test: gate the complete plan-to-deployment workflow

Makefile

@@ -1,4 +1,4 @@
-.PHONY: install install-core run clean lint test test-core tui-test check docs-check shell-check benchmark wheel-smoke docker-smoke git-status git-diff git-log web
+.PHONY: install install-core run clean lint test test-core tui-test check docs-check shell-check benchmark agent-acceptance wheel-smoke docker-smoke git-status git-diff git-log web
 
 # Tests parse the benchmark target's stdout as JSON; do not inject GNU make's
 # recursive directory banners into that machine-readable output.
@@ -64,7 +64,7 @@ clean:
 
 # Syntax check
 lint:
-	$(VENV_PYTHON) -m compileall -q main.py tui_launcher.py tui_backend.py main_debug.py server.py tools.py memory.py event_store.py interaction.py task_engine.py learning_engine.py learning_benchmark.py learning_worker.py migration.py scheduler.py skill_registry.py browser_manager.py instruction_loader.py agent_config.py subagents.py capability_tokens.py tool_registry.py dynamic_tools.py plugin_runtime.py workspace_context.py project_graph.py github_cli.py scripts/generate_tool_inventory.py scripts/check_docs.py scripts/check_zsh_extras.py scripts/wheel_smoke.py
+	$(VENV_PYTHON) -m compileall -q main.py tui_launcher.py tui_backend.py main_debug.py server.py tools.py memory.py event_store.py interaction.py task_engine.py learning_engine.py learning_benchmark.py learning_worker.py migration.py scheduler.py skill_registry.py browser_manager.py instruction_loader.py agent_config.py subagents.py capability_tokens.py tool_registry.py dynamic_tools.py plugin_runtime.py workspace_context.py project_graph.py github_cli.py scripts/generate_tool_inventory.py scripts/check_docs.py scripts/check_zsh_extras.py scripts/agent_workflow_acceptance.py scripts/wheel_smoke.py
 	@echo "Python syntax OK."
 	@echo "All files pass syntax check."
 
@@ -91,6 +91,9 @@ wheel-smoke:
 	$(VENV_PYTHON) -m pip install build -q
 	$(VENV_PYTHON) scripts/wheel_smoke.py
 
+agent-acceptance:
+	$(VENV_PYTHON) scripts/agent_workflow_acceptance.py
+
 docs-check:
 	$(VENV_PYTHON) scripts/generate_tool_inventory.py --check
 	$(VENV_PYTHON) scripts/check_docs.py
@@ -107,7 +110,7 @@ docker-smoke:
 # Quick verification
 check:
 	@echo "Checking Python syntax..."
-	@$(VENV_PYTHON) -m py_compile main.py tui_launcher.py tui_backend.py main_debug.py server.py tools.py memory.py event_store.py interaction.py task_engine.py learning_engine.py learning_benchmark.py learning_worker.py migration.py scheduler.py skill_registry.py browser_manager.py instruction_loader.py agent_config.py subagents.py capability_tokens.py tool_registry.py dynamic_tools.py plugin_runtime.py workspace_context.py project_graph.py github_cli.py scripts/generate_tool_inventory.py scripts/check_docs.py scripts/check_zsh_extras.py scripts/wheel_smoke.py
+	@$(VENV_PYTHON) -m py_compile main.py tui_launcher.py tui_backend.py main_debug.py server.py tools.py memory.py event_store.py interaction.py task_engine.py learning_engine.py learning_benchmark.py learning_worker.py migration.py scheduler.py skill_registry.py browser_manager.py instruction_loader.py agent_config.py subagents.py capability_tokens.py tool_registry.py dynamic_tools.py plugin_runtime.py workspace_context.py project_graph.py github_cli.py scripts/generate_tool_inventory.py scripts/check_docs.py scripts/check_zsh_extras.py scripts/agent_workflow_acceptance.py scripts/wheel_smoke.py
 	@echo "  Python modules: OK"
 	@echo "Checking git tools..."
 	@$(VENV_PYTHON) -c "from tools import AVAILABLE_TOOLS; git = [k for k in AVAILABLE_TOOLS if k.startswith('git_')]; print(f'  {len(git)} git tools, {len(AVAILABLE_TOOLS)} total tools')"

docs/agentic-runtime-roadmap.md

@@ -21,3 +21,24 @@ is intentionally deferred rather than hidden behind speculative abstractions:
 
 MCP remains intentionally non-interactive. Tool approvals remain distinct from
 clarification questions on every future surface.
+
+## Release acceptance
+
+Run `make agent-acceptance` for the deterministic plan-to-deployment workflow.
+`make wheel-smoke` repeats the same check against a newly installed wheel from
+outside the checkout. It verifies plan approval, Agent-mode execution, durable
+receipts, generated files, and a real loopback HTTP deployment without using a
+provider key.
+
+Live-provider validation is optional and must be reported as unverified unless
+it is actually run. To exercise it manually, create a temporary project and
+state database, then run the installed UI with the configured provider:
+
+```bash
+work_dir="$(mktemp -d)"
+mkdir "$work_dir/project"
+KYROZEN_DB_PATH="$work_dir/state.sqlite3" kyrozen --project "$work_dir/project"
+```
+
+Select Plan mode, request a small locally served page, accept the proposal, and
+confirm the task receipts and deployment check before reporting a live pass.

docs/self-evolution.md

@@ -238,7 +238,7 @@ diagnostic only and is not treated as a release claim.
 
 ## Verification snapshot commands
 
-Current repository test count at this snapshot: **233 unittest cases**.
+Current repository test count at this snapshot: **239 unittest cases**.
 
 The post-#54 snapshot ran the repository's current checks and smoke coverage:
 
@@ -275,7 +275,7 @@ git diff --check
 ```
 
 The historical post-#54 verification passed the 130 discovered tests; the current
-repository contains 233 discovered tests. The historical run also covered the API
+repository contains 239 discovered tests. The historical run also covered the API
 health/scoping smoke, the CLI command-loop smoke, and the five-case
 clean/evolved benchmark described above. `make check` reports the live 39-tool
 runtime inventory, including 14 `git_` tools. A new artifact is not immediate:

interaction.py

@@ -182,6 +182,53 @@ def parse_control_block(text: str, name: str) -> Any | None:
         raise InteractionError(f"{name} contains invalid JSON: {exc.msg}") from exc
 
 
+def normalize_provider_control(text: str) -> tuple[str, dict[str, Any]] | None:
+    """Map a bounded provider JSON shape to one non-executable control."""
+    source = str(text or "").strip()
+    fenced = re.fullmatch(r"```(?:json)?\s*([\s\S]*?)\s*```", source, re.IGNORECASE)
+    if fenced:
+        source = fenced.group(1).strip()
+    try:
+        value = json.loads(source)
+    except json.JSONDecodeError:
+        return None
+    if not isinstance(value, dict):
+        return None
+
+    executable_keys = {"action", "args", "arguments", "command", "tool", "tool_calls"}
+    stack: list[Any] = [value]
+    while stack:
+        item = stack.pop()
+        if isinstance(item, dict):
+            if executable_keys & {str(key).lower() for key in item}:
+                return None
+            stack.extend(item.values())
+        elif isinstance(item, list):
+            stack.extend(item)
+
+    mode = str(value.get("mode") or "").strip().lower()
+    if mode == "plan" or value.get("do_not_execute_until_approved") is True:
+        steps = []
+        for index, step in enumerate(value.get("steps") or [], 1):
+            if not isinstance(step, dict):
+                return None
+            steps.append({
+                "id": str(step.get("id") or step.get("step_id") or f"step-{step.get('step', index)}"),
+                "title": step.get("title") or f"Step {index}",
+                "description": step.get("description") or step.get("details"),
+                "acceptance": step.get("acceptance") or step.get("acceptance_criteria"),
+            })
+        return "PlanProposal", {
+            "title": value.get("title") or value.get("plan_name"),
+            "summary": value.get("summary") or value.get("overview"),
+            "assumptions": value.get("assumptions") or [],
+            "steps": steps,
+        }
+    if mode == "ask" or value.get("ask_user") is True:
+        return "AskUser", {"questions": value.get("questions")}
+    return None
+
+
 def route_mode(preference: str, user_input: str, *, pending_plan: bool = False,
                executing_plan: bool = False) -> str:
     if executing_plan:

main.py

@@ -105,7 +105,7 @@ def _terminal_supports_unicode() -> bool:
 from event_store import stable_hash, utc_now
 from interaction import (
     InteractionController, InteractionError, is_plan_acceptance, mode_capabilities,
-    parse_control_block, render_plan, render_question, validate_plan_proposal,
+    normalize_provider_control, parse_control_block, render_plan, render_question, validate_plan_proposal,
     validate_question_request,
 )
 from learning_engine import LearningEngine
@@ -2748,16 +2748,25 @@ def interaction_envelope(user_input: str = "") -> dict[str, Any]:
     return _interaction_controller.envelope(user_input)
 
 
+def interaction_workspace_id(context: LaunchContext | None = None) -> str:
+    """Keep project interaction state separate without splitting global memory."""
+    active = context if context is not None else _launch_context
+    if isinstance(active, LaunchContext) and not active.is_global:
+        return active.source_scope_id
+    return memory_bank.workspace_id
+
+
 def bind_interaction_scope(session_id: str, *, user_id: str | None = None) -> None:
     """Bind the shared task and interaction facades to one local surface."""
     global tasks, _interaction_controller
     owner = user_id or memory_bank.user_id
+    workspace_id = interaction_workspace_id()
     tasks = TaskManager(
-        memory_bank.store, workspace_id=memory_bank.workspace_id,
+        memory_bank.store, workspace_id=workspace_id,
         session_id=session_id, user_id=owner,
     )
     _interaction_controller = InteractionController(
-        memory_bank.store, user_id=owner, workspace_id=memory_bank.workspace_id,
+        memory_bank.store, user_id=owner, workspace_id=workspace_id,
         session_id=session_id,
     )
     _restore_ponytail_level()
@@ -3879,6 +3888,17 @@ def _execute_turn_action(action: str, args: Any, *, operation_scope: str,
         )
     args = str(args)
     operation_id = _operation_id(operation_scope, canonical, args)
+    if (_is_state_changing_action(canonical, args)
+            and _interaction_controller.state().get("executing_plan")):
+        allowed, reason = tasks.mutation_matches_current_task(canonical, args)
+        if not allowed:
+            result = f"Error: action does not match the accepted plan; {reason}."
+            _notify_tool_execute(canonical, args, result)
+            return _make_execution_receipt(
+                action=canonical, args=args, authorized=False, started_at=started_at,
+                success=False, result=result, operation_scope=operation_scope,
+                failure="plan_action_mismatch",
+            )
     if _is_state_changing_action(canonical, args) and operation_id in successful_operations:
         result = "Error: duplicate successful state-changing action refused for this turn."
         _notify_tool_execute(canonical, args, result)
@@ -4337,14 +4357,17 @@ class DeepSeekDSMLFilter:
         r"(?i)(?<![\w])(?P<kind>Action|Thought|Plan|TaskList|TaskDone|DefineTool)\s*:"
     )
     _GENERIC_OPEN_RE = re.compile(
-        r"<\s*(?P<kind>invoke|parameter|calls|tool_calls|function_calls)\b[^>]*>",
+        r"<\s*(?P<kind>invoke|parameter|calls|tool_calls|function_calls|tool_use|notes|thought|reasoning)\b[^>]*>",
         re.IGNORECASE,
     )
     _GENERIC_CLOSE_RE = re.compile(
-        r"</\s*(?P<kind>invoke|parameter|calls|tool_calls|function_calls)\s*>",
+        r"</\s*(?P<kind>invoke|parameter|calls|tool_calls|function_calls|tool_use|notes|thought|reasoning)\s*>",
         re.IGNORECASE,
     )
-    _GENERIC_KINDS = ("invoke", "parameter", "calls", "tool_calls", "function_calls")
+    _GENERIC_KINDS = (
+        "invoke", "parameter", "calls", "tool_calls", "function_calls",
+        "tool_use", "notes", "thought", "reasoning",
+    )
     _ACTION_MARKER_RE = _ACTION_MARKER_RE
     _CONTROL_PREFIXES = tuple(
         item[:length].lower()
@@ -4474,7 +4497,10 @@ def _action_end(cls, value: str, start: int, *, final: bool) -> int | None:
     @classmethod
     def _generic_block_end(cls, value: str, match: re.Match[str]) -> int | None:
         kind = match.group("kind").lower()
-        close_kinds = "parameter" if kind == "parameter" else "invoke|calls|tool_calls|function_calls"
+        close_kinds = (
+            "parameter" if kind == "parameter" else
+            "invoke|calls|tool_calls|function_calls|tool_use|notes|thought|reasoning"
+        )
         close = re.compile(rf"</\s*(?:{close_kinds})\s*>", re.IGNORECASE).search(value, match.end())
         return close.end() if close else None
 
@@ -4740,6 +4766,12 @@ def _parse_model_response(text: str) -> dict[str, Any]:
     try:
         question_value = parse_control_block(raw, "AskUser")
         plan_value = parse_control_block(raw, "PlanProposal")
+        if question_value is None and plan_value is None:
+            normalized = normalize_provider_control(raw)
+            if normalized is not None:
+                name, value = normalized
+                question_value = value if name == "AskUser" else None
+                plan_value = value if name == "PlanProposal" else None
         if question_value is not None:
             question = validate_question_request(question_value)
         if plan_value is not None:
@@ -4856,6 +4888,8 @@ def _safe_fstring(s: str) -> str:
 
 def _tasks_from_plan(text: str) -> None:
     """Parse a Plan block and create pending tasks."""
+    if tasks.tasks:
+        return
     plan_match = re.search(
         r"Plan:\s*\n(.*?)(?=\n\s*(?:Action|TaskList|$))",
         text,
@@ -5745,6 +5779,10 @@ def _chat_turn(user_input: str, clear_tasks: bool = False, profile: str | None =
             "Execute the accepted plan below in Agent mode. Complete each durable task with evidence.\n\n"
             + json.dumps(accepted_plan, ensure_ascii=False)
         )
+        _emit_stream_event({
+            "event": "interaction",
+            "interaction": _interaction_controller.envelope(user_input),
+        })
         clear_tasks = False
     elif mode_override is None and pending_plan:
         user_input = (