feat: resolve sandbox file paths and show tool description in status

- Extract resolve_sandbox_script utility in sandbox_tool.py to detect
file paths and read content from sandbox via cat
- Resolve script paths before sending approval_required event so the
frontend receives actual file content for display and editing
- Show tool description arg in status chip and activity bar instead of
generic "running" / "Running bash" text

agent/core/agent_loop.py

@@ -491,6 +491,16 @@ class Handlers:
                             tool_args = json.loads(tc.function.arguments)
                         except (json.JSONDecodeError, TypeError):
                             tool_args = {}
+
+                        # Resolve sandbox file paths for hf_jobs scripts so the
+                        # frontend can display & edit the actual file content.
+                        if tool_name == "hf_jobs" and isinstance(tool_args.get("script"), str):
+                            from agent.tools.sandbox_tool import resolve_sandbox_script
+                            sandbox = getattr(session, "sandbox", None)
+                            content, _ = await resolve_sandbox_script(sandbox, tool_args["script"])
+                            if content:
+                                tool_args = {**tool_args, "script": content}
+
                         tools_data.append(
                             {
                                 "tool": tool_name,

agent/tools/sandbox_tool.py

@@ -12,6 +12,7 @@ a cpu-basic sandbox is auto-created (no approval needed).
 from __future__ import annotations
 
 import asyncio
+import shlex
 from typing import Any
 
 from huggingface_hub import HfApi, SpaceHardware
@@ -19,6 +20,37 @@ from huggingface_hub import HfApi, SpaceHardware
 from agent.core.session import Event
 from agent.tools.sandbox_client import Sandbox
 
+
+def _looks_like_path(script: str) -> bool:
+    """Return True if the script string looks like a file path (not inline code)."""
+    return (
+        isinstance(script, str)
+        and script.strip() == script
+        and not any(c in script for c in "\r\n\0")
+        and (script.startswith("/") or script.startswith("./") or script.startswith("../"))
+    )
+
+
+async def resolve_sandbox_script(sandbox: Any, script: str) -> tuple[str | None, str | None]:
+    """Read a file from the sandbox if *script* looks like a path.
+
+    Returns:
+        (content, error) — content is the file text on success,
+        error is a message on failure.  Both None means *script*
+        is not a path (caller should use it as-is).
+    """
+    if not sandbox or not _looks_like_path(script):
+        return None, None
+    try:
+        result = await asyncio.to_thread(
+            sandbox.bash, f"cat {shlex.quote(script)}"
+        )
+        if result.success and result.output:
+            return result.output, None
+        return None, f"Failed to read {script} from sandbox: {result.error}"
+    except Exception as e:
+        return None, f"Failed to read {script} from sandbox: {e}"
+
 # ── Tool name mapping (short agent names → Sandbox client names) ──────
 
 

frontend/src/components/Chat/ActivityStatusBar.tsx

@@ -21,7 +21,7 @@ function statusLabel(status: ActivityStatus): string {
   switch (status.type) {
     case 'thinking': return 'Thinking';
     case 'streaming': return 'Writing';
-    case 'tool': return TOOL_LABELS[status.toolName] || `Running ${status.toolName}`;
+    case 'tool': return status.description || TOOL_LABELS[status.toolName] || `Running ${status.toolName}`;
     case 'waiting-approval': return 'Waiting for approval';
     default: return '';
   }

frontend/src/hooks/useAgentChat.ts

@@ -195,8 +195,8 @@ export function useAgentChat({ sessionId, isActive, onReady, onError, onSessionD
       onStreaming: () => {
         if (isActiveRef.current) setActivityStatus({ type: 'streaming' });
       },
-      onToolRunning: (toolName: string) => {
-        if (isActiveRef.current) setActivityStatus({ type: 'tool', toolName });
+      onToolRunning: (toolName: string, description?: string) => {
+        if (isActiveRef.current) setActivityStatus({ type: 'tool', toolName, description });
       },
     }),
     // sessionId is the only real dependency — Zustand setters are stable

frontend/src/lib/ws-chat-transport.ts

@@ -35,7 +35,7 @@ export interface SideChannelCallbacks {
   /** Called when assistant text starts streaming */
   onStreaming: () => void;
   /** Called when a tool starts running (non-plan) */
-  onToolRunning: (toolName: string) => void;
+  onToolRunning: (toolName: string, description?: string) => void;
 }
 
 // ---------------------------------------------------------------------------
@@ -499,7 +499,7 @@ export class WebSocketChatTransport implements ChatTransport<UIMessage> {
         this.enqueue({ type: 'tool-input-start', toolCallId, toolName, dynamic: true });
         this.enqueue({ type: 'tool-input-available', toolCallId, toolName, input: args, dynamic: true });
 
-        this.sideChannel.onToolRunning(toolName);
+        this.sideChannel.onToolRunning(toolName, (args as Record<string, unknown>)?.description as string | undefined);
         this.sideChannel.onToolCallPanel(toolName, args as Record<string, unknown>);
         break;
       }

frontend/src/store/agentStore.ts

@@ -41,7 +41,7 @@ export interface LLMHealthError {
 export type ActivityStatus =
   | { type: 'idle' }
   | { type: 'thinking' }
-  | { type: 'tool'; toolName: string }
+  | { type: 'tool'; toolName: string; description?: string }
   | { type: 'waiting-approval' }
   | { type: 'streaming' };