feat(dashboard): página Ejecutar con trace timeline y violations rendering

- components/trace_view.render_trace: decision_path como timeline con duración.
- components/violation_view.render_violations: violaciones con badge de severidad.
- pages/2_▶️_Ejecutar.py: selector de agente, botones de escenarios pregrabados,
  textarea, invoke y render de status/output/violations/traza. Aviso si queda en
  awaiting_approval. (Se omite la variable muerta chosen_example del plan.)

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

dashboard/src/agentforge_dashboard/components/trace_view.py

@@ -0,0 +1,24 @@
+"""Renderiza el decision_path como timeline."""
+
+from __future__ import annotations
+
+import streamlit as st
+
+
+def render_trace(decision_path: list[dict]) -> None:
+    if not decision_path:
+        st.info("Aún no hay traza disponible.")
+        return
+    st.subheader("📜 Decision path")
+    for step in decision_path:
+        with st.container(border=True):
+            cols = st.columns([2, 1, 1])
+            with cols[0]:
+                st.markdown(f"**{step['step']}**")
+            with cols[1]:
+                st.caption(step.get("timestamp", "—"))
+            with cols[2]:
+                st.metric("ms", step.get("duration_ms", 0), label_visibility="collapsed")
+            if step.get("detail"):
+                with st.expander("detalle"):
+                    st.json(step["detail"])

dashboard/src/agentforge_dashboard/components/violation_view.py

@@ -0,0 +1,26 @@
+"""Renderiza violaciones de guardrails con badges de severidad."""
+
+from __future__ import annotations
+
+import streamlit as st
+
+_SEV_BADGE = {
+    "info": "🟦",
+    "warning": "🟧",
+    "block": "🟥",
+}
+
+
+def render_violations(violations: list[dict]) -> None:
+    if not violations:
+        st.success("Sin violaciones.")
+        return
+    st.subheader("🛡️ Violaciones de guardrails")
+    for v in violations:
+        with st.container(border=True):
+            badge = _SEV_BADGE.get(v["severity"], "⬜")
+            st.markdown(
+                f"{badge} **{v['validator']}** — `{v['stage']}` — "
+                f"severity=`{v['severity']}` — blocked=`{v['blocked']}`"
+            )
+            st.caption(v["message"])

dashboard/src/agentforge_dashboard/pages/2_▶️_Ejecutar.py

@@ -0,0 +1,72 @@
+"""Página para invocar un agente y visualizar la traza completa."""
+
+from __future__ import annotations
+
+from pathlib import Path
+
+import streamlit as st
+
+from agentforge_dashboard.client import CoreClient
+from agentforge_dashboard.components.trace_view import render_trace
+from agentforge_dashboard.components.violation_view import render_violations
+
+
+@st.cache_resource
+def get_client() -> CoreClient:
+    return CoreClient()
+
+
+client = get_client()
+st.title("▶️ Ejecutar Agente")
+st.caption("Lanza una ejecución con la cadena completa de gobierno.")
+
+agents = client.list_agents()
+if not agents:
+    st.warning("No hay agentes registrados.")
+    st.stop()
+
+names = [a["name"] for a in agents]
+agent_name = st.selectbox("Agente", names)
+
+# Cargar escenarios pregrabados si existen en el disco montado.
+examples_dir = Path("agents") / agent_name / "examples"
+example_files = sorted(examples_dir.glob("*.txt")) if examples_dir.exists() else []
+
+col_l, col_r = st.columns([3, 1])
+with col_r:
+    st.markdown("**Escenarios**")
+    for ef in example_files:
+        if st.button(ef.stem, use_container_width=True):
+            st.session_state["input_text"] = ef.read_text(encoding="utf-8")
+
+with col_l:
+    user_input = st.text_area(
+        "Descripción del incidente",
+        height=240,
+        key="input_text",
+        placeholder="Pega aquí la descripción del incidente o usa un escenario...",
+    )
+
+if st.button("🚀 Invocar agente", type="primary", disabled=not user_input):
+    with st.spinner("Ejecutando..."):
+        result = client.invoke_agent(agent_name, {"input": user_input})
+    st.session_state["last_execution"] = result
+
+execution = st.session_state.get("last_execution")
+if execution and "error" not in execution:
+    st.divider()
+    st.markdown(f"**trace_id:** `{execution['trace_id']}`")
+    st.markdown(f"**Status:** `{execution['status']}`")
+
+    if execution["status"] == "awaiting_approval":
+        st.warning(
+            "Esta ejecución requiere aprobación humana. "
+            "Ve a la página **Aprobaciones** para revisar y decidir."
+        )
+    if execution.get("final_output"):
+        st.subheader("📦 Output final")
+        st.json(execution["final_output"])
+    render_violations(execution.get("violations", []))
+    render_trace(execution.get("decision_path", []))
+elif execution and "error" in execution:
+    st.error(f"Error de la API: {execution['error']}")