Baseline & scope
Characterize HEAD. Bound the blast radius before an agent touches code.
PRODUCT
The control plane around AI coding
Agents write. RIVA scopes, gates, and proves. Same contract in the IDE, MCP, and CI.
01
Prepare → Agent → Gate → Advance
Progress requires met criteria. The model’s reply is never done.
Bounded change
Handoffs stay inside the map. Callers outside the radius stay green.
Unit · Integration · Journey
Three depths — isolated behavior, joins, and cross-layer paths.
Human approve
Lock evidence. A person decides what ships.
02
One quality bar
Unit
Isolated function and symbol behavior after edits.
Integration
Composition checks across module joins — at unit speed.
Journey
Cross-layer paths, with results in the evidence pack.
Surfaces: IDE · MCP · CLI / CI — not a second editor.
Cursor
Claude Code
Windsurf
Copilot
Gemini
03
How the system fits together
Workspace governance, surfaces, workflows, quality tiers, and an engine — anchored in the live repository.
System Overview
RIVA Workspace
Shared Baseline, Gates, Evidence
IDE
Operator Controls
Agent MCP
Structured Tools
CLI / CI
Pipeline Gates
Workflows
Baseline · Quality · Change · Team
Unit
Isolated Behavior
Integration
Composition
Journey
Cross-Layer Path
Engine
Analyze · Handoff · Gate · Evidence
Codebase
System of Record · Characterized HEAD
Layer 01
RIVA Workspace
The shared control plane for quality across the team. Establishes a single baseline, acceptance criteria, and evidence contract used by developers, AI agents, and CI.
Baseline Policy
Pins verified HEAD behavior as the reference for later change.
Acceptance Gates
Work advances only when criteria pass, not on model claims.
Evidence Contract
Standard provenance and regression packs for PRs and CI.
Layer 02
Control Surfaces
A consistent interface for humans, coding agents, and automation. The same prepare, verify, and lock semantics apply in the IDE, in agent sessions, and in CI.
IDE
Initialize, loops, impact, and next steps inside the editor.
Agent tools (MCP)
Context, handoffs, and verify actions for coding agents.
CLI and CI
Same gates and evidence artifacts in automated pipelines.
Layer 03
Workflows
End-to-end operating sequences for AI-assisted development on existing systems, from establishing trust in the codebase through change delivery and organizational enforcement.
Baseline Initialization
Map the repo, characterize behavior, lock a baseline before AI edits.
Quality Verification
Run Unit, Integration, and Journey as one quality bar.
Change Delivery
Blast-radius scope, gated handoffs, re-verify, evidence pack.
Team Enforcement
One definition of done for contributors, review, and CI.
Layer 04
Quality Tiers
A structured quality model derived from system structure, not line coverage alone. Each tier answers a distinct question about correctness under change.
Unit
Isolated function and symbol behavior after edits.
Integration
Composition checks across module joins at unit speed.
Journey
Cross-Layer Paths, with results in the evidence pack.
Layer 05
Engine
Local-first primitives that map the system, prepare agent work, enforce gates, and produce verifiable artifacts on enterprise repositories.
Structural Analysis
Symbol and dependency index for blast radius and context.
Handoff Orchestration
Scoped packages, agent sessions, completion tracking.
Gate Evaluation
Verify results before the workflow can advance.
Evidence and Lock
Provenance, regression output, optional baseline lock.
Layer 06
Codebase as System of Record
All analysis and verification are grounded in the live repository. Documentation informs; source code decides.
Application Repositories
Live production code with modules, deps, and constraints.
Characterized HEAD
Pinned revision that later AI changes are measured against.