Trestle
The setup tool I wish existed when I started building with Claude Code.
Most people install Claude Code and start chatting. That works - for about a week. Then you realize the AI doesn't remember your name, keeps suggesting React when you use Vue, and has no idea you've told it three times to stop adding emojis.
The fix isn't a better prompt. It's a better workspace.
That workspace is a three-legged stool: a Brain (the model you talk to), Agents (skills plus the tools to do real work), and Context (everything it should remember and know). The Brain is Claude Code itself - Trestle builds the other two legs.
Trestle is an interactive setup tool that scaffolds your Context and wires up your Agents in 5 minutes. It asks who you are, what you build, and how you like to work, then sets up the knowledge base, memory, and skills that make every conversation smarter than the last.
Quick Start
Prerequisites
- Claude Code installed (CLI, VS Code extension, or desktop app)
- macOS, Linux, or Windows
- Node.js 18+ (for the guardrail hooks and the
/73 health check - no packages, just node)
Install
macOS / Linux:
git clone https://github.com/liormesh/trestle /tmp/trestle && /tmp/trestle/install.sh
Windows (PowerShell):
git clone https://github.com/liormesh/trestle $env:TEMP\trestle; & $env:TEMP\trestle\install.ps1
This copies the starter skills (/onboard, /cq, /73, /visualize), three small guardrail tools (~/.claude/trestle/) and a bootstrap file to ~/.claude/. That's it - read the script (bash | powershell), it's ~90 lines.
Run
Open Claude Code from your home directory and type:
cd ~
/onboard
~8 questions, ~5 minutes, fully automated setup.
What It Does
Claude asks about your role, projects, tech stack, and preferences, then builds the two legs that sit on the Brain (Claude Code):
Context - what it knows & remembers
A cold layer (knowledge base) - structured folders for everything Claude should know about you, pulled in on demand:
~/Documents/knowledge-base/ (or wherever you choose)
├── me/
│ ├── profile.md ← who you are, what you do
│ └── tone-of-voice.md ← how you communicate
├── projects/
│ └── {your-projects}/ ← one folder per project
│ └── overview.md ← stable description + state: project | area | archive | icebox
├── pm/
│ └── handoff/ ← live status per project - what /cq reads first
├── career/ ← CV, job search, interviews
├── books/
│ ├── README.md ← how to structure knowledge books
│ └── dataviz/ ← shipped example book (powers /visualize)
├── resources/ ← guides, external docs
├── _private/
│ └── credentials.md ← API keys (never leaves your machine)
├── _analytics/
│ ├── usage.log ← one line per skill run (the sweep reads it)
│ └── baseline.json ← the health-check ratchet (created by the first /73)
├── claude-memory/ → persistent memory (symlinked)
├── claude-skills/ → reusable skills (symlinked)
├── _first-run-check.md ← "is it alive?" check: read it back in a fresh session
└── .gitignore
_first-run-check.md is a cold file holding a distinctive magic word. In a fresh session, ask "Read
_first-run-check.md and tell me the magic word" - a correct reply proves the setup works end to end: the
engine runs and it can pull a file on demand. (It's also the workshop's first checkpoint.)
A hot layer (memory) - persistent context loaded into every conversation:
~/.claude/.../memory/
├── MEMORY.md ← index: identity, rules, preferences, pointers - no live status
├── MEMORY-extended.md ← overflow context (loaded on demand)
├── user-profile.md ← your role, expertise, communication style
├── feedback-health-check.md ← pre-session build check
├── feedback-memory-size.md ← signal density + no live status in the index
├── feedback-no-standalone-feedback.md ← new feedback goes inline, not into new files
├── feedback-memory-decay.md ← the sweep (active / archive / promote) - /73 says when it's due
└── project-skill-backlog.md ← logs 3x-rule skill candidates as they recur
Agents - how it acts
Skills (reusable workflows) plus the tools to run them, wired up under ~/.claude/skills/ with an _index.md catalog. Four skills ship out of the box so you're not staring at an empty catalog:
/onboard - the setup interview (re-runnable)./cq <project> - the opening call. Tunes a session into one project's full context (its handoff doc first, then the overview, state, gotchas, health check) before you start. This is the Context leg as a one-command ritual./73 - the sign-off. Runs the workspace health check, then writes the session's learnings back to memory and the KB. This is the ritual that runs the growth loop - corrections become memory and finished work accumulates, on purpose, instead of by luck. It also runs the 3x rule: if a workflow recurred enough to be worth encoding, /73 offers to turn it into a skill on the spot. Built to be cheap: three tool calls, a smaller model./visualize - a ready working skill: turns a plan, report, or dataset into a self-contained HTML or PDF artifact (one file, opens offline). It's also the worked example of the skill + book pattern - it ships paired with a dataviz book in books/, and loads that book's chapters on demand instead of carrying the craft in its own body.
This is the beginning of the README. Read the rest at the original. Read the full README ↗
