development-standards: one shared repo for AI agent rules

The setup draws a line between two ways instructions reach an AI: work delegated to an agent like Codex, and the generate button in VS Code. Those use different entry points, but both point back to the same source text — PR creation, for example, uses the same pull-request.instructions.md whether an agent or the button is asking.

The repository layout reflects that split. AGENTS.md is kept as a reference list rather than a place to write rule text, and .github/copilot-instructions.md only tells the reader to read AGENTS.md instead of duplicating the list. Rule bodies live under docs/rules/, while form structure sits in Issue forms and PR templates and writing style in the generate-button files.

Syncing is deliberate rather than automatic. The script is run first without --write to preview, then with --write to apply, and finally re-run to confirm zero additions, updates, or deletions. Some files are overwritten wholesale, while AGENTS.md and the Copilot guidance are updated only inside management markers, and VS Code settings are partially updated with Microsoft's jsonc-parser so project-specific keys survive. The stated aim is to return judgments from real work back into the source over time, so the same standards carry across projects and even across different agents.