fix(backend/copilot): use working_dir in prompt examples instead of hardcoded /home/user

The storage supplement template and _persist_and_summarize had hardcoded
/home/user/ paths in save_to_path examples. In local (bubblewrap) mode
the working dir is /tmp/copilot-<session>/, not /home/user/. Use the
{working_dir} template variable in prompting.py and a generic
<working_dir> placeholder in base.py so the model gets correct paths
regardless of execution mode.

autogpt_platform/backend/backend/copilot/prompting.py

@@ -221,7 +221,7 @@ def _build_storage_supplement(
 
 ### Moving files between storages
 - **{file_move_name_1_to_2}**: `write_workspace_file(filename="output.json", source_path="/path/to/local/file")`
-- **{file_move_name_2_to_1}**: `read_workspace_file(path="tool-outputs/data.json", save_to_path="/home/user/data.json")`
+- **{file_move_name_2_to_1}**: `read_workspace_file(path="tool-outputs/data.json", save_to_path="{working_dir}/data.json")`
 
 ### File persistence
 Important files (code, configs, outputs) should be saved to workspace to ensure they persist.
@@ -235,7 +235,7 @@ To read these files, use `Read` — it reads from the host filesystem.
 When a tool output contains `<tool-output-truncated workspace_path="...">`, the
 full output is in workspace storage (NOT on the local filesystem). To access it:
 - Use `read_workspace_file(path="...", offset=..., length=50000)` for reading sections.
-- To process in the sandbox, use `read_workspace_file(path="...", save_to_path="/home/user/file.json")` first, then use `bash_exec` on the local copy.
+- To process in the sandbox, use `read_workspace_file(path="...", save_to_path="{working_dir}/file.json")` first, then use `bash_exec` on the local copy.
 {_SHARED_TOOL_NOTES}{extra_notes}"""
 
 

autogpt_platform/backend/backend/copilot/tools/base.py

@@ -92,9 +92,9 @@ async def _persist_and_summarize(
         f"Use read_workspace_file("
         f'path="{file_path}", offset=<char_offset>, length=50000) '
         f"to read any section. "
-        f"To process the file in the sandbox, use "
+        f"To process the file in the sandbox/working dir, use "
         f"read_workspace_file("
-        f'path="{file_path}", save_to_path="/home/user/{tool_call_id}.json") '
+        f'path="{file_path}", save_to_path="<working_dir>/{tool_call_id}.json") '
         f"first, then use bash_exec to work with the local copy."
     )
     # Use workspace:// prefix so the model doesn't confuse the workspace path