Skip to content

ADR 0002: Multi-surface hybrid execution model

  • Status: Accepted
  • Date: 2026-05-24

Context

The framework is designed for coordinated work across VS Code Copilot, GitHub Copilot Chat/Coding Agent, GH CLI, GitHub Mobile, and external AI agents. Real workflows move between discovery, planning, implementation, review, and follow-up on different surfaces. A single-surface or chat-only approach breaks continuity and weakens handoff quality.

Decision

Adopt and preserve a multi-surface hybrid execution model. Contributors may choose the best surface per task stage, while preserving a durable handoff contract and normalizing execution context into GitHub artifacts.

The framework explicitly supports:

  • VS Code Copilot (local)
  • GitHub Copilot Chat/Coding Agent (GitHub cloud)
  • GH CLI
  • GitHub Mobile
  • External AI agents (for discovery/synthesis, then normalization)

Consequences

  • Improves delivery flexibility while keeping governance consistency.
  • Requires explicit handoff packets (objective, context, constraints, acceptance criteria, validation, next owner).
  • Increases importance of normalization and artifact discipline across surfaces.
  • Prevents regression to single-surface or transcript-dependent execution.

Alternatives considered

  • Single-surface model: rejected because it does not match real operational workflows.
  • GitHub-only without external synthesis: rejected because discovery often starts outside repository context.
  • External-agent-led implementation without normalization: rejected due to control-plane and reviewability risks.

References

  • docs/gh-agents-and-automation.md
  • docs/multi-agent-handoff-playbook.md