Adopt AGENTS.md as the canonical AI brief; wire all harnesses to it

[ChrisonSimtian]

Why

Multiple maintainers use different AI tools — Chris uses Claude Code, Dennis uses GitHub Copilot, and future contributors may use Aider, Cursor, Continue, Codex, or whatever appears next. Each tool reads from a different default file (CLAUDE.md, .github/copilot-instructions.md, etc.), which means the same instructions get duplicated and drift apart.

The emerging community convention (started by OpenAI Codex, adopted by Aider and others, increasingly supported across tools) is a single AGENTS.md at the repo root as the canonical source of truth. Tool-specific files become thin pointers that import or reference the canonical doc.

Action

• Create AGENTS.md at the repo root containing the canonical brief — what the project is, how the build works, conventions, do-not-do list, useful pointers (i.e. the content currently in CLAUDE.md)
• Replace CLAUDE.md with a thin pointer (one-line "see AGENTS.md") or symlink, depending on what Claude Code respects on Windows + Unix
• Add .github/copilot-instructions.md as a thin pointer to AGENTS.md
• Document the convention in CONTRIBUTING.md so any contributor using any tool knows where to look
• Verify the round-trip with both Claude and Copilot picking up the same brief
• Establish a discipline: any future tool-specific instruction file in this repo must be a pointer, not a copy

Coordinate with

• Architecture brief issue — that doc will be cross-linked from AGENTS.md
• Data-handling policy issue — same
• Conventions RFC issue — outcome of that discussion lands in AGENTS.md

Done when

• One canonical AGENTS.md exists at the repo root
• Every harness-specific instruction file is a thin pointer, not a duplicate
• Both Claude Code and Copilot are confirmed to read from it
• CONTRIBUTING.md tells contributors where the brief lives