Agentic Guidance

Base system prompt, onboarding kit, and benchmark framework for AI-assisted engineering at Moonsong Labs.

Quick start

bash <(curl -sSL https://github.com/Moonsong-Labs/agentic-guidance/releases/latest/download/setup.sh)

This configures your machine for AI-assisted engineering at MSL:

1. Prompt-Train — connects Claude Code to MSL's proxy for monitoring and shared account pooling
2. Core-engineering plugin — skills for brainstorming, TDD, debugging, code review, and more
3. MSL engineering values — injected into every session automatically

Supports Claude Code, Cursor, and Codex. Idempotent — safe to re-run.

How it works per platform

Platform	Values delivery	Skills	Updates
Claude Code	Plugin session-start hook	Plugin marketplace	Auto-updates on session start
Cursor	Global rules file (~/.cursor/rules/)	Via Claude Code plugin	Re-run setup.sh
Codex	Plugin session-start hook (via clone)	Symlinked from clone	Re-run setup.sh

Cursor's sessionStart hooks from plugins don't fire reliably as of v2.6.x. The script uses global rules as a workaround.

Flags

./setup.sh --install-claude    # Prompt to install Claude Code even if other tools exist
./setup.sh --help

The prompt

The MSL engineering values prompt lives in knowledge-work-plugins (source of truth) and is distributed via the core-engineering plugin's session-start hook.

Why values over rules

See benchmark/README.md for the full benchmark methodology, results, and rationale.

Repository structure

agentic-guidance/
├── setup.sh              <- zero-to-one installer
├── benchmark/            <- validation framework
└── README.md