Streaming TUI — bubbletea terminal UI with a scrollable transcript, live memory panel, status bar, and input history
Loop detection — monitors agent output for repeating patterns; warns in the status bar and auto-interrupts when the agent gets stuck looping (configurable, works with all providers)
Aider and smolagents — plug in aider-chat or smolagents as either the primary or escalation agent
Evaluation harness — run the same scenarios against different agents (or against the Claude Code CLI directly) and compare LLM-ju