Merlino · Internal Playbook

The Clief System, set up to use.

One method for giving AI agents exactly the context they need and nothing they don't. Shipped as 5 reusable templates plus a one-click launcher. Same engine for agency clients, coding, and the mastermind.

The one idea

A context delivery system, not a folder system.

Every task needs three answers, and most setups fail the third: what am I doing, how, and what should I NOT load. Three layers keep the window clean and the agent on the rails.

Layer 1
CLAUDE.md - the map

Always loaded. Folder map, naming rules, routing table. Under ~200 lines.

Layer 2
CONTEXT.md - the router

"What's your task? Go to workspace X, also load Y." 30 to 50 lines.

Layer 3
workspace/CONTEXT.md - the work

Self-contained. Says exactly what to load and what to skip per task.

The 60 / 30 / 10 split

60% infrastructure · 30% orchestration · 10% AI judgment. Route bulk work down a layer before spending Opus tokens.

Why it works

A blog-writing agent never loads the deploy pipeline. A coding agent never loads brand voice. Less wasted context means more one-shots and less babysitting.

What's in the folder

Five templates + a one-click launcher.

Double-click New Clief Project, pick a template, name it. It builds the folder, opens it, and offers to launch Claude Code. Or run new-clief-project.cmd.

workspace-blueprint

The blank canvas

Generic 3-layer starter with worked example workspaces.

map → router → workspaces
client-delivery

Agency engagements

Isolated context per client so brand never bleeds across accounts.

discovery → build → review → handoff
content-production

Content, any format

Locked voice per project. Wire blog, PAA, and sourcing skills per stage.

research → script → production
small-business-ops

Recurring ops engine

The "same work every week" machine. Built for local-SEO lead gen.

intake → process → deliver
custom-frontend

Build a UI on Claude

Tool Ladder + PRD + a repo index to fork. For dashboards and tools.

PRD → fork → build in sessions
+ extras

Starters & reference

3 interview-style skill builders plus reference docs.

skill-starters/ · reference/

Applied

Same engine, across the business.

Only two things change per use: which workspaces you define, and which skills you wire into each stage.

Agencies

Per client → client-delivery. Recurring local SEO → small-business-ops. Onboarding = copy + one routing row. Maps to the fleet: Oliver scopes, Carlos sequences.

Coding

New app → workspace-blueprint. Dashboards and tools → custom-frontend, PRD before code. Research → PRD → build in sessions → screenshot before done.

Mastermind

Run the cohort → content-production + a community/ workspace. Package SOPs as done-for-you workspaces. Members productive in 15 minutes is the sellable outcome.

SEO fleet

One scoped workspace per campaign for Einstein, Tommy, Ghost, Sherlock, Shakespeare. Wire each agent's skills into the stage where they belong.

Cheat sheet

Task in, template out.

Most work stops at Level 3 (VS Code + Claude Code). Build a custom front-end only when native tools genuinely don't fit.

Start anything
Double-click New Clief Project
Onboard an agency client
client-delivery, fill _config/
Run recurring SEO ops
small-business-ops
Produce content
content-production
New app or repo
workspace-blueprint
Build a dashboard / UI
custom-frontend, fill PRD.md first
Package an SOP
a file in skill-starters/
Terminal instead of shortcut
init-workspace.sh "<dir>" "<Name>" "<tag>" "<template>"