todo1: Report adopted Harness learnings before every agent push #5

Goal

Before any agent pushes changes in this repository, show the user the Harness improvements actually adopted in that work and the concrete evidence behind each one. If none were adopted, state "none" explicitly. Do not present unreviewed audit suggestions as learned or adopted improvements.

Settled decisions

• This applies to every agent-initiated push in this repository, not only $ai-workflow delivery.
• Put the short, always-applicable rule in root AGENTS.md.
• Add supporting guidance to relevant repository Skills for identifying adopted improvements and reporting evidence.
• Do not require make harness-audit on every push; it is an optional, trusted-checkout review tool.
• This is a separate follow-up to the merged AI workflow harness integration in Complete AI workflow harness integration #4.

Acceptance criteria

• AGENTS.md requires an agent to report adopted Harness improvements and evidence before each push, or explicitly report that there were none.
• Relevant repository Skills explain how to distinguish an implemented, evidence-backed improvement from a suggestion that has not been adopted.
• The rule covers agent pushes outside $ai-workflow as well as its Delivery Gate.
• Documentation does not imply this is a Git hook or that it blocks manual human pushes.
• make verify passes.

Out of scope

• Automatically running make harness-audit before every push.
• Automatically applying audit suggestions.
• Blocking manual human pushes with a Git hook.