argus-code-review 0.2.3__tar.gz → 0.2.4__tar.gz

{argus_code_review-0.2.3 → argus_code_review-0.2.4}/CHANGELOG.md

@@ -7,6 +7,21 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [0.2.4] - 2026-09-17
+
+### Changed
+
+- Decoupled the Gemini reviewer's turn budget from the shared
+  `argus.runners._MAX_TURNS` constant (TECH-6453). `argus/gemini_runner.py`
+  now has its own `_MAX_TURNS_GEMINI = 45` (30 \* 1.5), independent of the
+  Claude Agent SDK and OpenAI runner paths, which remain at `_MAX_TURNS = 30`
+  unchanged. Gemini-backed bulk reviewer sessions were exhausting the
+  previously-shared 30-turn budget before calling `finish_review`,
+  truncating exploration; raising only Gemini's budget avoids widening the
+  cost/blast-radius ceiling on the more expensive Claude-Opus-tier
+  cross-cutting/blocking-validator/feedback-verifier roles, which showed no
+  evidence of needing more turns.
+
 ## [0.2.3] - 2026-09-14
 
 ### Added
@@ -244,7 +259,8 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
   packaged set.
 - `argus --version`, `argus prompts list`, and `argus prompts export`.
 
-[Unreleased]: https://github.com/redesignhealth/argus-review/compare/v0.2.3...HEAD
+[Unreleased]: https://github.com/redesignhealth/argus-review/compare/v0.2.4...HEAD
+[0.2.4]: https://github.com/redesignhealth/argus-review/compare/v0.2.3...v0.2.4
 [0.2.3]: https://github.com/redesignhealth/argus-review/compare/v0.2.2...v0.2.3
 [0.2.2]: https://github.com/redesignhealth/argus-review/compare/v0.2.1...v0.2.2
 [0.2.1]: https://github.com/redesignhealth/argus-review/compare/v0.2.0...v0.2.1

{argus_code_review-0.2.3 → argus_code_review-0.2.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: argus-code-review
-Version: 0.2.3
+Version: 0.2.4
 Summary: Self-orchestrated PR review agent using LangGraph + Claude Agent SDK
 Project-URL: Repository, https://github.com/redesignhealth/argus-review
 Project-URL: Issues, https://github.com/redesignhealth/argus-review/issues

{argus_code_review-0.2.3 → argus_code_review-0.2.4}/argus/gemini_runner.py

@@ -40,8 +40,9 @@ then loop ``client.aio.models.generate_content(...)`` -- executing every
 function call the model requests each turn (there can be more than
 one), feeding all of their results back as a single follow-up turn --
 until the model stops requesting function calls, calls
-``finish_review``, or the turn budget (``argus.runners._MAX_TURNS``, the
-same constant the Claude path uses) is exhausted. Findings arrive via
+``finish_review``, or the turn budget (``_MAX_TURNS_GEMINI``, this
+module's own budget, independent of the Claude/OpenAI paths' shared
+``argus.runners._MAX_TURNS``) is exhausted. Findings arrive via
 ``report_finding`` tool calls into ``review_tools``' per-session sink,
 not as a JSON blob embedded in the model's own text -- so, to keep this
 task's blast radius contained to this file plus ``argus.bench``'s
@@ -158,7 +159,6 @@ from argus.gemini_cache import GeminiCacheKeeper
 from argus.llm.models import estimate_cost_usd
 from argus.llm.models import resolve as resolve_model_alias
 from argus.runners import (
-    _MAX_TURNS,
     _SUBPROCESS_TIMEOUT_S,
     SessionResult,
     _resolve_repo_root,
@@ -166,6 +166,8 @@ from argus.runners import (
 
 logger = logging.getLogger(__name__)
 
+_MAX_TURNS_GEMINI = 45  # 30 * 1.5 — see TECH-6453
+
 # Both "auto" and "on" attempt explicit caching today -- there is no
 # separate heuristic distinguishing them yet (Track 1 defined the
 # `caching` values without one; see argus.bench's `_VALID_CACHING`). The
@@ -687,7 +689,7 @@ async def _run_turns(
         usage_tool_use_prompt_total = 0
 
         with review_tools.review_session(repo_root) as findings_sink:
-            for _turn in range(_MAX_TURNS):
+            for _turn in range(_MAX_TURNS_GEMINI):
                 try:
                     response = await client.aio.models.generate_content(
                         model=model, contents=contents, config=config
@@ -817,13 +819,13 @@ async def _run_turns(
                 if finished:
                     break
             else:
-                # `for...else`: only reached if every one of _MAX_TURNS
+                # `for...else`: only reached if every one of _MAX_TURNS_GEMINI
                 # iterations executed a function call and none of them was
                 # finish_review -- i.e. the turn budget was exhausted.
                 logger.warning(
                     "Gemini session [%s] exhausted its %d-turn budget without a finish_review call",
                     label or "unlabeled",
-                    _MAX_TURNS,
+                    _MAX_TURNS_GEMINI,
                 )
 
             result_text = _build_result_text(findings_sink, files_explored)

{argus_code_review-0.2.3 → argus_code_review-0.2.4}/tests/test_gemini_runner.py

@@ -323,18 +323,19 @@ class TestMultipleSimultaneousFunctionCalls:
 
 class TestTurnBudgetExhaustion:
     async def test_exhausting_max_turns_still_returns_a_result(self) -> None:
-        from argus.runners import _MAX_TURNS
+        from argus.gemini_runner import _MAX_TURNS_GEMINI
 
         entry = _make_entry(caching="off")
         settings = _make_settings()
 
         # Every turn keeps requesting a function call, never finish_review --
-        # the loop must stop after _MAX_TURNS turns rather than looping forever.
+        # the loop must stop after _MAX_TURNS_GEMINI turns rather than looping
+        # forever.
         responses = [
             _make_response(
                 calls=[("report_finding", {"file": None, "line": None, "description": "x"})]
             )
-            for _ in range(_MAX_TURNS)
+            for _ in range(_MAX_TURNS_GEMINI)
         ]
         fake_client = _make_fake_client(responses)
 
@@ -348,9 +349,9 @@ class TestTurnBudgetExhaustion:
             )
 
         assert result.failure_reason is None
-        assert fake_client.aio.models.generate_content.await_count == _MAX_TURNS
+        assert fake_client.aio.models.generate_content.await_count == _MAX_TURNS_GEMINI
         payload = _parse_result_json(result.result_text)
-        assert len(payload["findings"]) == _MAX_TURNS
+        assert len(payload["findings"]) == _MAX_TURNS_GEMINI
 
 
 # ---------------------------------------------------------------------------