Add huggingface tools and improve execution modes

backend/configs/prompts/system_prompt.yaml

@@ -11,35 +11,39 @@ prompt: |
 
   {% if mode == "plan" %}
   ## CURRENT MODE: PLAN
-  
+
   You are in **Plan mode**. Your job is to understand the task, ask clarifying
   questions, gather context, and produce a comprehensive plan.
-
-  **Available tools**: ask_user, plan_tool, read_file, list_dir, glob_files,
-  grep_search, web_search, papers, github_search, github_read_file, github_read_repo,
-  github_search_repos, github_get_readme
-
-  **NOT available**: writing, research sub-agent, sandbox/code execution tools.
-  Calls to unavailable tools will be rejected.
-
+
+  **Available tools**: ask_user, plan_tool, read (local files), web_search,
+  papers, github_read_file, github_find_examples, github_search_repos,
+  github_get_readme, github_list_repos, hf_search_models, hf_model_info,
+  hf_search_datasets, hf_dataset_info, hf_read_file, compute_list,
+  compute_plan, compute_probe, workspace
+
+  **NOT available**: bash, write, edit, writing, research sub-agent,
+  sandbox/code execution, compute_select, compute_sync_up, compute_sync_down.
+  Calls to unavailable tools will be **rejected by the system**.
+
   **Rules**:
   1. Ask clarifying questions using `ask_user` before making assumptions
-  2. Search the web, papers, and code repos to gather context
-  3. Create a structured plan using `plan_tool` with clear, actionable tasks
-  4. The plan is auto-saved as PLAN.md in resources — the user can see it
-  5. Do NOT execute any work — plan only
-  6. Do NOT write content, run code, or make changes
-  7. Be thorough in your plan — it will be the blueprint for Execute mode
-
+  2. Search the web, papers, GitHub, and Hugging Face to gather context
+  3. Read local project files with `read` to understand the codebase
+  4. Create a structured plan using `plan_tool` with clear, actionable tasks
+  5. The plan is auto-saved as PLAN.md in resources — the user can see it
+  6. Do NOT execute any work — plan only
+  7. Do NOT write content, run code, or make changes
+  8. Be thorough in your plan — it will be the blueprint for Execute mode
+
   {% elif mode == "execute" %}
   ## CURRENT MODE: EXECUTE
-  
+
   You are in **Execute mode**. Your job is to follow the plan and do the work.
   Do NOT ask questions — just execute.
-  
+
   **Available tools**: ALL tools EXCEPT ask_user.
-  Calls to ask_user will be rejected.
-  
+  Calls to ask_user will be **rejected by the system**.
+
   **Rules**:
   1. Follow the task plan — check it with `plan_tool get` if unsure
   2. Work through tasks one at a time, marking them in_progress then completed
@@ -49,10 +53,10 @@ prompt: |
      decision and document it in your completion report
   6. Keep pushing through the task list until done or interrupted
   7. Generate completion reports for each task
-  
+
   {% else %}
-  ## CURRENT MODE: EXECUTE (default)
-  All tools available except ask_user. Execute the work.
+  ## CURRENT MODE: PLAN (default)
+  Plan only — ask questions, gather context, create plan. No execution.
   {% endif %}
 
   # Task Management

backend/openmlr/agent/loop.py

@@ -47,8 +47,15 @@ async def _run_agent(session: Session, tool_router, user_message: str, mode: str
     if session.pending_approval:
         session.pending_approval = None
 
-    # Set the mode on the tool router for strict enforcement
-    effective_mode = mode if mode in ("plan", "execute") else "execute"
+    # Set the mode on the tool router for strict enforcement.
+    # Default to plan (safe) if mode is missing or invalid.
+    # If mode is explicitly provided, use it and persist on session.
+    # If not provided (e.g. approval continuation), fall back to session's stored mode.
+    if mode in ("plan", "execute"):
+        effective_mode = mode
+        session.current_mode = mode
+    else:
+        effective_mode = session.current_mode  # preserved from the last explicit mode
     tool_router.set_mode(effective_mode)
 
     # Inject per-message mode hint (short reinforcement of system prompt rules)

backend/openmlr/agent/session.py

@@ -25,6 +25,9 @@ class Session:
     # Cancellation
     _cancelled: asyncio.Event = field(default_factory=asyncio.Event)
 
+    # Mode tracking (plan/execute) — persists across approval continuations
+    current_mode: str = "plan"
+
     # Approval flow
     pending_approval: dict | None = None
 

backend/openmlr/models.py

@@ -1,7 +1,7 @@
 """Pydantic models for API requests and responses."""
 
 from datetime import datetime
-from typing import Any
+from typing import Any, Literal
 
 from pydantic import BaseModel, Field
 
@@ -71,7 +71,9 @@ class ConversationDetail(BaseModel):
 
 class MessageSend(BaseModel):
     message: str
-    mode: str | None = None  # plan, research, write — per-message mode override
+    mode: Literal["plan", "execute"] | None = (
+        None  # per-message mode; only plan or execute accepted
+    )
 
 
 class ApprovalRequest(BaseModel):

backend/openmlr/routes/settings.py

@@ -64,6 +64,7 @@ async def update_setting(
             "openalex_api_key": "OPENALEX_API_KEY",
             "modal_token_id": "MODAL_TOKEN_ID",
             "modal_token_secret": "MODAL_TOKEN_SECRET",
+            "hf_token": "HF_TOKEN",
         }
         env_key = env_key_map.get(key)
         if env_key and isinstance(value, str):
@@ -127,6 +128,7 @@ def _is_provider_configured(provider_id: str, provider_settings: dict) -> bool:
         "semantic_scholar": "SEMANTIC_SCHOLAR_API_KEY",
         "openalex": "OPENALEX_API_KEY",
         "modal": "MODAL_TOKEN_ID",
+        "huggingface": "HF_TOKEN",
     }
     env_key = env_map.get(provider_id)
     if env_key and os.environ.get(env_key):
@@ -143,6 +145,7 @@ def _is_provider_configured(provider_id: str, provider_settings: dict) -> bool:
         "semantic_scholar": "semantic_scholar_api_key",
         "openalex": "openalex_api_key",
         "modal": "modal_token_id",
+        "huggingface": "hf_token",
     }.get(provider_id)
     if setting_key and provider_settings.get(setting_key):
         return True
@@ -258,6 +261,14 @@ async def list_providers(
             "categories": ["compute"],
             "docs_url": "https://modal.com/docs",
         },
+        {
+            "id": "huggingface",
+            "name": "Hugging Face",
+            "key_env": "HF_TOKEN",
+            "configured": _is_provider_configured("huggingface", provider_settings),
+            "categories": ["models", "papers"],
+            "docs_url": "https://huggingface.co/docs/hub/security-tokens",
+        },
     ]
 
     # Add custom providers
@@ -759,6 +770,7 @@ async def save_config(
         "OPENALEX_API_KEY",
         "MODAL_TOKEN_ID",
         "MODAL_TOKEN_SECRET",
+        "HF_TOKEN",
     }
 
     body = await request.json()

backend/openmlr/tasks/agent_tasks.py

@@ -116,7 +116,7 @@ async def _async_process_message(
     # Build and set system prompt
     session.context_manager.system_prompt = build_system_prompt(
         tool_specs=tool_router.get_raw_specs(),
-        mode=mode or "general",
+        mode=mode if mode in ("plan", "execute") else "plan",
         username="user",
     )
 

backend/openmlr/tools/ask_user.py

@@ -66,8 +66,8 @@ def create_ask_user_tool() -> ToolSpec:
                 },
                 "suggest_mode": {
                     "type": "string",
-                    "description": "If confident, suggest the user switch to this mode after answering (e.g. 'research', 'write')",
-                    "enum": ["research", "write"],
+                    "description": "If the plan is ready and the user should start executing, set this to 'execute' to suggest switching to Execute mode after answering.",
+                    "enum": ["execute"],
                 },
             },
             "required": ["questions"],