fix(4.3): revert active_skills behavioral change per GPT-5.4 buyoff

Extraction must preserve original behavior. The fallback path reports
injected skills in active_skills (matching monolith behavior). Behavioral
improvement (distinguishing attempted vs active) tracked for future epic.

Added test assertion: skills_used reflects attempted skills on fallback.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Author: Brian Krafft

openspace/execution_engine.py

@@ -336,8 +336,7 @@ async def _execute_skill_first(
         execution_context_p2["max_iterations"] = max_iterations
 
         result = await self._grounding_agent.process(execution_context_p2)
-        result["active_skills"] = []
-        result["attempted_skills"] = injected_skill_ids
+        result["active_skills"] = injected_skill_ids
         logger.info(
             f"[Phase 2 — Fallback] {result.get('status', 'unknown')} "
             f"({result.get('iterations', 0)} iterations)"

tests/test_execution_engine.py

@@ -245,6 +245,8 @@ async def test_skill_failure_triggers_fallback(self, engine, mock_grounding_agen
         result = await engine.execute("test task")
         assert mock_grounding_agent.process.call_count == 2
         mock_grounding_agent.clear_skill_context.assert_called_once()
+        # Backward compat: skills_used reflects attempted skills even on fallback
+        assert result["skills_used"] == ["failing-skill"]
 
 
 # ---------------------------------------------------------------------------