fix(skills): prevent empty responses after load_skill

The skill system instruction's rule 2 told the model to follow loaded
skill instructions "before replying to the user". Some models (notably
Gemini) read this as license to treat the load_skill tool call as the
entire turn and stop with no visible output, producing empty responses.
This was most acute for tool-heavy skills, whose next correct action
after load_skill is to call more tools rather than reply.

Two changes to _build_skill_system_instruction:
- Reword rule 2 to drop the "before replying" framing.
- Add rule 7 stating that load_skill only retrieves instructions and
  does NOT complete the turn; the model must continue in the same turn
  (calling whatever tools the skill requires) and never end with an
  empty response right after loading a skill.

Verified in a production environment: the empty-response rate for
Gemini after skill loading dropped substantially with this guidance.

Adds tests asserting the new prompt guarantees on both the default and
prefixed system instruction.

Author: lwangverizon

src/google/adk/tools/skill_toolset.py

@@ -80,9 +80,9 @@ def _build_skill_system_instruction(prefix: str | None = None) -> str:
       f'the `{p}load_skill` tool with `skill_name="<SKILL_NAME>"` to read '
       "its full instructions before proceeding.\n"
       "2. Once you have read the instructions, follow them exactly as "
-      "documented before replying to the user. For example, If the "
-      "instruction lists multiple steps, please make sure you complete all "
-      "of them in order.\n"
+      "documented. For example, if the instruction lists multiple steps, "
+      "please make sure you complete all of them in order, and only then "
+      "reply to the user.\n"
       f"3. The `{p}load_skill_resource` tool is for viewing files within a "
       "skill's directory (e.g., `references/*`, `assets/*`, `scripts/*`). "
       "It is ONLY for skill-bundled files — do NOT use it to access "
@@ -97,6 +97,11 @@ def _build_skill_system_instruction(prefix: str | None = None) -> str:
       f"6. If `{p}run_skill_script` returns an error (for example "
       f"`SCRIPT_NOT_FOUND`), do not retry the same script or guess a "
       "different script path. Report the error to the user and stop.\n"
+      f"7. Loading a skill only retrieves its instructions; it does NOT "
+      f"complete your turn. After a `{p}load_skill` call returns, continue "
+      "in the SAME turn: call whatever tools the skill's steps require "
+      "(search, data retrieval, render), then write your reply. Never end "
+      "your turn with an empty response right after loading a skill.\n"
   )
 
 

tests/unittests/tools/test_skill_toolset.py

@@ -1758,6 +1758,36 @@ def test_system_instruction_references_run_skill_script():
   )
 
 
+def test_system_instruction_marks_load_skill_as_non_terminal():
+  """Rule 7 must tell the model load_skill does not complete the turn.
+
+  Without it, some models (notably Gemini) treat the load_skill tool call as
+  the entire turn and stop with no visible output, producing empty responses.
+  """
+  instruction = skill_toolset.DEFAULT_SKILL_SYSTEM_INSTRUCTION
+  assert "does NOT complete your turn" in instruction
+  assert "empty response" in instruction
+
+
+def test_system_instruction_rule_2_avoids_before_replying_framing():
+  """Rule 2 must not say to follow instructions "before replying".
+
+  That framing is read as license to end the turn after load_skill, which is
+  the original trigger for the empty-response regression.
+  """
+  assert (
+      "before replying"
+      not in skill_toolset.DEFAULT_SKILL_SYSTEM_INSTRUCTION
+  )
+
+
+def test_prefixed_system_instruction_includes_continue_after_load_rule():
+  """The prefixed builder variant must also carry rule 7 (with the prefix)."""
+  instruction = skill_toolset._build_skill_system_instruction(prefix="my")
+  assert "does NOT complete your turn" in instruction
+  assert "my_load_skill" in instruction
+
+
 # ── Finding 2: empty files are mounted (not silently dropped) ──