Restore assign-to-agent (copilot) after semantics clarification

Re-add the assign-to-agent safe output (name copilot, allowed [copilot],
max 1, target triggering, issue-intent true) and the outcome-only prompt line
suggesting Copilot for well-scoped, actionable issues. Taxonomy stays
runtime-discovered (no issue-type or label allow-lists, no hard-coded names),
and the prompt remains outcome-only with no intent or tool mechanics.

Compiled with stable gh-aw v0.83.1.

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Copilot-Session: c058b80f-b718-49c7-96e5-f336efcbdc34

Author: alondahari

.github/workflows/issue-triage.lock.yml

@@ -3,7 +3,8 @@ description: |
   Agentic issue triage for new and reopened issues. Reads the issue and its
   comments, classifies the issue type when it is unset, adds relevant existing
   labels, checks for likely duplicate or related issues, asks the author for
-  missing information only when needed. Keeps agent-generated comments to a
+  missing information only when needed, and flags well-scoped issues that could
+  be handed to the Copilot coding agent. Keeps agent-generated comments to a
   minimum and never closes issues.
 
 on:
@@ -24,6 +25,12 @@ safe-outputs:
     issue-intent: true
   add-comment:
     max: 1
+  assign-to-agent:
+    name: copilot
+    allowed: [copilot]
+    max: 1
+    target: triggering
+    issue-intent: true
 
 tools:
   github:
@@ -53,6 +60,8 @@ Work toward the following outcomes:
   labels over many uncertain ones.
 - **Duplicates and related issues:** when the issue clearly duplicates or relates to an
   existing one, note that connection for maintainers. Never close an issue.
+- **Ready for Copilot:** when an issue is well-scoped and actionable, suggest assigning
+  it to the Copilot coding agent.
 
 Keep noise low: do not post a routine triage summary. When nothing needs to change,
 make no change.