Tune general task execution guidance

src/context_system/prompt_assembly.py

@@ -883,12 +883,29 @@ def _emit_group(
     "the error, check your assumptions, try a focused fix. Don't retry "
     "the identical action blindly, but don't abandon a viable approach "
     "after a single failure either.\n"
+    "- For ambiguous bugs, failures, or unfamiliar data, form a small set "
+    "of plausible hypotheses from the available evidence and rank them by "
+    "likelihood and the cost of checking them. Start with the cheapest "
+    "check that can distinguish among the leading hypotheses. Broaden the "
+    "investigation only when results rule them out or reveal new evidence; "
+    "do not replace focused diagnosis with an exhaustive search of the "
+    "environment.\n"
     "- Before finishing, audit the result against every explicit requirement "
     "in the user's request and verify the requirements that can be checked. "
     "Do not treat producing a plausible result as proof that the task is "
     "complete. When the user asks for all, every, multiple, or an exhaustive "
     "set of results, actively search for additional valid results instead of "
     "stopping after the first one.\n"
+    "- Use the narrowest sufficient verification for the change. Prefer an "
+    "existing focused check, add a regression test when it provides lasting "
+    "value, and run broader checks when the scope or risk warrants them. "
+    "Do not create multiple temporary tests or repeat equivalent checks once "
+    "the same behavior has been established.\n"
+    "- Stop when the explicit requirements are satisfied and proportionate "
+    "verification has passed. Do not re-run passing checks or expand into "
+    "unrequested audits without concrete evidence of an unresolved "
+    "requirement, a new failure, or material risk. If verification is not "
+    "possible, state that clearly instead of continuing speculative work.\n"
     "- Be cautious not to introduce security vulnerabilities such as "
     "command injection, XSS, SQL injection, and other OWASP top 10 "
     "vulnerabilities.\n"
@@ -905,7 +922,12 @@ def _emit_group(
     "- Don't create helpers, utilities, or abstractions for one-time "
     "operations. Don't design for hypothetical future requirements. "
     "The right amount of complexity is what the task actually requires. "
-    "Three similar lines of code is better than a premature abstraction."
+    "Three similar lines of code is better than a premature abstraction.\n"
+    "- When the same well-understood mechanical change is needed in multiple "
+    "places, inspect the affected scope and make a coherent grouped edit "
+    "when it is safe and reviewable. Avoid a long sequence of tiny edits "
+    "with equivalent verification after each one. Keep unrelated or "
+    "semantically distinct changes separate."
 )
 
 # Module 4: Cautious operations

tests/test_system_prompt_full.py

@@ -65,6 +65,28 @@ def test_task_prompt_requires_requirement_audit_and_exhaustive_results(self):
         assert "all, every, multiple, or an exhaustive set" in prompt
         assert "stopping after the first one" in prompt
 
+    def test_task_prompt_prioritizes_evidence_and_cheap_discriminating_checks(self):
+        prompt = build_full_system_prompt(use_cache=False)
+        assert "small set of plausible hypotheses" in prompt
+        assert "cheapest check that can distinguish" in prompt
+        assert "Broaden the investigation only when results rule them out" in prompt
+        assert "exhaustive search of the environment" in prompt
+
+    def test_task_prompt_uses_proportionate_verification_and_stops(self):
+        prompt = build_full_system_prompt(use_cache=False)
+        assert "narrowest sufficient verification" in prompt
+        assert "Do not create multiple temporary tests" in prompt
+        assert "Stop when the explicit requirements are satisfied" in prompt
+        assert "Do not re-run passing checks" in prompt
+        assert "material risk" in prompt
+
+    def test_task_prompt_groups_related_mechanical_edits(self):
+        prompt = build_full_system_prompt(use_cache=False)
+        assert "same well-understood mechanical change" in prompt
+        assert "coherent grouped edit" in prompt
+        assert "long sequence of tiny edits" in prompt
+        assert "Keep unrelated or semantically distinct changes separate" in prompt
+
     def test_identity_prompt_backward_compat(self):
         """_IDENTITY_PROMPT is an alias for _INTRO_SECTION."""
         assert _IDENTITY_PROMPT is _INTRO_SECTION