Provide AI coding assistant support: AGENTS.md + Agent Skills

[linyuqing222]

Summary

Add two complementary AI development aids to the template:

1. A top-level AGENTS.md — always-on repo orientation.
2. One or more Agent Skills (SKILL.md folders) — on-demand,
   task-specific packages for the repeatable parts of add-on work.

Both are open formats supported across major coding agents, so the same files
work regardless of which assistant a contributor uses.

Motivation

A growing share of add-on developers, including blind and low-vision developers,
now work with AI coding assistants. NVDA add-on conventions are niche
and thinly documented compared with mainstream frameworks,
so these tools can't be assumed to have reliable knowledge of them.
Without project-specific guidance they are likely to guess, producing code
that needs rework before it builds or passes review.
The areas where guidance would help most:

• manifest.ini fields and buildVars.py configuration
• The scons-based build and packaging step
• The gettext localization flow
• Placement and structure of globalPlugins vs appModules
• minimumNVDAVersion / lastTestedNVDAVersion and API compatibility
• Add-on Store submission requirements

The template already exists to lower the barrier to starting an add-on.
Shipping maintained AI aids extends that same benefit to the now-common case
of developing with an assistant.

Why both, and how they divide

AGENTS.md and skills are complementary, not alternatives:

• AGENTS.md loads as soon as the agent opens the repo.
  It answers "how does this repo work and what must I not break."
• Skills load only when relevant and can bundle scripts and templates.
  They answer "do this specific repeatable task correctly."

Proposal

AGENTS.md (orientation):

• Repo layout and what each key file does (manifest.ini, buildVars.py,
  sconstruct, addon/, etc.)
• How to build and package the add-on
• API version fields and how to set them
• Conventions/constraints not to violate (e.g. avoid global plugins that run
  on every key press, for performance)
• Pointers to the NVDA Add-on Development Guide and Add-on Store submission docs

Skills (candidates, one SKILL.md folder each):

• Scaffold a globalPlugin / appModule
• Run the gettext localization workflow
• Build and package into a .nvda-addon
• Generate Add-on Store submission metadata

[Keesan12]

AGENTS.md support would help a lot if it stays practical and short.

The highest-value sections are usually: project layout, build/test commands, generated files to avoid, release/public-surface rules, and verification expectations. Long agent docs get ignored by humans and compressed weirdly by agents; a small set of sharp rules tends to work better.

[makhlwf]

I don't think this is needed. You just need to point the agent to the README file and the developer guide, and this is really sufficient.