The synthesis work system gives human-AI work a durable home: shared methods, project memory, safeguards, coordination, and evidence you can inspect. It is open, vendor-neutral, and designed to keep working when you change tools.
One door must account for every layer. A partial setup is legitimate when it is deliberate. A silent gap is not. Each layer should read as installed, declined, or missing, and a check that cannot run must never look like a pass.
02 / the system
More than a skill pack.
The plugin is the public core. The working system includes every layer below. This map shows what the current bootstrap reaches and where the one-door work is still incomplete.
Reached today Exists through separate setup One-door completion work
01
Skills
Portable methods for coding, writing, project management, governance, and operations.
Reached today
02
Session context
Verified time, active project state, and recovery at the beginning of a session.
Reached today
03
Hooks and gates
Fail-closed checks for commits, messages, coordination, publishing, and task completion.
Separate setup
04
Agent kernel
A thin instruction core that carries identity, invariants, routing, and enforcement across clients.
Completion work
05
Runtime engines
Durable machinery for receipts, safeguards, daily operations, and repository health.
Separate setup
06
Coordination
Claims, handoffs, and shared project state when several agents work at once.
Separate setup
07
Doctors and conformance
Checks that distinguish a working system from a system that merely has files installed.
Separate setup
08
Personal policy
Your voice, preferences, disclosure boundaries, and agent identities in files you own.
Completion work
09
Organization and knowledge
Personal and shared knowledge bases, plus organization configuration without installer forks.
Separate setup
10
Lifecycle
Update notices, safe convergence, and receipts that show every installed machine is current.
Completion work
03 / start
Install the public core.
The current bootstrap installs the native public plugin into supported local clients, activates verified session context, and runs its onboarding doctor. It can also scaffold a personal project workspace. It does not yet claim to produce every layer in the system map.
Option 1
Ask your AI agent
Copy this prompt into a local coding agent that can inspect files and run terminal commands on your computer.
Set up the public synthesis work system on this computer from the stable release.
First inspect https://github.com/synthesisengineering/synthesis-skills/blob/stable/onboard.sh and the onboarding code it invokes, then summarize what the installer will change. Run the guided installer with the full profile unless I tell you to install skills only. Do not overwrite files I have changed. Run the onboarding doctor and report every layer as installed, declined, missing, or unverifiable.
Make installation the final action in this task. If a client restart is required, tell me exactly what to restart. If you cannot inspect local files and run terminal commands, say so instead of claiming the installation succeeded.
Option 2
Use the terminal
On macOS or Linux, or on Windows through WSL, close active agent sessions and run the stable bootstrap.
curl -fsSL https://raw.githubusercontent.com/synthesisengineering/synthesis-skills/stable/onboard.sh | sh
Installed means the doctor verified the component it names.
Declined means you chose not to install that layer.
Missing or unverifiable means the run is not complete.
The public catalog works as native plugins in supported clients and as directAgent Skills in compatible tools. It is licensed under CC0. Browse the catalog on GitHub.
04 / where this leads
A system you can leave, and return to.
Native agent memory helps inside one product. Synthesis keeps the work itself in durable, versioned records that different tools and people can read. The harness becomes an adapter. Your project remains yours.