One CLI. Versioned database in your repo. The same interface for the human and the agent doing the work.
go install gitlab.com/nomograph/synthesist@latest Spec work front-runs context.
Writing a spec before executing forces you to name the goal, the dependencies, the stakeholders, and the acceptance criteria. That structured context is exactly what an AI agent needs to do good work across sessions. The spec IS the context. Synthesist makes spec work a first-class operation so both humans and agents work from the same source of truth.
Every body of work follows a six-phase cycle. The critical constraint: no execution without human approval.
Read the landscape. Stakeholder dispositions, upstream specs, what changed since last session.
Create specs and task DAGs. Dependencies, acceptance criteria, human gates.
Human reviews and approves the plan. The only mandatory checkpoint. Nothing executes without this.
Claim tasks. Do the work. Record discoveries. Write findings to disk before context fills.
Check propagation. Did upstream data change? Are downstream artifacts stale? Fix before moving on.
Close the body of work. Retro with arc and transforms. Merge the session.
Work organized into trees and specs. Each spec has a goal, decisions, tasks, and discoveries. Trees span repos. Our estate has 3 trees covering 19 repos.
Tasks with dependencies, human gates, and acceptance criteria. The system enforces ordering. You can't claim what isn't ready. You can't complete what isn't verified.
When data changes upstream, propagation check finds every downstream artifact that references it. Papers, websites, READMEs. Nothing ships with outdated numbers.
Record signals (what someone said or did), dispositions (what it means for us), and stances (the relationship). Alignment without guessing. Evidence-linked.
All writes require an active session. Sessions branch like git branches for your spec data. Merge atomically when done. No partial state, no conflicts.
Findings that emerge during work. Attached to the spec they came from. The things you learn that weren't in the plan. They persist. They propagate.