AI-Assisted Workflow

This repository uses OpenSpec-first development with a small set of explicit AI instruction files.

Canonical guidance files

openspec/ — requirements, change intent, and task order
AGENTS.md — shared repo guidance for terminal copilots and Codex-style agents
CLAUDE.md — Claude Code guidance
.github/copilot-instructions.md — GitHub Copilot repository guidance

When these surfaces drift, fix the drift instead of inventing a fourth workflow.

Recommended flow

Read the affected capability specs in openspec/specs/.
Create or update one change under openspec/changes/<change-id>/.
Implement tasks in dependency order.
Run the relevant preset-driven validation commands.
Use /review before merge or after a major cleanup phase.
Archive the change when the work is complete.

Tool choices

LSP: clangd
Build/test entry points: cmake --preset=... and ctest --preset=...
Hooks: repository-managed .githooks/
Autopilot strategy: prefer a longer single session over /fleet unless the work is truly parallel

What to avoid

reintroducing GitBook / HonKit / .kiro assumptions
adding generic AI docs that duplicate existing instructions
adding heavy MCP or plugin layers without a repository-specific payoff