Add Claude Opus 4.7, simplify pre-commit and LGTM workflow

CLAUDE.md

@@ -119,30 +119,27 @@ assert find_test_files("foo.ts", all_files, None) == ["foo.test.ts"]
 
 **CRITICAL: NEVER start without explicit user request. PR must be clean — don't ignore failures.**
 
-1. `git status` to see ALL changes
-2. `scripts/git/list_changed_files.sh` — store for staging reference
-3. Verify not on main: `git branch --show-current`
-4. `git fetch origin main && git merge origin/main`
-5. `git commit -m "descriptive message"` — user has already run `git add` before saying "lgtm"
+1. `git fetch origin main && git merge origin/main`
+2. `git commit -m "descriptive message"` — user has already run `git add` before saying "lgtm"
    - Pre-commit hook runs automatically (see `scripts/git/pre_commit_hook.sh`): pip-freeze, generate-types, black, ruff, print/logging checks, then pylint + pyright + pytest concurrently
    - Install: `ln -sf ../../scripts/git/pre_commit_hook.sh .git/hooks/pre-commit`
    - **If hooks fail**: fix, re-stage, commit again. Don't stage other sessions' files.
    - **`--no-verify`** only for trivial non-code changes
    - Unused mock params: `# pyright: reportUnusedVariable=false` at top
    - NO co-author lines or `[skip ci]`
-6. Check for existing PR: `gh pr list --head $(git branch --show-current) --state open` — if exists, **STOP and ask**
-7. `git push`
-8. `gh pr create --title "PR title" --body "" --assignee @me` — create PR immediately, no body
-9. Check recent posts: `scripts/git/recent_social_posts.sh gitauto` and `scripts/git/recent_social_posts.sh wes`
-10. `gh pr edit <number> --body "..."` — add summary and social posts after checking recent posts
+3. Check for existing PR: `gh pr list --head $(git branch --show-current) --state open` — if exists, **STOP and ask**
+4. `git push`
+5. `gh pr create --title "PR title" --body "" --assignee @me` — create PR immediately, no body
+6. Check recent posts: `scripts/git/recent_social_posts.sh gitauto` and `scripts/git/recent_social_posts.sh wes`
+7. `gh pr edit <number> --body "..."` — add summary and social posts after checking recent posts
     - Technical, descriptive title. **No `## Test plan`**.
     - **Two posts** (last section, customer-facing only): GitAuto (changelog) + Wes (personal voice, don't emphasize "GitAuto")
     - Format: `## Social Media Post (GitAuto)` and `## Social Media Post (Wes)` headers (parsed by `extract-social-posts.js`)
     - **GitAuto post**: Changelog format — one-liner headline + change bullets. No storytelling.
     - **Wes post**: Honest stories. Vary openers — check recent posts first.
     - Guidelines: No em dashes (—). Under 280 chars. No marketing keywords. No negative framing. No internal names. No small numbers — use relative language.
-11. If Sentry issue: `python3 scripts/sentry/get_issue.py AGENT-XXX` then `python3 scripts/sentry/resolve_issue.py AGENT-XXX ...`
-12. **Blog post** in `../website/app/blog/posts/`:
+8. If Sentry issue: `python3 scripts/sentry/get_issue.py AGENT-XXX` then `python3 scripts/sentry/resolve_issue.py AGENT-XXX ...`
+9. **Blog post** in `../website/app/blog/posts/`:
     - `YYYY-MM-DD-kebab-case-title.mdx`. Universal dev lesson, not GitAuto internals (exception: deep technical content).
     - **Skip if lesson is thin** — argue back if no real insight.
     - `metadata.title`: **34-44 chars** (layout appends `- GitAuto Blog` for 50-60 total). Verify no duplicate slug.
@@ -172,7 +169,7 @@ assert find_test_files("foo.ts", all_files, None) == ["foo.test.ts"]
       - Unsplash API: `source .env && curl "https://api.unsplash.com/search/photos?query=QUERY&orientation=landscape&client_id=$UNSPLASH_ACCESS_KEY"`, download with `?w=1200&h=630&fit=crop&crop=entropy`
       - Convert to PNG: `sips -s format png downloaded.jpg --out ../website/public/og/blog/{slug}.png`
       - Dev.to crops to 1000x420 — keep important content centered.
-13. **Docs page** in `../website/app/docs/`: Create new or update existing. Browse for best-fit category. New pages: 3 files (`page.tsx`, `layout.tsx`, `jsonld.ts`).
+10. **Docs page** in `../website/app/docs/`: Create new or update existing. Browse for best-fit category. New pages: 3 files (`page.tsx`, `layout.tsx`, `jsonld.ts`).
 
 ## CRITICAL: Fixing Foxquilt PRs
 

constants/claude.py

@@ -7,6 +7,7 @@
 
 # https://platform.claude.com/docs/en/docs/about-claude/models/all-models#model-comparison-table
 CONTEXT_WINDOW: dict[ClaudeModelId, int] = {
+    ClaudeModelId.OPUS_4_7: 1_000_000,
     ClaudeModelId.OPUS_4_6: 1_000_000,
     ClaudeModelId.SONNET_4_6: 1_000_000,
     ClaudeModelId.OPUS_4_5: 200_000,
@@ -15,6 +16,7 @@
 }
 
 MAX_OUTPUT_TOKENS: dict[ClaudeModelId, int] = {
+    ClaudeModelId.OPUS_4_7: 128_000,
     ClaudeModelId.OPUS_4_6: 128_000,
     ClaudeModelId.SONNET_4_6: 64_000,
     ClaudeModelId.OPUS_4_5: 64_000,

constants/models.py

@@ -5,6 +5,7 @@
 class ClaudeModelId(StrEnum):
     """Claude models — user-selectable and fallback-only."""
 
+    OPUS_4_7 = "claude-opus-4-7"
     OPUS_4_6 = "claude-opus-4-6"
     SONNET_4_6 = "claude-sonnet-4-6"
     OPUS_4_5 = "claude-opus-4-5"
@@ -41,9 +42,9 @@ class ModelInfo(TypedDict):
 
 MODEL_REGISTRY: dict[ModelId, ModelInfo] = {
     # Claude (user-selectable)
-    ClaudeModelId.OPUS_4_6: ModelInfo(
+    ClaudeModelId.OPUS_4_7: ModelInfo(
         provider=ModelProvider.CLAUDE,
-        display_name="Claude Opus 4.6",
+        display_name="Claude Opus 4.7",
         credit_cost_usd=8,
         user_selectable=True,
         free_tier=False,
@@ -56,6 +57,13 @@ class ModelInfo(TypedDict):
         free_tier=True,
     ),
     # Claude (fallback-only, same cost as their newer versions)
+    ClaudeModelId.OPUS_4_6: ModelInfo(
+        provider=ModelProvider.CLAUDE,
+        display_name="Claude Opus 4.6",
+        credit_cost_usd=8,
+        user_selectable=False,
+        free_tier=False,
+    ),
     ClaudeModelId.OPUS_4_5: ModelInfo(
         provider=ModelProvider.CLAUDE,
         display_name="Claude Opus 4.5",
@@ -100,6 +108,6 @@ class ModelInfo(TypedDict):
     m for m, r in MODEL_REGISTRY.items() if r["user_selectable"] and r["free_tier"]
 ]
 DEFAULT_FREE_MODEL = GoogleModelId.GEMMA_4_31B
-DEFAULT_PAID_MODEL = ClaudeModelId.OPUS_4_6
+DEFAULT_PAID_MODEL = ClaudeModelId.OPUS_4_7
 MAX_CREDIT_COST_USD = max(entry["credit_cost_usd"] for entry in MODEL_REGISTRY.values())
 CREDIT_GRANT_AMOUNT_USD = MAX_CREDIT_COST_USD * 3

constants/test_models.py

@@ -64,3 +64,19 @@ def test_anthropic_models_exist():
         m for m, r in MODEL_REGISTRY.items() if r["provider"] == ModelProvider.CLAUDE
     ]
     assert len(anthropic_models) >= 1
+
+
+def test_opus_47_is_user_selectable():
+    info = MODEL_REGISTRY[ClaudeModelId.OPUS_4_7]
+    assert info["user_selectable"] is True
+    assert info["credit_cost_usd"] == 8
+
+
+def test_opus_46_is_fallback_only():
+    info = MODEL_REGISTRY[ClaudeModelId.OPUS_4_6]
+    assert info["user_selectable"] is False
+    assert info["credit_cost_usd"] == 8
+
+
+def test_default_paid_model_is_opus_47():
+    assert DEFAULT_PAID_MODEL == ClaudeModelId.OPUS_4_7

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "GitAuto"
-version = "1.6.6"
+version = "1.6.10"
 requires-python = ">=3.14"
 dependencies = [
     "annotated-doc==0.0.4",

services/claude/chat_with_claude.py

@@ -14,7 +14,6 @@
 from services.claude.remove_outdated_file_edit_attempts import (
     remove_outdated_file_edit_attempts,
 )
-from services.claude.strip_strict_from_tools import strip_strict_from_tools
 from services.claude.trim_messages import trim_messages_to_token_limit
 from services.llm_result import LlmResult, ToolCall
 from services.supabase.llm_requests.insert_llm_request import insert_llm_request
@@ -43,14 +42,6 @@ def chat_with_claude(
         messages=messages, client=claude, model=model_id, max_input=max_input
     )
 
-    # Strip "strict" from tools for models that don't support it
-    if model_id not in (
-        ClaudeModelId.SONNET_4_6,
-        ClaudeModelId.SONNET_4_5,
-        ClaudeModelId.OPUS_4_6,
-    ):
-        tools = strip_strict_from_tools(tools)
-
     # https://docs.anthropic.com/en/api/messages
     start_time = time.time()
     try:

services/claude/evaluate_condition.py

@@ -46,8 +46,8 @@ def evaluate_condition(
         return EvaluationResult(False, "empty input")
 
     response = claude.beta.messages.create(
-        model=ClaudeModelId.OPUS_4_6,
-        max_tokens=MAX_OUTPUT_TOKENS[ClaudeModelId.OPUS_4_6],
+        model=ClaudeModelId.OPUS_4_7,
+        max_tokens=MAX_OUTPUT_TOKENS[ClaudeModelId.OPUS_4_7],
         temperature=0,
         system=system_prompt,
         messages=[{"role": "user", "content": content}],

services/claude/is_code_untestable.py

@@ -106,8 +106,8 @@ def is_code_untestable(
 Is this code dead (unreachable/redundant) or genuinely untestable (reachable at runtime but impossible to test)?"""
 
     response = claude.beta.messages.create(
-        model=ClaudeModelId.OPUS_4_6,
-        max_tokens=MAX_OUTPUT_TOKENS[ClaudeModelId.OPUS_4_6],
+        model=ClaudeModelId.OPUS_4_7,
+        max_tokens=MAX_OUTPUT_TOKENS[ClaudeModelId.OPUS_4_7],
         temperature=0,
         system=SYSTEM_PROMPT,
         messages=[{"role": "user", "content": content}],

services/claude/test_chat_with_claude.py

@@ -147,3 +147,40 @@ def test_chat_with_claude_calls_optimization_functions(
     )
 
     mock_remove_outdated_file_edit_attempts.assert_called_once()
+
+
+@patch("services.claude.chat_with_claude.insert_llm_request")
+@patch("services.claude.chat_with_claude.claude")
+def test_strict_tools_passed_through_unchanged(mock_claude, mock_insert_llm_request):
+    """Strict tools must not be stripped — all current models support strict."""
+    mock_response = Mock()
+    mock_response.content = [Mock(type="text", text="ok")]
+    mock_response.usage = Mock(output_tokens=5)
+    mock_insert_llm_request.return_value = {"total_cost_usd": 0.01}
+    mock_claude.messages.create.return_value = mock_response
+    mock_claude.messages.count_tokens.return_value = Mock(input_tokens=10)
+
+    tools = cast(
+        list[ToolUnionParam],
+        [
+            {
+                "name": "test_tool",
+                "description": "Test",
+                "strict": True,
+                "input_schema": {"type": "object", "properties": {}},
+            }
+        ],
+    )
+
+    chat_with_claude(
+        messages=cast(list[MessageParam], [{"role": "user", "content": "test"}]),
+        system_content="system",
+        tools=tools,
+        model_id=ClaudeModelId.HAIKU_4_5,
+        usage_id=999,
+        created_by="4:test-user",
+    )
+
+    call_args = mock_claude.messages.create.call_args
+    passed_tools = call_args.kwargs["tools"]
+    assert passed_tools[0]["strict"] is True

services/claude/test_evaluate_condition.py

@@ -66,6 +66,18 @@ def test_returns_evaluation_failed_on_invalid_json(self, mock_claude):
 
         assert result == EvaluationResult(False, "evaluation failed")
 
+    def test_uses_opus_47_model(self, mock_claude):
+        mock_response = MagicMock()
+        mock_response.content = [
+            MagicMock(text='{"result": true, "reason": "testable"}')
+        ]
+        mock_claude.beta.messages.create.return_value = mock_response
+
+        evaluate_condition(content="code", system_prompt="Check this.")
+
+        call_args = mock_claude.beta.messages.create.call_args
+        assert call_args.kwargs["model"] == "claude-opus-4-7"
+
     def test_uses_structured_output_schema(self, mock_claude):
         mock_response = MagicMock()
         mock_response.content = [

services/claude/test_evaluate_quality_checks.py

@@ -1,12 +1,12 @@
 # pyright: reportArgumentType=false
 from unittest.mock import MagicMock, patch
 
+from constants.claude import MAX_OUTPUT_TOKENS
 from constants.models import ClaudeModelId
 from services.claude.evaluate_quality_checks import evaluate_quality_checks
 
 
-@patch("services.claude.evaluate_quality_checks.claude")
-def test_uses_opus_model(mock_claude):
+def _mock_claude_call(mock_claude, model: ClaudeModelId):
     mock_content = MagicMock()
     mock_content.text = '{"business_logic": {}}'
     mock_claude.messages.create.return_value = MagicMock(content=[mock_content])
@@ -15,26 +15,49 @@ def test_uses_opus_model(mock_claude):
         source_content="const x = 1;",
         source_path="src/foo.ts",
         test_files=[("test/foo.spec.ts", "it('works', () => {})")],
-        model=ClaudeModelId.OPUS_4_6,
+        model=model,
     )
 
-    call_kwargs = mock_claude.messages.create.call_args.kwargs
-    assert call_kwargs["model"] == "claude-opus-4-6"
+    return mock_claude.messages.create.call_args.kwargs
 
 
 @patch("services.claude.evaluate_quality_checks.claude")
-def test_uses_max_tokens_matching_model(mock_claude):
-    mock_content = MagicMock()
-    mock_content.text = '{"business_logic": {}}'
-    mock_claude.messages.create.return_value = MagicMock(content=[mock_content])
+def test_opus_47_passes_model_and_max_tokens(mock_claude):
+    kwargs = _mock_claude_call(mock_claude, ClaudeModelId.OPUS_4_7)
+    assert kwargs["model"] == "claude-opus-4-7"
+    assert kwargs["max_tokens"] == MAX_OUTPUT_TOKENS[ClaudeModelId.OPUS_4_7]
 
-    evaluate_quality_checks(
-        source_content="const x = 1;",
-        source_path="src/foo.ts",
-        test_files=[("test/foo.spec.ts", "it('works', () => {})")],
-        model=ClaudeModelId.OPUS_4_6,
-    )
 
-    call_kwargs = mock_claude.messages.create.call_args.kwargs
-    # Opus 4.6 has 128_000 max tokens
-    assert call_kwargs["max_tokens"] == 128_000
+@patch("services.claude.evaluate_quality_checks.claude")
+def test_opus_46_passes_model_and_max_tokens(mock_claude):
+    kwargs = _mock_claude_call(mock_claude, ClaudeModelId.OPUS_4_6)
+    assert kwargs["model"] == "claude-opus-4-6"
+    assert kwargs["max_tokens"] == MAX_OUTPUT_TOKENS[ClaudeModelId.OPUS_4_6]
+
+
+@patch("services.claude.evaluate_quality_checks.claude")
+def test_sonnet_46_passes_model_and_max_tokens(mock_claude):
+    kwargs = _mock_claude_call(mock_claude, ClaudeModelId.SONNET_4_6)
+    assert kwargs["model"] == "claude-sonnet-4-6"
+    assert kwargs["max_tokens"] == MAX_OUTPUT_TOKENS[ClaudeModelId.SONNET_4_6]
+
+
+@patch("services.claude.evaluate_quality_checks.claude")
+def test_opus_45_passes_model_and_max_tokens(mock_claude):
+    kwargs = _mock_claude_call(mock_claude, ClaudeModelId.OPUS_4_5)
+    assert kwargs["model"] == "claude-opus-4-5"
+    assert kwargs["max_tokens"] == MAX_OUTPUT_TOKENS[ClaudeModelId.OPUS_4_5]
+
+
+@patch("services.claude.evaluate_quality_checks.claude")
+def test_sonnet_45_passes_model_and_max_tokens(mock_claude):
+    kwargs = _mock_claude_call(mock_claude, ClaudeModelId.SONNET_4_5)
+    assert kwargs["model"] == "claude-sonnet-4-5"
+    assert kwargs["max_tokens"] == MAX_OUTPUT_TOKENS[ClaudeModelId.SONNET_4_5]
+
+
+@patch("services.claude.evaluate_quality_checks.claude")
+def test_haiku_45_passes_model_and_max_tokens(mock_claude):
+    kwargs = _mock_claude_call(mock_claude, ClaudeModelId.HAIKU_4_5)
+    assert kwargs["model"] == "claude-haiku-4-5"
+    assert kwargs["max_tokens"] == MAX_OUTPUT_TOKENS[ClaudeModelId.HAIKU_4_5]

services/claude/test_is_code_untestable.py

@@ -28,6 +28,19 @@ def _set_mock_response(mock_claude, result: bool, category: str, reason: str):
     ]
 
 
+def test_uses_opus_47_model(mock_claude):
+    _set_mock_response(mock_claude, False, "testable", "testable")
+
+    is_code_untestable(
+        file_path="src/app.tsx",
+        file_content="const x = 1;",
+        uncovered_lines="1",
+    )
+
+    call_args = mock_claude.beta.messages.create.call_args
+    assert call_args.kwargs["model"] == "claude-opus-4-7"
+
+
 def test_returns_testable_when_no_uncovered_code(mock_claude):
     result = is_code_untestable(
         file_path="src/app.tsx",

services/get_fallback_models.py

@@ -10,6 +10,7 @@
 
 # Full fallback chains for resilience — includes non-user-selectable models
 CLAUDE_FALLBACK_MODELS: list[ModelId] = [
+    ClaudeModelId.OPUS_4_7,
     ClaudeModelId.OPUS_4_6,
     ClaudeModelId.OPUS_4_5,
     ClaudeModelId.SONNET_4_6,

services/supabase/credits/test_get_credit_price.py

@@ -16,9 +16,33 @@ def test_returns_max_cost_for_none():
     assert get_credit_price(None) == MAX_CREDIT_COST_USD
 
 
-def test_opus_costs_8():
+def test_opus_47_costs_8():
+    assert get_credit_price(ClaudeModelId.OPUS_4_7) == 8
+
+
+def test_opus_46_costs_8():
     assert get_credit_price(ClaudeModelId.OPUS_4_6) == 8
 
 
+def test_sonnet_46_costs_4():
+    assert get_credit_price(ClaudeModelId.SONNET_4_6) == 4
+
+
+def test_opus_45_costs_8():
+    assert get_credit_price(ClaudeModelId.OPUS_4_5) == 8
+
+
+def test_sonnet_45_costs_4():
+    assert get_credit_price(ClaudeModelId.SONNET_4_5) == 4
+
+
+def test_haiku_45_costs_2():
+    assert get_credit_price(ClaudeModelId.HAIKU_4_5) == 2
+
+
+def test_gemini_25_flash_costs_4():
+    assert get_credit_price(GoogleModelId.GEMINI_2_5_FLASH) == 4
+
+
 def test_gemma_costs_2():
     assert get_credit_price(GoogleModelId.GEMMA_4_31B) == 2

services/supabase/llm_requests/calculate_costs.py

@@ -11,6 +11,7 @@ def calculate_costs(
     # Pricing per 1M tokens (input/output)
     pricing = {
         "claude": {
+            "claude-opus-4-7": {"input": 5.00, "output": 25.00},
             "claude-opus-4-6": {"input": 5.00, "output": 25.00},
             "claude-opus-4-5": {"input": 5.00, "output": 25.00},
             "claude-sonnet-4-6": {"input": 3.00, "output": 15.00},