overkill 0.5.4__tar.gz → 0.6.1__tar.gz

{overkill-0.5.4 → overkill-0.6.1}/.gitignore

@@ -8,6 +8,9 @@
 # Legacy directory (pre-migration)
 .review-loop/
 
+# Environment
+.env
+
 # Python
 __pycache__/
 *.py[cod]

{overkill-0.5.4 → overkill-0.6.1}/AGENTS.md

@@ -49,6 +49,7 @@ git log HEAD..origin/<target-branch> --oneline
 2. **Release branch → main**
    - Create `release/x.y.z` from `develop`
    - Bump version in `pyproject.toml` (`[project] version`)
+   - Run `uv lock` to sync `uv.lock` with the new version
    - Commit: `chore: bump version to x.y.z`
    - Push and open PR targeting `main`
    - Run review loop — if the review produces fix commits, merge them into `develop` first (via PR) before merging the release PR into `main`

{overkill-0.5.4 → overkill-0.6.1}/PKG-INFO

@@ -1,7 +1,7 @@
 Metadata-Version: 2.4
 Name: overkill
-Version: 0.5.4
-Summary: AI-powered code review loop — automates Codex review + Claude fix cycles
+Version: 0.6.1
+Summary: AI-powered code review loop — automates Codex/Gemini review + Claude fix cycles
 Project-URL: Repository, https://github.com/modocai/mr-overkill
 Project-URL: Issues, https://github.com/modocai/mr-overkill/issues
 Author: ModocAI
@@ -73,10 +73,11 @@ git clone --depth 1 https://github.com/modocai/mr-overkill.git /tmp/overkill \
 
 You need these to participate in the madness:
 
-**Accounts** (yes, you need both — that's the point):
+**Accounts** (yes, you need all three — that's the point):
 
 - [OpenAI](https://platform.openai.com/) account (paid plan) — because free tier is for weak code
 - [Anthropic](https://console.anthropic.com/) account (Pro/Max plan or API credits) — because Claude needs to think *deeply* about your variable names
+- [Google AI](https://aistudio.google.com/) account (API key) — because Gemini wants in on the overkill too
 
 **Runtime**:
 
@@ -87,8 +88,9 @@ You need these to participate in the madness:
 **CLI Tools**:
 
 ```bash
-npm install -g @openai/codex        # Codex CLI
+npm install -g @openai/codex             # Codex CLI
 npm install -g @anthropic-ai/claude-code  # Claude Code CLI
+npm install -g @google/gemini-cli         # Gemini CLI
 ```
 
 - [jq](https://jqlang.github.io/jq/) — JSON processor

{overkill-0.5.4 → overkill-0.6.1}/README.md

@@ -50,10 +50,11 @@ git clone --depth 1 https://github.com/modocai/mr-overkill.git /tmp/overkill \
 
 You need these to participate in the madness:
 
-**Accounts** (yes, you need both — that's the point):
+**Accounts** (yes, you need all three — that's the point):
 
 - [OpenAI](https://platform.openai.com/) account (paid plan) — because free tier is for weak code
 - [Anthropic](https://console.anthropic.com/) account (Pro/Max plan or API credits) — because Claude needs to think *deeply* about your variable names
+- [Google AI](https://aistudio.google.com/) account (API key) — because Gemini wants in on the overkill too
 
 **Runtime**:
 
@@ -64,8 +65,9 @@ You need these to participate in the madness:
 **CLI Tools**:
 
 ```bash
-npm install -g @openai/codex        # Codex CLI
+npm install -g @openai/codex             # Codex CLI
 npm install -g @anthropic-ai/claude-code  # Claude Code CLI
+npm install -g @google/gemini-cli         # Gemini CLI
 ```
 
 - [jq](https://jqlang.github.io/jq/) — JSON processor

overkill-0.6.1/prompts/active/claude-fix-execute.prompt.md

@@ -0,0 +1,14 @@
+좋아, 네 판단대로 고쳐줘.
+
+- 네가 동의한 finding만 고쳐
+- SKIP하겠다고 한 finding 자체는 고치지 마. 단, opinion에서 같은 false positive 재발을 방지하는 작은 조치를 제안했으면 (주석 추가, @deprecated 표시, dead code 제거 등) 그건 해줘.
+- 최소한의 변경만 해. 기존 코드 스타일을 따라
+- 끝나면 아래 형식으로 요약해줘:
+
+```
+## Fix Summary
+
+- [FIXED] <finding title>: <변경 내용>
+- [SKIPPED] <finding title>: <SKIP 사유>
+- [SKIPPED] <finding title>: <SKIP 사유> → 예방 조치: <조치 내용>
+```

{overkill-0.5.4 → overkill-0.6.1}/prompts/active/claude-refactor-fix-execute.prompt.md

@@ -1,7 +1,7 @@
 좋아, 네 판단대로 리팩토링해줘.
 
 - 네가 동의한 finding만 고쳐
-- SKIP하겠다고 한 건 고치지 마
+- SKIP하겠다고 한 finding 자체는 고치지 마. 단, opinion에서 같은 false positive 재발을 방지하는 작은 조치를 제안했으면 (주석 추가, @deprecated 표시, dead code 제거 등) 그건 해줘.
 - refactoring_plan의 steps 순서대로 진행해
 - 각 step이 끝날 때마다 코드가 정상 동작하는 상태를 유지해
 - 최소한의 변경만 해. 기존 코드 스타일을 따라
@@ -12,6 +12,7 @@
 
 - [FIXED] <finding title>: <변경 내용>
 - [SKIPPED] <finding title>: <SKIP 사유>
+- [SKIPPED] <finding title>: <SKIP 사유> → 예방 조치: <조치 내용>
 ```
 
 ## Safety Guards

{overkill-0.5.4 → overkill-0.6.1}/pyproject.toml

@@ -1,7 +1,7 @@
 [project]
 name = "overkill"
-version = "0.5.4"
-description = "AI-powered code review loop — automates Codex review + Claude fix cycles"
+version = "0.6.1"
+description = "AI-powered code review loop — automates Codex/Gemini review + Claude fix cycles"
 readme = "README.md"
 license = { text = "MIT" }
 requires-python = ">=3.11"

{overkill-0.5.4 → overkill-0.6.1}/src/mr_overkill/agents.py

@@ -385,7 +385,7 @@ class GeminiReviewAgent(ReviewAgent):
         return retry_gemini_cmd(
             output_path,
             "Gemini review",
-            ["gemini", "--sandbox", "--approval-mode", "plan", "-p", "-"],
+            ["gemini", "--sandbox", "--approval-mode", "yolo", "-p", "-"],
             stdin=prompt_text,
             max_wait=config.retry_max_wait,
             initial_wait=config.retry_initial_wait,
@@ -441,7 +441,7 @@ class GeminiRefactorReviewAgent(ReviewAgent):
         return retry_gemini_cmd(
             output_path,
             "Gemini analysis",
-            ["gemini", "--sandbox", "--approval-mode", "plan", "-p", "-"],
+            ["gemini", "--sandbox", "--approval-mode", "yolo", "-p", "-"],
             stdin=prompt_text,
             max_wait=config.retry_max_wait,
             initial_wait=config.retry_initial_wait,

{overkill-0.5.4 → overkill-0.6.1}/src/mr_overkill/budget/claude.py

@@ -34,11 +34,14 @@ def _sum_usage(usage: dict[str, int]) -> float:
     )
 
 
-# Rough 5-hour token limits per tier (empirical estimates).
+# Approximate 5-hour token limits per tier.
+# Derived from comparing local weighted-token sums against OAuth utilization
+# percentages (e.g. 5.4M weighted tokens ≈ 60% → ~9-10M limit for max5).
+# These are best-effort estimates; the authoritative value comes from OAuth.
 _TIER_LIMITS: dict[str, int] = {
-    "pro": 1_000_000,
-    "max5": 5_000_000,
-    "max20": 20_000_000,
+    "pro": 2_000_000,
+    "max5": 10_000_000,
+    "max20": 40_000_000,
 }
 
 _TIER_MAP: dict[str, str] = {
@@ -129,11 +132,13 @@ def check_oauth() -> BudgetStatus | None:
             timeout=5,
         )
         if result.returncode != 0 or not result.stdout.strip():
+            logger.debug("OAuth: keychain access failed (rc=%d)", result.returncode)
             return None
 
         creds = json.loads(result.stdout.strip())
         token = creds.get("claudeAiOauth", {}).get("accessToken")
         if not token:
+            logger.debug("OAuth: no access token in credentials")
             return None
 
         resp = subprocess.run(
@@ -153,12 +158,25 @@ def check_oauth() -> BudgetStatus | None:
             timeout=15,
         )
         if resp.returncode != 0 or not resp.stdout.strip():
+            logger.debug("OAuth: API request failed (rc=%d)", resp.returncode)
             return None
 
         data = json.loads(resp.stdout)
+
+        # Check for API error response (e.g. rate_limit_error)
+        if "error" in data:
+            err = data["error"]
+            logger.warning(
+                "OAuth: API error — %s: %s",
+                err.get("type", "unknown"),
+                err.get("message", ""),
+            )
+            return None
+
         five_hour = data.get("five_hour", {})
         utilization = five_hour.get("utilization")
         if utilization is None:
+            logger.debug("OAuth: no utilization in response")
             return None
 
         tier = detect_tier()
@@ -179,7 +197,8 @@ def check_oauth() -> BudgetStatus | None:
         json.JSONDecodeError,
         KeyError,
         OSError,
-    ):
+    ) as exc:
+        logger.debug("OAuth: %s", exc)
         return None
 
 
@@ -243,6 +262,16 @@ def check_local(projects_dir: Path | None = None) -> BudgetStatus:
 
     pct = (int(total_tokens) * 100 // limit) if total_tokens > 0 and limit > 0 else 0
 
+    logger.info(
+        "Local estimate: %d%% (%d weighted tokens / %d limit, "
+        "tier=%s, %d deduplicated messages)",
+        pct,
+        int(total_tokens),
+        limit,
+        tier,
+        len(seen_ids),
+    )
+
     return BudgetStatus(
         five_hour_used_pct=pct,
         seven_day_used_pct=None,
@@ -311,6 +340,24 @@ def check_token_budget() -> BudgetStatus:
     result = check_oauth()
     if result is not None:
         _save_cache(result)
+        # Run local estimate for calibration only when debug logging is active;
+        # check_local() does a full rglob scan which is too expensive for the
+        # normal OAuth fast path.
+        if logger.isEnabledFor(logging.DEBUG):
+            try:
+                local = check_local()
+                if local.five_hour_used_pct is not None:
+                    logger.debug(
+                        "Budget calibration: oauth=%d%% local=%d%% "
+                        "(diff=%+dpp, %d weighted tokens, tier=%s)",
+                        result.five_hour_used_pct or 0,
+                        local.five_hour_used_pct,
+                        (local.five_hour_used_pct) - (result.five_hour_used_pct or 0),
+                        local.tokens_used,
+                        local.tier,
+                    )
+            except Exception:
+                logger.debug("Calibration check_local() failed", exc_info=True)
         return result
 
     cached = _load_cache()

{overkill-0.5.4 → overkill-0.6.1}/src/mr_overkill/budget_report.py

@@ -11,7 +11,7 @@ from mr_overkill.budget.claude import check_token_budget as claude_check
 from mr_overkill.budget.codex import check_token_budget as codex_check
 from mr_overkill.models import BudgetScope, BudgetStatus
 
-_HEADER = "\u2550" * 24  # ════════════════════════
+_HEADER = "\u2550" * 33  # ═════════════════════════════════
 
 
 def _fmt_pct(pct: int | None) -> str:
@@ -58,10 +58,17 @@ def _print_codex(status: BudgetStatus) -> None:
     print()
 
 
+def _print_gemini() -> None:
+    print("Gemini Budget")
+    print(_HEADER)
+    print("  No local budget data (API-key billing).")
+    print()
+
+
 def _print_scope_table(claude_st: BudgetStatus, codex_st: BudgetStatus) -> None:
     print("Scope Thresholds")
     print(_HEADER)
-    print("           Claude  Codex")
+    print("           Claude  Codex   Gemini")
     scopes = (
         BudgetScope.MICRO, BudgetScope.MODULE,
         BudgetScope.LAYER, BudgetScope.FULL,
@@ -72,7 +79,7 @@ def _print_scope_table(claude_st: BudgetStatus, codex_st: BudgetStatus) -> None:
             x_label = "GO" if budget_sufficient(scope, codex_st) else "NOGO"
         else:
             c_label = x_label = "—"
-        print(f"  {scope.value:<8} {c_label:<7} {x_label}")
+        print(f"  {scope.value:<8} {c_label:<7} {x_label:<7} —")
 
 
 def print_budget_report(*, json_mode: bool = False) -> int:
@@ -88,11 +95,13 @@ def print_budget_report(*, json_mode: bool = False) -> int:
         payload = {
             "claude": asdict(claude_st),
             "codex": asdict(codex_st),
+            "gemini": {"mode": "api-key", "note": "no local budget data"},
         }
         print(json.dumps(payload, indent=2))
         return 0
 
     _print_claude(claude_st)
     _print_codex(codex_st)
+    _print_gemini()
     _print_scope_table(claude_st, codex_st)
     return 0

{overkill-0.5.4 → overkill-0.6.1}/uv.lock

@@ -247,7 +247,7 @@ wheels = [
 
 [[package]]
 name = "overkill"
-version = "0.5.0"
+version = "0.6.1"
 source = { editable = "." }
 
 [package.dev-dependencies]

overkill-0.5.4/prompts/active/claude-fix-execute.prompt.md

@@ -1,13 +0,0 @@
-좋아, 네 판단대로 고쳐줘.
-
-- 네가 동의한 finding만 고쳐
-- SKIP하겠다고 한 건 고치지 마
-- 최소한의 변경만 해. 기존 코드 스타일을 따라
-- 끝나면 아래 형식으로 요약해줘:
-
-```
-## Fix Summary
-
-- [FIXED] <finding title>: <변경 내용>
-- [SKIPPED] <finding title>: <SKIP 사유>
-```