Tiny fixes: escape sequence, dead code removal, docs

AGENTS.md

@@ -89,6 +89,21 @@ Always check the package-level `AGENTS.md` for file-specific rules
 - Formatting/linting: no repo-wide formatter is configured; follow existing
   file style and do not add new lint tools unless requested.
 
+## LLM structured-output schemas (Literal vs Enum)
+- In Pydantic models used for LLM structured output, use `Literal["a", "b"]`
+  instead of an `Enum` type for field annotations.
+- Reason: `Enum` fields cause `model_json_schema()` to emit `$defs`/`$ref`
+  indirection. Some structured-output backends (e.g. MLX Outlines for local
+  inference) cannot resolve `$defs`/`$ref` and fail silently or error out.
+  `Literal` produces a flat schema with an inline `enum` array, which is
+  universally supported.
+- Keep the canonical `str(Enum)` class in the same module as the single source
+  of truth for valid values; the `Literal` duplicates those values in the
+  schema annotation only.
+- A CI parity test (`test_enum_literal_parity.py`) asserts that every `Literal`
+  field stays in sync with its corresponding `Enum`. If you add or rename an
+  Enum member, update the matching `Literal` and vice versa.
+
 ## Python version
 - Canonical version is defined in each package `pyproject.toml`.
 

presets/README.md

@@ -0,0 +1,18 @@
+# Presets
+
+LM Studio preset files that configure local models for use with PlanExe.
+
+## What are LM Studio presets?
+
+LM Studio presets are JSON configs that tune model parameters (temperature, context length, max tokens, system prompt, thinking mode, etc.) for a specific use case. They can be published to the LM Studio Hub so users can apply them with one click.
+
+## lmstudio-planexe.json
+
+Optimizes any local LM Studio model for PlanExe's structured JSON pipeline:
+
+- **Temperature 0.2** — deterministic, schema-compliant output.
+- **Context window 32768 / max tokens 4096** — avoids llama_index's low defaults that silently truncate prompts.
+- **System prompt** — instructs the model to return only valid JSON matching the requested schema, with no markdown fences or filler.
+- **Thinking disabled** — turns off thinking/reasoning mode for models that support it (e.g. Qwen 3, GLM-4), since PlanExe needs raw JSON, not chain-of-thought.
+
+This preset complements the `llm_config/local.json` config, which tells PlanExe *how to call* LM Studio (via `OpenAILike` with `should_use_structured_outputs: true` for grammar-enforced JSON). The preset tells LM Studio *how to run* the model.

worker_plan/worker_plan_internal/assume/identify_purpose.py

@@ -108,7 +108,7 @@ def execute(cls, llm: LLM, user_prompt: str) -> 'IdentifyPurpose':
         if plan_purpose_instance is None:
             raise ValueError("LLM returned empty structured response (chat_response.raw is None).")
         json_response = plan_purpose_instance.model_dump()
-        purpose_value = plan_purpose_instance.purpose.value
+        purpose_value = plan_purpose_instance.purpose
         json_response['purpose'] = purpose_value
 
         metadata = dict(llm.metadata)

worker_plan/worker_plan_internal/markdown_util/escape_markdown.py

@@ -1,4 +1,4 @@
-"""
+r"""
 Escape special markdown characters to prevent syntax interference.
 
 This module provides functionality to escape markdown special characters that

worker_plan/worker_plan_internal/plan/data_collection.py

@@ -240,7 +240,7 @@ def convert_to_markdown(document_details: DocumentDetails) -> str:
             responsible_parties = DataCollection._format_bullet_list(data_collection_item.responsible_parties)
             rows.append(f"\n### Responsible Parties\n\n{responsible_parties}")
 
-            assumption_list = [f"**{item.sensitivity_score.human_readable()}:** {item.assumption}" for item in data_collection_item.assumptions]
+            assumption_list = [f"**{item.sensitivity_score.capitalize()}:** {item.assumption}" for item in data_collection_item.assumptions]
             assumptions = DataCollection._format_bullet_list(assumption_list)
             rows.append(f"\n### Assumptions\n\n{assumptions}")
 

worker_plan/worker_plan_internal/team/enrich_team_members_with_contract_type.py

@@ -166,7 +166,7 @@ def cleanup_and_merge_with_team_members(document_details: DocumentDetails, team_
             id = team_member['id']
             enriched_team_member = id_to_enriched_team_member.get(id)
             if enriched_team_member:
-                team_member['contract_type'] = enriched_team_member.contract_type.value
+                team_member['contract_type'] = enriched_team_member.contract_type
                 team_member['contract_type_justification'] = enriched_team_member.justification
         return result_team_member_list