context-compiler 0.4.2__tar.gz → 0.4.4__tar.gz

{context_compiler-0.4.2 → context_compiler-0.4.4}/AGENTS.md

@@ -61,6 +61,7 @@ Prefer modern typing syntax:
 - PR descriptions should include:
   - what changed
   - why the change was needed
+- A dedicated "Validation" section in PR text is optional and not required.
 - Keep PR scope aligned to the requested task; if scope grows, ask for guidance before expanding.
 
 ## CI

{context_compiler-0.4.2 → context_compiler-0.4.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: context-compiler
-Version: 0.4.2
+Version: 0.4.4
 Summary: Deterministic conversational state engine for LLM applications.
 Project-URL: Homepage, https://github.com/rlippmann/context-compiler
 Project-URL: Repository, https://github.com/rlippmann/context-compiler
@@ -73,19 +73,7 @@ User sets a constraint once:
 User: don't use peanuts
 ```
 
-State becomes:
-
-```json
-{
-  "facts": {
-    "focus.primary": null
-  },
-  "policies": {
-    "prohibit": ["peanuts"]
-  },
-  "version": 1
-}
-```
+Outcome: prohibited items now include `"peanuts"`.
 
 Later in the conversation:
 
@@ -157,35 +145,35 @@ else:
 
 | API | Description |
 |---|---|
-| `create_engine(...)` | Create a new compiler engine, optionally with replacement initial state. |
+| `create_engine(state=None)` | Create a new compiler engine; optional `state` provides initial authoritative state (validated/canonicalized). |
 | `step(user_input)` | Parse one user turn and return a deterministic `Decision`. |
 | `compile_transcript(messages)` | Replay a transcript from a fresh engine and return either final state or a confirmation prompt. |
 | `engine.apply_transcript(messages)` | Replay a transcript onto the current engine state and return either final state or a confirmation prompt. |
-| `engine.state` | Read or replace full in-memory authoritative state. |
+| `engine.state` | Read current authoritative in-memory state snapshot. |
+| `get_focus_value(state)` | Read the current focus value from a state snapshot. |
+| `get_prohibited_items(state)` | Read prohibited items from a state snapshot. |
 | `export_json()` | Export current state as JSON for persistence/transport. |
-| `import_json(payload)` | Load state from exported JSON payload. |
+| `import_json(payload)` | Load/restore state from exported JSON payload. |
 
 ---
 
 ## State Model
 
-The compiler maintains an authoritative state:
-
-```json
-{
-  "facts": {
-    "focus.primary": null
-  },
-  "policies": {
-    "prohibit": []
-  },
-  "version": 1
-}
-```
+The compiler maintains an authoritative state snapshot.
+Hosts should treat this state as structured application data and avoid coupling
+to internal field names or nested layout.
 
 ## State Access and Persistence
 
-Hosts may inspect or replace in-memory state (`engine.state`) or persist it using `export_json()` and `import_json()`. State changes occur only through directives processed by `step()`. Storage is managed by the host application.
+Hosts can provide initial state at engine creation (`create_engine(state=...)`),
+read current in-memory state via `engine.state`, and persist/restore via
+`export_json()` and `import_json()`. Semantic state mutations occur through
+directives processed by `step()`. Storage is managed by the host application.
+
+Use the returned state snapshot as structured host input for prompt
+construction, policy enforcement, or replay/storage workflows.
+For host code that needs typed reads without direct nested key lookups, use
+`get_focus_value(state)` and `get_prohibited_items(state)`.
 
 ### Transcript Replay
 
@@ -200,7 +188,7 @@ Transcript replay compiles conversational history by reusing the same determinis
 
 ### Fact Schema
 
-The current schema contains a single exclusive slot: `facts["focus.primary"]`.
+The current behavior includes one exclusive focus value.
 This demonstrates deterministic fact replacement and correction behavior.
 Richer schemas may be introduced in future releases.
 
@@ -221,9 +209,9 @@ User: use corn oil
 ```
 
 Result:
-facts.focus.primary = "corn oil"
+the current focus value becomes `"corn oil"`
 
-Because `focus.primary` is an exclusive slot, later `use ...` directives replace earlier values.
+Because the focus value is exclusive (last write wins), later `use ...` directives replace earlier values.
 
 This may differ from human expectations, where the intent may be interpreted as additive (e.g., ingredient + cooking medium). The current schema models a single focus value. See [issue #45](https://github.com/rlippmann/context-compiler/issues/45) for discussion.
 
@@ -238,14 +226,7 @@ User: don't use peanuts
 ```
 
 Result:
-
-```json
-{
-  "policies": {
-    "prohibit": ["peanuts"]
-  }
-}
-```
+prohibited items include `"peanuts"`.
 
 Fact configuration:
 
@@ -254,10 +235,7 @@ User: use vegetarian curry
 ```
 
 State update:
-
-```text
-facts.focus.primary = "vegetarian curry"
-```
+the current focus value becomes `"vegetarian curry"`
 
 Correction:
 
@@ -266,10 +244,7 @@ User: actually vegan curry
 ```
 
 Result:
-
-```text
-facts.focus.primary = "vegan curry"
-```
+the current focus value becomes `"vegan curry"`
 
 Ambiguous mutation:
 
@@ -289,40 +264,14 @@ No state mutation occurs until confirmation.
 
 Two explicit reset commands are supported:
 
-- `reset policies` clears `policies.prohibit` but preserves the current fact (`facts["focus.primary"]`)
+- `reset policies` clears prohibited items but preserves the current focus value
 - `clear state` resets the full state to initial values
 
 Example:
 
-Before:
-
-```json
-{
-  "facts": {"focus.primary": "vegetarian curry"},
-  "policies": {"prohibit": ["peanuts"]},
-  "version": 1
-}
-```
-
-After `reset policies`:
-
-```json
-{
-  "facts": {"focus.primary": "vegetarian curry"},
-  "policies": {"prohibit": []},
-  "version": 1
-}
-```
-
-After `clear state`:
-
-```json
-{
-  "facts": {"focus.primary": null},
-  "policies": {"prohibit": []},
-  "version": 1
-}
-```
+- If current focus is `"vegetarian curry"` and prohibited items include `"peanuts"`:
+- after `reset policies`, prohibited items are empty and focus remains `"vegetarian curry"`.
+- after `clear state`, both focus and prohibited items return to initial defaults.
 
 ---
 

{context_compiler-0.4.2 → context_compiler-0.4.4}/README.md

@@ -44,19 +44,7 @@ User sets a constraint once:
 User: don't use peanuts
 ```
 
-State becomes:
-
-```json
-{
-  "facts": {
-    "focus.primary": null
-  },
-  "policies": {
-    "prohibit": ["peanuts"]
-  },
-  "version": 1
-}
-```
+Outcome: prohibited items now include `"peanuts"`.
 
 Later in the conversation:
 
@@ -128,35 +116,35 @@ else:
 
 | API | Description |
 |---|---|
-| `create_engine(...)` | Create a new compiler engine, optionally with replacement initial state. |
+| `create_engine(state=None)` | Create a new compiler engine; optional `state` provides initial authoritative state (validated/canonicalized). |
 | `step(user_input)` | Parse one user turn and return a deterministic `Decision`. |
 | `compile_transcript(messages)` | Replay a transcript from a fresh engine and return either final state or a confirmation prompt. |
 | `engine.apply_transcript(messages)` | Replay a transcript onto the current engine state and return either final state or a confirmation prompt. |
-| `engine.state` | Read or replace full in-memory authoritative state. |
+| `engine.state` | Read current authoritative in-memory state snapshot. |
+| `get_focus_value(state)` | Read the current focus value from a state snapshot. |
+| `get_prohibited_items(state)` | Read prohibited items from a state snapshot. |
 | `export_json()` | Export current state as JSON for persistence/transport. |
-| `import_json(payload)` | Load state from exported JSON payload. |
+| `import_json(payload)` | Load/restore state from exported JSON payload. |
 
 ---
 
 ## State Model
 
-The compiler maintains an authoritative state:
-
-```json
-{
-  "facts": {
-    "focus.primary": null
-  },
-  "policies": {
-    "prohibit": []
-  },
-  "version": 1
-}
-```
+The compiler maintains an authoritative state snapshot.
+Hosts should treat this state as structured application data and avoid coupling
+to internal field names or nested layout.
 
 ## State Access and Persistence
 
-Hosts may inspect or replace in-memory state (`engine.state`) or persist it using `export_json()` and `import_json()`. State changes occur only through directives processed by `step()`. Storage is managed by the host application.
+Hosts can provide initial state at engine creation (`create_engine(state=...)`),
+read current in-memory state via `engine.state`, and persist/restore via
+`export_json()` and `import_json()`. Semantic state mutations occur through
+directives processed by `step()`. Storage is managed by the host application.
+
+Use the returned state snapshot as structured host input for prompt
+construction, policy enforcement, or replay/storage workflows.
+For host code that needs typed reads without direct nested key lookups, use
+`get_focus_value(state)` and `get_prohibited_items(state)`.
 
 ### Transcript Replay
 
@@ -171,7 +159,7 @@ Transcript replay compiles conversational history by reusing the same determinis
 
 ### Fact Schema
 
-The current schema contains a single exclusive slot: `facts["focus.primary"]`.
+The current behavior includes one exclusive focus value.
 This demonstrates deterministic fact replacement and correction behavior.
 Richer schemas may be introduced in future releases.
 
@@ -192,9 +180,9 @@ User: use corn oil
 ```
 
 Result:
-facts.focus.primary = "corn oil"
+the current focus value becomes `"corn oil"`
 
-Because `focus.primary` is an exclusive slot, later `use ...` directives replace earlier values.
+Because the focus value is exclusive (last write wins), later `use ...` directives replace earlier values.
 
 This may differ from human expectations, where the intent may be interpreted as additive (e.g., ingredient + cooking medium). The current schema models a single focus value. See [issue #45](https://github.com/rlippmann/context-compiler/issues/45) for discussion.
 
@@ -209,14 +197,7 @@ User: don't use peanuts
 ```
 
 Result:
-
-```json
-{
-  "policies": {
-    "prohibit": ["peanuts"]
-  }
-}
-```
+prohibited items include `"peanuts"`.
 
 Fact configuration:
 
@@ -225,10 +206,7 @@ User: use vegetarian curry
 ```
 
 State update:
-
-```text
-facts.focus.primary = "vegetarian curry"
-```
+the current focus value becomes `"vegetarian curry"`
 
 Correction:
 
@@ -237,10 +215,7 @@ User: actually vegan curry
 ```
 
 Result:
-
-```text
-facts.focus.primary = "vegan curry"
-```
+the current focus value becomes `"vegan curry"`
 
 Ambiguous mutation:
 
@@ -260,40 +235,14 @@ No state mutation occurs until confirmation.
 
 Two explicit reset commands are supported:
 
-- `reset policies` clears `policies.prohibit` but preserves the current fact (`facts["focus.primary"]`)
+- `reset policies` clears prohibited items but preserves the current focus value
 - `clear state` resets the full state to initial values
 
 Example:
 
-Before:
-
-```json
-{
-  "facts": {"focus.primary": "vegetarian curry"},
-  "policies": {"prohibit": ["peanuts"]},
-  "version": 1
-}
-```
-
-After `reset policies`:
-
-```json
-{
-  "facts": {"focus.primary": "vegetarian curry"},
-  "policies": {"prohibit": []},
-  "version": 1
-}
-```
-
-After `clear state`:
-
-```json
-{
-  "facts": {"focus.primary": null},
-  "policies": {"prohibit": []},
-  "version": 1
-}
-```
+- If current focus is `"vegetarian curry"` and prohibited items include `"peanuts"`:
+- after `reset policies`, prohibited items are empty and focus remains `"vegetarian curry"`.
+- after `clear state`, both focus and prohibited items return to initial defaults.
 
 ---
 

{context_compiler-0.4.2 → context_compiler-0.4.4}/demos/04_llm_tool_governance.py

@@ -2,8 +2,7 @@
 
 import re
 
-from context_compiler import create_engine
-from context_compiler.const import POLICY_PROHIBIT, STATE_POLICIES
+from context_compiler import create_engine, get_prohibited_items
 from demos.common import (
     build_baseline_messages,
     build_mediated_messages,
@@ -72,7 +71,7 @@ def main() -> None:
     baseline_output = complete_messages(baseline_messages)
     print_model_output("Baseline", baseline_output)
 
-    prohibited = engine.state[STATE_POLICIES][POLICY_PROHIBIT]
+    prohibited = get_prohibited_items(engine.state)
     candidate_tools = ["docker", "kubectl"]
     filtered_tools = [tool for tool in candidate_tools if tool not in prohibited]
     if is_verbose():

{context_compiler-0.4.2 → context_compiler-0.4.4}/demos/06_context_compaction.py

@@ -1,7 +1,6 @@
 """Demo 6: host-side prompt replacement from authoritative compiled state."""
 
-from context_compiler import compile_transcript
-from context_compiler.const import FOCUS_PRIMARY, STATE_FACTS
+from context_compiler import compile_transcript, get_focus_value
 from demos.common import is_verbose, print_info_report
 
 DEMO_NAME = "06_context_compaction — superseded directives eliminated"
@@ -44,7 +43,7 @@ def _compile_focus(turns: list[str]) -> str:
     messages: list[dict[str, object]] = [{"role": "user", "content": turn} for turn in turns]
     result = compile_transcript(messages)
     assert result["kind"] == "state"
-    compiled_focus = result["state"][STATE_FACTS][FOCUS_PRIMARY]
+    compiled_focus = get_focus_value(result["state"])
     assert compiled_focus is not None
     return compiled_focus
 

{context_compiler-0.4.2 → context_compiler-0.4.4}/demos/common.py

@@ -5,8 +5,7 @@ import os
 import re
 from typing import Any, TypedDict
 
-from context_compiler import Decision, State
-from context_compiler.const import FOCUS_PRIMARY, POLICY_PROHIBIT, STATE_FACTS, STATE_POLICIES
+from context_compiler import Decision, State, get_focus_value, get_prohibited_items
 from demos.llm_client import Message
 
 VERBOSE_ENV_VAR = "CONTEXT_COMPILER_DEMO_VERBOSE"
@@ -179,8 +178,8 @@ def consume_last_info_report() -> InfoReport | None:
 
 
 def build_compiled_system_prompt(state: State) -> str:
-    focus_value = state[STATE_FACTS][FOCUS_PRIMARY]
-    prohibit = state[STATE_POLICIES][POLICY_PROHIBIT]
+    focus_value = get_focus_value(state)
+    prohibit = get_prohibited_items(state)
     prohibit_text = ", ".join(prohibit) if prohibit else "(none)"
     focus_text = focus_value if focus_value is not None else "(unset)"
     return (

{context_compiler-0.4.2 → context_compiler-0.4.4}/docs/DirectiveGrammarSpec.md

@@ -43,6 +43,15 @@ The host:
 - Displays clarification prompts
 - Calls the LLM when allowed
 - Formats prompts using provided state
+- May read state snapshots directly, but should prefer public helper accessors where available.
+
+Current helpers:
+- `get_focus_value(state)`
+- `get_prohibited_items(state)`
+
+These helpers are read-only conveniences for state snapshots to reduce direct
+coupling to nested layout. They do not modify compiler state and are not
+semantic/compiler primitives.
 
 ### 4. Decision API Contract
 
@@ -291,11 +300,12 @@ directive parsing, and does not mutate state.
 Adding duplicate policy is a no-op.
 Policies stored in sorted lexical order.
 
-Administrative state replacement is also supported through public host APIs:
-- `engine.state = ...` (object replacement)
+Administrative state initialization/replacement is supported through:
+- constructor input (`create_engine(state=...)` / `Engine(state=...)`) for initial load
 - `engine.import_json(payload)` (JSON replacement)
 
-Both replacement paths clear pending clarification state and must behave like live state for subsequent `step()` calls.
+Import-based replacement clears pending clarification state and must behave like
+live state for subsequent `step()` calls.
 
 ### 10. Context Serialization Contract
 

{context_compiler-0.4.2 → context_compiler-0.4.4}/examples/01_persistent_guardrails.py

@@ -2,12 +2,11 @@
 
 from _util import print_json
 
-from context_compiler import State, create_engine
-from context_compiler.const import POLICY_PROHIBIT, STATE_POLICIES
+from context_compiler import State, create_engine, get_prohibited_items
 
 
 def build_prompt(state: State, user_input: str) -> str:
-    prohibit = state[STATE_POLICIES][POLICY_PROHIBIT]
+    prohibit = get_prohibited_items(state)
     prohibit_text = ", ".join(prohibit) if prohibit else "(none)"
     return (
         "System: Follow authoritative conversation state.\n"

{context_compiler-0.4.2 → context_compiler-0.4.4}/examples/04_tool_governance_denylist.py

@@ -4,8 +4,7 @@ from dataclasses import dataclass
 
 from _util import print_json
 
-from context_compiler import create_engine
-from context_compiler.const import POLICY_PROHIBIT, STATE_POLICIES
+from context_compiler import create_engine, get_prohibited_items
 
 
 @dataclass
@@ -35,7 +34,7 @@ def main() -> None:
     print()
 
     print("Host-side tool denylist behavior:")
-    prohibit = state[STATE_POLICIES][POLICY_PROHIBIT]
+    prohibit = get_prohibited_items(state)
     tools = [Tool("docker"), Tool("kubectl")]
     for tool in tools:
         if tool.name in prohibit:

{context_compiler-0.4.2 → context_compiler-0.4.4}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "context-compiler"
-version = "0.4.2"
+version = "0.4.4"
 description = "Deterministic conversational state engine for LLM applications."
 readme = "README.md"
 requires-python = ">=3.11"

context_compiler-0.4.4/src/context_compiler/__init__.py

@@ -0,0 +1,25 @@
+from importlib.metadata import version
+
+from .engine import (
+    ApplyResult,
+    Decision,
+    Engine,
+    State,
+    compile_transcript,
+    create_engine,
+    get_focus_value,
+    get_prohibited_items,
+)
+
+__version__ = version("context-compiler")
+
+__all__ = [
+    "ApplyResult",
+    "Decision",
+    "Engine",
+    "State",
+    "compile_transcript",
+    "create_engine",
+    "get_focus_value",
+    "get_prohibited_items",
+]

{context_compiler-0.4.2 → context_compiler-0.4.4}/src/context_compiler/engine.py

@@ -123,6 +123,16 @@ def compile_transcript(messages: list[dict[str, object]]) -> ApplyResult:
     return engine.apply_transcript(messages)
 
 
+def get_focus_value(state: State) -> str | None:
+    """Return the current exclusive focus value from a state snapshot."""
+    return state[STATE_FACTS][FOCUS_PRIMARY]
+
+
+def get_prohibited_items(state: State) -> list[str]:
+    """Return prohibited items from a state snapshot as a defensive list copy."""
+    return list(state[STATE_POLICIES][POLICY_PROHIBIT])
+
+
 class Engine:
     """Deterministic state engine implementing directive semantics.
 
@@ -132,7 +142,7 @@ class Engine:
       directive input to ``step()``.
     - Host code should not rely on imperative helpers such as
       ``reset_policies()`` or ``clear_state()``.
-    - State may be administratively replaced via ``engine.state = ...`` and
+    - State may be administratively replaced via constructor input and
       ``engine.import_json(...)``.
     """
 
@@ -153,15 +163,6 @@ class Engine:
         """Return a defensive copy of the current authoritative in-memory state."""
         return deepcopy(self._state)
 
-    @state.setter
-    def state(self, value: State) -> None:
-        """Replace authoritative in-memory state from a supplied object.
-
-        The supplied value is validated and canonicalized. Replacement is full,
-        and pending clarification state is cleared.
-        """
-        self._replace_state(_load_state_obj(value))
-
     def export_json(self) -> str:
         """Serialize authoritative state for persistence or transport."""
         return json.dumps(self._state, sort_keys=True, separators=(",", ":"))

{context_compiler-0.4.2 → context_compiler-0.4.4}/tests/test_engine.py

@@ -2,7 +2,7 @@ import json
 
 import pytest
 
-from context_compiler import create_engine
+from context_compiler import create_engine, get_focus_value, get_prohibited_items
 from context_compiler.engine import DecisionKind, Engine
 
 
@@ -20,6 +20,23 @@ def test_state_getter_returns_defensive_copy() -> None:
     assert engine.state["facts"]["focus.primary"] is None
 
 
+def test_get_focus_value_reads_current_focus_from_state_snapshot() -> None:
+    engine = create_engine()
+    engine.step("use Nord Stage 4")
+
+    assert get_focus_value(engine.state) == "Nord Stage 4"
+
+
+def test_get_prohibited_items_returns_defensive_list_copy() -> None:
+    engine = create_engine()
+    engine.step("don't use docker")
+
+    prohibited = get_prohibited_items(engine.state)
+    prohibited.append("kubernetes")
+
+    assert get_prohibited_items(engine.state) == ["docker"]
+
+
 def test_export_json_returns_complete_representation_of_state() -> None:
     engine = create_engine()
     engine.step("use Nord Stage 4")
@@ -195,7 +212,7 @@ def test_create_engine_with_state_initializes_from_normalized_state() -> None:
             },
         ),
         (
-            "setter",
+            "import_json",
             {
                 "facts": {"focus.primary": None},
                 "policies": {"prohibit": "docker"},
@@ -212,22 +229,36 @@ def test_object_state_replacement_paths_reject_invalid_state(path: str, bad_stat
 
     engine = create_engine()
     with pytest.raises(ValueError):
-        engine.state = bad_state
+        engine.import_json(json.dumps(bad_state))
 
 
-def test_state_setter_replaces_state_and_clears_pending_clarification() -> None:
+def test_state_property_is_read_only() -> None:
+    engine = create_engine()
+    with pytest.raises(AttributeError):
+        engine.state = {
+            "facts": {"focus.primary": None},
+            "policies": {"prohibit": []},
+            "version": 1,
+        }
+
+
+def test_import_json_replaces_state_and_clears_pending_clarification() -> None:
     engine = create_engine()
     decision = engine.step("no use docker")
     assert decision["kind"] == "clarify"
 
-    engine.state = {
-        "facts": {"focus.primary": "Nord Stage 4"},
-        "policies": {"prohibit": ["kubernetes", "docker", "docker"]},
-        "version": 1,
-    }
+    engine.import_json(
+        json.dumps(
+            {
+                "facts": {"focus.primary": "Nord Stage 4"},
+                "policies": {"prohibit": ["kubernetes", "docker", "docker"]},
+                "version": 1,
+            }
+        )
+    )
 
-    decision_after_setter = engine.step("yes")
-    assert decision_after_setter["kind"] == "passthrough"
+    decision_after_import = engine.step("yes")
+    assert decision_after_import["kind"] == "passthrough"
     assert engine.state == {
         "facts": {"focus.primary": "Nord Stage 4"},
         "policies": {"prohibit": ["docker", "kubernetes"]},
@@ -235,7 +266,7 @@ def test_state_setter_replaces_state_and_clears_pending_clarification() -> None:
     }
 
 
-def test_constructor_setter_and_import_json_share_normalization_behavior() -> None:
+def test_constructor_and_import_json_share_normalization_behavior() -> None:
     raw_state = {
         "facts": {"focus.primary": "  MacBook   M3`  "},
         "policies": {"prohibit": ["kubernetes", "docker", "docker"]},
@@ -244,13 +275,10 @@ def test_constructor_setter_and_import_json_share_normalization_behavior() -> No
 
     from_ctor = Engine(state=json.loads(json.dumps(raw_state)))
 
-    from_setter = create_engine()
-    from_setter.state = json.loads(json.dumps(raw_state))
-
     from_import = create_engine()
     from_import.import_json(json.dumps(raw_state))
 
-    assert from_ctor.state == from_setter.state == from_import.state
+    assert from_ctor.state == from_import.state
 
 
 def test_import_json_clears_pending_clarification_state() -> None:
@@ -881,9 +909,9 @@ def test_pending_clarification_rejected_by_explicit_no_is_passthrough() -> None:
     ("path", "bad_state"),
     [
         ("constructor", []),
-        ("setter", "not-a-dict"),
+        ("import_json", "not-a-dict"),
         ("constructor", {"facts": [], "policies": {"prohibit": []}, "version": 1}),
-        ("setter", {"facts": {"focus.primary": None}, "policies": [], "version": 1}),
+        ("import_json", {"facts": {"focus.primary": None}, "policies": [], "version": 1}),
         (
             "constructor",
             {"facts": {"focus.primary": 123}, "policies": {"prohibit": []}, "version": 1},
@@ -898,7 +926,7 @@ def test_object_state_paths_reject_malformed_state_inputs(path: str, bad_state:
 
     engine = create_engine()
     with pytest.raises(ValueError):
-        engine.state = bad_state
+        engine.import_json(json.dumps(bad_state))
 
 
 def test_allow_suffix_removes_existing_prohibition() -> None:
@@ -909,3 +937,20 @@ def test_allow_suffix_removes_existing_prohibition() -> None:
 
     assert decision["kind"] == "update"
     assert engine.state["policies"]["prohibit"] == []
+
+
+def test_correction_payload_that_invokes_other_directive_family_is_clarified_without_mutation() -> (
+    None
+):
+    engine = create_engine()
+    engine.step("use Nord Stage 4")
+    before = engine.state
+
+    decision = engine.step("actually don't use docker")
+
+    assert decision["kind"] == "clarify"
+    assert (
+        decision["prompt_to_user"]
+        == "Your directive mixes multiple directive types. Please provide one."
+    )
+    assert engine.state == before

{context_compiler-0.4.2 → context_compiler-0.4.4}/uv.lock

@@ -53,7 +53,7 @@ wheels = [
 
 [[package]]
 name = "context-compiler"
-version = "0.4.2"
+version = "0.4.4"
 source = { editable = "." }
 
 [package.optional-dependencies]

context_compiler-0.4.2/src/context_compiler/__init__.py

@@ -1,7 +0,0 @@
-from importlib.metadata import version
-
-from .engine import ApplyResult, Decision, Engine, State, compile_transcript, create_engine
-
-__version__ = version("context-compiler")
-
-__all__ = ["ApplyResult", "Decision", "Engine", "State", "compile_transcript", "create_engine"]