[pull] main from Alishahryar1:main

e2e/test_admin_vscode_chat.py

@@ -73,6 +73,11 @@ def retry(route):
         page.route("**/admin/api/integrations/vscode-chat", status)
         page.route("**/admin/api/integrations/vscode-chat/refresh", retry)
         page.goto(f"{admin_base_url}/admin/integrations")
+        # Startup reconciliation can temporarily disable Retry during the click.
+        # Exercise the action after initial polling has settled.
+        page.wait_for_function(
+            "state.startup && !state.startupRequest && state.startupTimer === null"
+        )
         main = page.locator("#openVSCodeChatIntegration")
         secondary = page.locator("#retryVSCodeChatIntegration")
         if status_read_failed:

e2e/test_code_sessions.py

@@ -2055,21 +2055,23 @@ def test_library_reconnect_removes_missed_deletion_and_searches_beyond_first_pag
 ):
     control_feed(page)
 
-    async def seed():
-        first = await code_control.service.create_session(
-            str(uuid.uuid4()), str(tmp_path)
-        )
-        await code_control.service.update_settings(
-            first.id, first.revision, {"title": "Needle project"}
-        )
-        for _ in range(26):
-            await code_control.service.create_session(str(uuid.uuid4()), str(tmp_path))
-        return first.id
-
     # Start the isolated service before seeding it.
     page.goto(f"{admin_base_url}/admin/code")
     expect(page.locator("#codeNew")).to_be_enabled()
-    session_id = code_control.run(seed())
+    # Bound each operation, not the cumulative cost of filling a whole page.
+    first = code_control.run(
+        code_control.service.create_session(str(uuid.uuid4()), str(tmp_path))
+    )
+    code_control.run(
+        code_control.service.update_settings(
+            first.id, first.revision, {"title": "Needle project"}
+        )
+    )
+    for _ in range(26):
+        code_control.run(
+            code_control.service.create_session(str(uuid.uuid4()), str(tmp_path))
+        )
+    session_id = first.id
     page.get_by_role("searchbox", name="Search titles and folders").fill("Needle")
     expect(page.locator(".session-card")).to_have_count(1)
     expect(page.locator(".session-card")).to_contain_text("Needle project")

smoke/capabilities.py

@@ -269,13 +269,16 @@ class CapabilityContract:
     ),
     CapabilityContract(
         "streaming_conversion",
-        "heuristic_tools",
-        "heuristic_tool_parser",
-        "free_claude_code.core.anthropic.tools",
-        "textual tool-call output",
+        "native_tools",
+        "native_tool_assembly",
+        "free_claude_code.providers.openai_chat.tool_calls",
+        "native structured tool-call output",
         "structured Anthropic tool_use blocks",
-        "text fallback when malformed",
-        ("tests/providers/test_parsers.py", "tests/contracts/test_stream_contracts.py"),
+        "raw arguments preserved for harness validation",
+        (
+            "tests/providers/test_native_tool_arguments.py",
+            "tests/contracts/test_stream_contracts.py",
+        ),
         (
             "test_live_tool_use_when_configured_model_supports_tools",
             "test_provider_reasoning_tool_continuation_e2e",
@@ -285,10 +288,10 @@ class CapabilityContract:
         "streaming_conversion",
         "subagent_task_control",
         "subagent_control",
-        "free_claude_code.core.anthropic.streaming.AnthropicStreamLedger",
+        "free_claude_code.providers.openai_chat.tool_calls",
         "Task tool call arguments",
-        "run_in_background=false",
-        "invalid JSON flushed as safe object",
+        "arguments preserved with normal alias restoration",
+        "ordinary tool validation and recovery",
         ("tests/providers/test_subagent_interception.py",),
     ),
     CapabilityContract(

smoke/features.py

@@ -325,9 +325,12 @@ def has_pytest_coverage(self) -> bool:
         "configured providers must not reject adaptive thinking payloads",
     ),
     FeatureCoverage(
-        "heuristic_tool_parser",
+        "native_tool_assembly",
         "Tool use and tool result continuation survive provider/client paths",
-        ("tests/providers/test_parsers.py", "tests/contracts/test_stream_contracts.py"),
+        (
+            "tests/providers/test_native_tool_arguments.py",
+            "tests/contracts/test_stream_contracts.py",
+        ),
         ("test_live_tool_use_when_configured_model_supports_tools",),
         (
             "test_provider_interleaved_thinking_tool_e2e",
@@ -405,7 +408,7 @@ def has_pytest_coverage(self) -> bool:
     ),
     FeatureCoverage(
         "subagent_control",
-        "Task-like tool output is rendered and controlled as foreground work",
+        "Task arguments are preserved and tool output is rendered in transcripts",
         ("tests/providers/test_subagent_interception.py",),
         (),
         ("test_messaging_subagent_control_e2e",),

src/free_claude_code/application/code_sessions/models.py

@@ -115,6 +115,27 @@ class CodeItemPage:
     next_before: tuple[int, int] | None = None
 
 
+@dataclass(frozen=True, slots=True)
+class CodeHistory:
+    session: CodeSession
+    run: CodeRun | None
+    items: tuple[CodeItem, ...]
+    prompts: tuple[CodePrompt, ...]
+    runs: tuple[CodeRun, ...]
+    active_prompt_ids: tuple[str, ...]
+    next_before: tuple[int, int] | None
+
+
+@dataclass(frozen=True, slots=True)
+class CodeExecutionSeed:
+    session: CodeSession
+    run: CodeRun | None
+    sequence: int
+    items: tuple[CodeItem, ...]
+    prompts: tuple[CodePrompt, ...]
+    runs: tuple[CodeRun, ...]
+
+
 class CodeModel(Record):
     id: str
     display_name: str

src/free_claude_code/application/code_sessions/ports.py

@@ -9,6 +9,8 @@
 from .models import (
     CodeCatalog,
     CodeDetail,
+    CodeExecutionSeed,
+    CodeHistory,
     CodeItem,
     CodeItemPage,
     CodeMode,
@@ -86,6 +88,37 @@ async def open_history(self, cwd: str, sink: EventSink) -> HarnessConnection: ..
 
 
 class CodeStore(Protocol):
+    async def execution_seed(self, session_id: str) -> CodeExecutionSeed: ...
+
+    async def read_history(
+        self,
+        session_id: str,
+        before: tuple[int, int] | None,
+        include_item_ids: Sequence[str],
+    ) -> CodeHistory: ...
+
+    async def get_native_item(
+        self,
+        session_id: str,
+        turn_id: str,
+        item_id: str,
+    ) -> CodeItem | None: ...
+
+    async def run_items(self, session_id: str, run_id: str) -> tuple[CodeItem, ...]: ...
+
+    async def get_prompt(
+        self, session_id: str, prompt_id: str
+    ) -> CodePrompt | None: ...
+
+    async def has_prompt_request(
+        self,
+        session_id: str,
+        generation: str,
+        request_id: str | int,
+    ) -> bool: ...
+
+    async def has_native_turn(self, session_id: str, turn_id: str) -> bool: ...
+
     async def start(self) -> None: ...
 
     async def close(self) -> None: ...