How do specifications and connected project records improve AI coding?
In brief
Spec-driven development defines the intended result and how to prove it. Graph engineering connects that work to dependencies, decisions and evidence. Together they help AI coding agents continue from a durable project record, with a clear next step and an observable definition of done.
- Write observable acceptance criteria before coding.
- Connect tasks to dependencies and decisions.
- Bind completion evidence to the delivered revision.
The problem: specifications lose their connections
Spec-driven development defines the intended result and how to verify it. Graph engineering connects that work to prerequisites, decisions and evidence. A graph need not be a graph database.
A specification can drift from priorities and approvals. Preserve the behavior and why the next step is permissible.
What you may have tried
- A detailed prompt: clarifies one session’s intent, but may not survive its replacement.
- A static specification: preserves requirements, but can drift from implementation and decisions.
- Tickets and dependency diagrams: expose relationships; they still need observable criteria and evidence attached to the delivered revision.
Define behavior and proof before coding; record prerequisite decisions.
From purpose to proof
How Arcadia connects intent to evidence
Arcadia maintains plans and actions with one work pointer, records decisions and approval gates, and checks completion evidence against criteria written before work starts.
- ProjectMission and OutcomeWhy it exists and the desired change.
- ProgressMilestone and PlanThe checkpoint and organized work toward it.
- ActionOne verifiable taskAcceptance criteria and prerequisite Decisions.
- ProofRevision and evidenceWhat demonstrates the delivered behavior.
For a signup product, a Plan can connect a form Action to a prerequisite consent Decision. Evidence shows whether invalid email is refused without creating a registration. A fresh session can follow those relationships to find the next permissible step.
Arcadia currently drives Claude Code, Codex, and OpenCode. Coding agents remain interchangeable execution tools under your direction. Continuity still needs engineering judgment and integrated verification. Start with one task.
Sources and further reading
- GitHub Spec Kit documentation — a structured specification workflow.
- GitHub: spec-driven development with AI — motivation and examples.
- W3C RDF 1.2 Primer — representing connected information as graphs.
- Arcadia canonical semantics — the project’s record model.
Current capabilities and supporting evidence
- Plans, actions and a work pointerWorking
Governed documents with one active plan and one current action.
Supporting evidence - Completion evidenceWorking
Acceptance criteria are checked before an action can be marked done.
Supporting evidence
Capability snapshot: . See what changed →