Skip to content

AI SDLC Operating Model

Purpose

This document defines the canonical operating model for AI-enabled software delivery in BaseCoat and consuming repositories.

Two Complementary Planes

BaseCoat uses a two-plane model:

  1. Guardrails plane (BaseCoat) — Governs how work is performed.
  2. Visibility plane (GitHub Projects + repo operations) — Tracks what work is happening and its status.

Both planes are required. Guardrails without visibility creates blind execution. Visibility without guardrails creates inconsistent execution.

Plane 1: Guardrails (BaseCoat)

BaseCoat provides the execution contract through four primitives:

  • Agents — role-specific workflows and execution logic
  • Skills — reusable procedures, templates, and decision aids
  • Instructions — ambient standards, safety, and quality policies
  • Prompts — structured entry points for repeatable tasks

Guardrails outcomes:

  • Consistent quality and policy compliance across sessions
  • Reusable patterns instead of ad hoc prompts
  • Safer automation with scoped instructions and validation

Plane 2: Visibility (GitHub Operating Surface)

Visibility is provided by GitHub-native artifacts:

  • Issues for scoped work items
  • Pull requests for reviewed change units
  • Workflow runs for operational evidence and enforcement
  • Milestones / Projects for sprint and portfolio tracking

Visibility outcomes:

  • Execution state is observable and auditable
  • Blockers and carryover are explicit
  • Planning and delivery telemetry are available for retrospectives

Operating Rule

Use both planes in every sprint:

  1. Define work in issues and milestones (visibility).
  2. Execute with BaseCoat agents/skills/instructions/prompts (guardrails).
  3. Merge only when required checks are green (guardrails + visibility gate).
  4. Capture evidence in PRs/issues/workflow runs (visibility).

Practical Mapping

Delivery need Guardrails plane Visibility plane
Build a feature Agent + paired skills + instructions Issue + PR + checks
Enforce standards Instructions + validation workflows Status checks + run logs
Coordinate sprint execution Prompt/agent routing patterns Milestones + project board
Close out and handoff DoD-style execution checks Carryover issues + summary links