refactor: adopt directive drafter dev5 integrations

PROVIDER_CONTRACT.md

@@ -16,7 +16,6 @@ export PROVIDER=openai
 export MODEL=openai/gpt-4o-mini
 export PREPROCESSOR_MODEL=openai/gpt-4o-mini
 export OPENAI_BASE_URL=...
-export PREPROCESSOR_PROMPT_PROFILE=default
 ```
 
 Provider mode contract (`PROVIDER`) is strict:
@@ -52,5 +51,5 @@ Startup emits one concise config line showing resolved `mode`, `base_url`,
 `MODEL` and `PREPROCESSOR_MODEL` use LiteLLM format: `<provider>/<model>`.
 `PREPROCESSOR_MODEL` is optional and defaults to `MODEL`.
 
-For heuristic-first usage, keep `PREPROCESSOR_PROMPT_PROFILE=default`.
-Use `llama` only for LLM-only preprocessing with Llama-family models.
+The directive-drafter integration always uses heuristic-first processing with
+the configured fallback model when needed.

pyproject.toml

@@ -47,12 +47,12 @@ Issues = "https://github.com/rlippmann/context-compiler-example-integrations/iss
 [project.optional-dependencies]
 all = [
   "chromadb",
-  "context-compiler-directive-drafter==0.2.0dev4",
+  "context-compiler-directive-drafter==0.2.0dev5",
   "fastapi",
   "litellm",
 ]
 drafter = [
-  "context-compiler-directive-drafter==0.2.0dev4",
+  "context-compiler-directive-drafter==0.2.0dev5",
 ]
 fastapi = [
   "fastapi",
@@ -67,7 +67,7 @@ retrieval = [
 [dependency-groups]
 dev = [
   "chromadb",
-  "context-compiler-directive-drafter==0.2.0dev4",
+  "context-compiler-directive-drafter==0.2.0dev5",
   "fastapi",
   "httpx2>=2.5.0",
   "httpx>=0.28.1",

python/examples/prompt_construction/litellm/README.md

@@ -68,7 +68,7 @@ For `with_directive_drafter.py`:
 pip install "context-compiler-example-integrations[all]"
 ```
 
-That variant requires `context-compiler-directive-drafter>=0.2.0dev4`.
+That variant requires `context-compiler-directive-drafter>=0.2.0dev5`.
 
 ## Quickstart (copy/paste)
 

python/examples/prompt_construction/litellm/with_directive_drafter.py

@@ -18,6 +18,7 @@
 import logging
 import os
 from collections.abc import Callable, Mapping, Sequence
+from functools import lru_cache
 from importlib import import_module
 from typing import TypedDict, cast
 
@@ -37,7 +38,9 @@
     RejectedDirective,
     UnknownDirective,
 )
-from context_compiler_directive_drafter.fallbacks import get_converter_prompt
+from context_compiler_directive_drafter.fallbacks.litellm import (
+    create_litellm_fallback,
+)
 
 from context_compiler_example_integrations.examples._shared.provider_mode import (
     print_startup_config,
@@ -82,12 +85,6 @@ def _extract_response_content(response: object) -> str | None:
     return None
 
 
-_DIRECTIVE_DRAFTER = DirectiveDrafter(
-    fallback=lambda message: _llm_fallback_candidate(message),
-    fallback_source="litellm_fallback",
-)
-
-
 def _render_state_lines(
     premise: str | None, policies: Mapping[str, PolicyValue]
 ) -> list[str]:
@@ -211,44 +208,31 @@ def _call_litellm(messages: list[dict[str, str]]) -> str:
     return content
 
 
-def _llm_fallback_candidate(message: str) -> str | None:
-    try:
-        completion = _get_litellm_completion()
-    except ModuleNotFoundError:
-        return None
+@lru_cache(maxsize=8)
+def _create_directive_drafter(
+    model: str, api_key: str | None, api_base: str
+) -> DirectiveDrafter:
+    return DirectiveDrafter(
+        fallback=create_litellm_fallback(
+            model=model,
+            api_key=api_key,
+            api_base=api_base,
+        ),
+        fallback_source="litellm_fallback",
+    )
 
-    try:
-        config = resolve_provider_config(default_model="openai/gpt-4o-mini")
-    except RuntimeError:
-        return None
-    if config.mode == "openai" and not config.api_key:
-        return None
-    preprocessor_model = os.getenv("PREPROCESSOR_MODEL", "").strip()
-    if not preprocessor_model:
-        preprocessor_model = os.getenv("MODEL", "openai/gpt-4o-mini")
 
-    kwargs: _LiteLLMCallKwargs = {
-        "model": preprocessor_model,
-        "messages": [
-            {"role": "system", "content": get_converter_prompt()},
-            {"role": "user", "content": message},
-        ],
-        "temperature": 0,
-        "api_base": config.base_url,
-    }
-    if config.api_key:
-        kwargs["api_key"] = config.api_key
-
-    try:
-        response = completion(**kwargs)
-        return _extract_response_content(response)
-    except Exception:
-        return None
+def _get_directive_drafter() -> DirectiveDrafter:
+    config = resolve_provider_config(default_model="openai/gpt-4o-mini")
+    preprocessor_model = os.getenv("PREPROCESSOR_MODEL", "").strip() or config.model
+    return _create_directive_drafter(
+        preprocessor_model, config.api_key, config.base_url
+    )
 
 
 def _preprocess_user_input(message: str) -> str | None:
     try:
-        drafted_result = _DIRECTIVE_DRAFTER.draft_directive(message)
+        drafted_result = _get_directive_drafter().draft_directive(message)
         logger.debug("preprocessor: drafted_result=%r", drafted_result)
         return _extract_drafted_text(drafted_result)
     except Exception:

python/reference_integrations/litellm_proxy/README.md

@@ -108,7 +108,7 @@ For `context_compiler_precall_hook_with_directive_drafter.py`:
 pip install "context-compiler-example-integrations[all]"
 ```
 
-That variant requires `context-compiler-directive-drafter>=0.2.0dev4`.
+That variant requires `context-compiler-directive-drafter>=0.2.0dev5`.
 
 For the opt-in runtime smoke test, install the proxy runtime extras:
 
@@ -202,13 +202,12 @@ Optional env vars for directive-drafter fallback:
 
 ```shell
 export PREPROCESSOR_MODEL=openai/gpt-4o-mini
-export PREPROCESSOR_PROMPT_PROFILE=default
 ```
 
 `PREPROCESSOR_MODEL` is optional and defaults to `MODEL`.
 
-For heuristic-first usage, keep `PREPROCESSOR_PROMPT_PROFILE=default`.
-Use `llama` only for LLM-only fallback drafting with Llama-family models.
+The directive-drafter integration always uses heuristic-first processing with
+the configured fallback model when needed.
 
 ## Notes
 

python/reference_integrations/litellm_proxy/context_compiler_precall_hook_with_directive_drafter.py

@@ -13,9 +13,8 @@
 
 import logging
 import os
-from collections.abc import Callable, Mapping, Sequence
-from importlib import import_module
-from typing import Any, cast
+from functools import lru_cache
+from typing import Any
 
 try:
     from litellm.integrations.custom_logger import CustomLogger
@@ -33,7 +32,9 @@ class CustomLogger:  # type: ignore[no-redef]
     DirectiveDrafter,
     DraftResult,
 )
-from context_compiler_directive_drafter.fallbacks import get_converter_prompt
+from context_compiler_directive_drafter.fallbacks.litellm import (
+    create_litellm_fallback,
+)
 from context_compiler_example_integrations.reference_integrations.litellm_proxy._checkpoint_support import (
     MODE_PERSISTENT,
     CheckpointStore,
@@ -60,75 +61,41 @@ class CustomLogger:  # type: ignore[no-redef]
 CHECKPOINT_STORE: CheckpointStore = InMemoryCheckpointStore()
 
 
-def _extract_response_content(response: object) -> str | None:
-    if isinstance(response, Mapping):
-        choices = response.get("choices")
-        if isinstance(choices, Sequence) and choices:
-            first = choices[0]
-            if isinstance(first, Mapping):
-                message = first.get("message")
-                if isinstance(message, Mapping):
-                    content = message.get("content")
-                    if isinstance(content, str):
-                        return content
-
-    choices_attr = getattr(response, "choices", None)
-    if isinstance(choices_attr, Sequence) and choices_attr:
-        first = choices_attr[0]
-        message_attr = getattr(first, "message", None)
-        content_attr = getattr(message_attr, "content", None)
-        if isinstance(content_attr, str):
-            return content_attr
-
-    return None
-
-
-def _get_litellm_completion() -> Callable[..., object]:
-    litellm_module = import_module("litellm")
-    return cast(Callable[..., object], litellm_module.completion)
+@lru_cache(maxsize=8)
+def _create_directive_drafter(
+    model: str, api_key: str, api_base: str | None
+) -> DirectiveDrafter:
+    return DirectiveDrafter(
+        fallback=create_litellm_fallback(
+            model=model,
+            api_key=api_key,
+            api_base=api_base,
+        ),
+        fallback_source="litellm_fallback",
+    )
 
 
-def _llm_fallback_candidate(message: str) -> str | None:
+def _get_directive_drafter() -> DirectiveDrafter:
     preprocessor_model = os.getenv("PREPROCESSOR_MODEL", "").strip()
     if not preprocessor_model:
         preprocessor_model = os.getenv("MODEL", "").strip()
     if not preprocessor_model:
-        return None
+        return DirectiveDrafter()
 
     api_key = os.getenv("OPENAI_API_KEY")
     if not api_key:
-        return None
+        return DirectiveDrafter()
+    return _create_directive_drafter(
+        preprocessor_model, api_key, os.getenv("OPENAI_BASE_URL") or None
+    )
 
-    try:
-        completion = _get_litellm_completion()
-    except ModuleNotFoundError:
-        return None
-
-    kwargs: dict[str, object] = {
-        "model": preprocessor_model,
-        "messages": [
-            {"role": "system", "content": get_converter_prompt()},
-            {"role": "user", "content": message},
-        ],
-        "api_key": api_key,
-        "temperature": 0,
-    }
-    api_base = os.getenv("OPENAI_BASE_URL")
-    if api_base:
-        kwargs["api_base"] = api_base
 
+def _draft_last_user_message(message: str) -> DraftResult:
     try:
-        response = completion(**kwargs)
-        return _extract_response_content(response)
+        return _get_directive_drafter().draft_directive(message)
     except Exception:
-        return None
-
-
-def _draft_last_user_message(message: str) -> DraftResult:
-    drafter = DirectiveDrafter(
-        fallback=_llm_fallback_candidate, fallback_source="litellm_fallback"
-    )
-    return drafter.draft_directive(message)
+        logger.debug("litellm_proxy: drafter_exception", exc_info=True)
+        return DirectiveDrafter().draft_directive(message)
 
 
 class ContextCompilerPreCallHookWithPreprocessor(CustomLogger):

python/reference_integrations/openwebui_pipe/README.md

@@ -68,11 +68,9 @@ These examples require `context-compiler>=0.9.0dev13`.
 If using `open_webui_pipe_with_directive_drafter.py`:
 
 - Install directive-drafter support if needed:
-  `pip install "context-compiler>=0.9.0dev13" "context-compiler-directive-drafter>=0.2.0dev4"`
-- Set `PREPROCESSOR_PROMPT_PROFILE=default` for heuristic-first behavior
+  `pip install "context-compiler>=0.9.0dev13" "context-compiler-directive-drafter>=0.2.0dev5"`
 - Optionally set `PREPROCESSOR_MODEL_ID` to use a separate fallback drafting model
 - If `PREPROCESSOR_MODEL_ID` is unset, fallback uses `BASE_MODEL_ID`
-- Use `llama` only for LLM-only fallback drafting with Llama-family models
 
 Model fallback output is structurally validated before handoff. This does not prove that the model interpreted the user correctly. The automated fallback path is experimental pending a separate source-aware acceptance policy and reviewed drafting workflow.
 
@@ -86,7 +84,7 @@ If frontmatter dependency installs are disabled, offline, or unavailable:
 1. Install the package manually:
 
 - Minimal pipe: `pip install "context-compiler>=0.9.0dev13"`
-- Directive Drafter pipe: `pip install "context-compiler>=0.9.0dev13" "context-compiler-directive-drafter>=0.2.0dev4"`
+- Directive Drafter pipe: `pip install "context-compiler>=0.9.0dev13" "context-compiler-directive-drafter>=0.2.0dev5"`
 
 1. Import and enable the function in Open WebUI, then configure valves.
 
@@ -246,7 +244,7 @@ rejection flows.
 - `PREPROCESSOR_MODEL_ID must not match the selected pipe model id`: choose a real backend model id, not the pipe model id itself.
 - `PREPROCESSOR_MODEL_ID is invalid or not configured in Open WebUI`: the fallback route hit a missing model; fix the configured fallback model or unset it to reuse `BASE_MODEL_ID`.
 - `ALLOW_MISSING_BASE_MODEL_FOR_DEBUG=true`: directive-only updates still run locally, but passthrough returns a deterministic debug message instead of calling a downstream model.
-- imports fail after function upload: install `context-compiler>=0.9.0dev13` in the Open WebUI runtime, and add `context-compiler-directive-drafter>=0.2.0dev4` only for the Directive Drafter pipe, because the copied function runs from a temp/cached location.
+- imports fail after function upload: install `context-compiler>=0.9.0dev13` in the Open WebUI runtime, and add `context-compiler-directive-drafter>=0.2.0dev5` only for the Directive Drafter pipe, because the copied function runs from a temp/cached location.
 
 ## Fallback notes
 

python/reference_integrations/openwebui_pipe/open_webui_pipe_with_directive_drafter.py

@@ -3,7 +3,7 @@
 author: rlippmann
 author_url: https://github.com/rlippmann/context-compiler-example-integrations
 version: 0.10.1
-requirements: context-compiler>=0.9.0dev13, context-compiler-directive-drafter>=0.2.0dev4
+requirements: context-compiler>=0.9.0dev13, context-compiler-directive-drafter>=0.2.0dev5
 
 Open WebUI integration with Context Compiler directive drafter.
 
@@ -24,7 +24,7 @@
 import logging
 import re
 from collections.abc import AsyncIterator
-from typing import Any, Literal, TypedDict, cast
+from typing import Any, TypedDict, cast
 
 from fastapi import Request  # type: ignore[import-not-found]
 from open_webui.models.users import Users  # type: ignore[import-not-found]
@@ -62,10 +62,12 @@ def Field(*, default: Any, description: str = "") -> Any:  # type: ignore[no-red
     RejectedDirective,
     UnknownDirective,
 )
-from context_compiler_directive_drafter.fallbacks import get_converter_prompt
+from context_compiler_directive_drafter.fallbacks import get_fallback_profile
 
 logger = logging.getLogger(__name__)
 
+_FALLBACK_PROFILE = get_fallback_profile()
+
 _CC_MARKER = "[[cc_state]]"
 _ENGINES_BY_CHAT_KEY: dict[str, Engine] = {}
 
@@ -381,10 +383,6 @@ class Valves(BaseModel):
                 "Optional model id for fallback drafting (defaults to BASE_MODEL_ID)."
             ),
         )
-        PREPROCESSOR_PROMPT_PROFILE: Literal["default", "llama"] = Field(
-            default="default",
-            description="Prompt profile for LLM fallback drafting.",
-        )
         ALLOW_MISSING_BASE_MODEL_FOR_DEBUG: bool = Field(
             default=False,
             description="Allow missing BASE_MODEL_ID for debug/testing only.",
@@ -634,7 +632,7 @@ async def _llm_fallback_candidate(
             "model": model_id,
             "stream": False,
             "messages": [
-                {"role": "system", "content": get_converter_prompt()},
+                {"role": "system", "content": _FALLBACK_PROFILE.system_prompt},
                 {"role": "user", "content": message},
             ],
         }
@@ -695,10 +693,8 @@ async def _preprocess_user_input(
         *,
         request: Request,
         user_payload: dict[str, Any],
-        prompt_profile: str,
         model_id: str | None,
     ) -> tuple[DraftResult, str | None]:
-        del prompt_profile
         self._last_preprocessor_error = None
         drafted_result = await self._draft_user_input(
             message,
@@ -892,7 +888,6 @@ async def pipe(
             latest_user_text,
             request=__request__,
             user_payload=__user__,
-            prompt_profile=self.valves.PREPROCESSOR_PROMPT_PROFILE,
             model_id=effective_preprocessor_model,
         )
         if preprocess_error is not None: