feat(community/agents): add AL agent quality guidance

- add 20 agent knowledge rules with good and bad AL samples
- clarify setup dialog shape, temporary persistence, permissions, profiles, instructions, capability registration, and interface wiring
- add the community-owned AL agents review skill
- make review fixture discovery layer-aware with custom, community, and Microsoft precedence
- document layer-aware evaluation behavior

community/knowledge/agents/use-documented-instruction-keywords.md

@@ -0,0 +1,30 @@
+---
+bc-version: [27..]
+domain: agents
+keywords: [instruction-keywords, request-a-review, memorize, write-an-email, invoke-action]
+technologies: [al]
+countries: [w1]
+application-area: [all]
+---
+
+# Use the toolkit instruction keywords for review, mail, and memory
+
+## Description
+
+The agent runtime looks for specific phrases: ask for assistance, request a review, reply, write an email, memorize, set a field, use lookup, invoke an action. Ordinary English such as get a human to look or remember this is weaker. Outbound reply and email always require review; that is platform policy, not optional tone.
+
+## Best Practice
+
+In the instruction resource, use those keywords at the decision points: request a review before posting; write an email only after stating that outbound mail is reviewed; memorize values the later steps need. Pair `Reply` / `Write an email` with an explicit review sentence.
+
+See sample: `use-documented-instruction-keywords.good.al`.
+
+## Anti Pattern
+
+Inventing tool-like verbs (call Copilot, click Post_Promoted) or omitting request a review before posting. Detection signal: instruction text that says email the customer with no review keyword.
+
+See sample: `use-documented-instruction-keywords.bad.al`.
+
+## See also
+
+`instruction-structure-is-role-rules-steps.md` defines the containing document structure, while `instructions-describe-work-not-tool-ids.md` keeps outcomes independent of UI tool identifiers.