For remote agents for slack chat bot or something, how to make agent not return results in Files/Documents? I want all the features like offloading large tools etc..

[Yashwanth (yash025)]

I'm seeing response like this
`Business Unit: Platform
Time Window: March 22, 2026, 03:30-05:30 UTC
📚 Investigation Package Created

I've built you 7 comprehensive documents (~250 pages) with everything you need:
🚨 For Immediate Action:

•   START_HERE.md - 30-second quick start guide
•   INCIDENT_RUNBOOK.md - Emergency mitigation procedures with 5 immediate fix options
🔬 For Investigation:

•   INVESTIGATION_QUERIES_EXECUTABLE.md - 30+ copy-paste ready queries for Splunk, Prometheus, Kubernetes
•   API_GATEWAY_THROTTLING_INVESTIGATION.md - Complete technical deep-dive
📊 For Reporting:

•   EXECUTIVE_SUMMARY.md - Leadership briefing template
•   INDEX.md - Navigation hub
⚡ What To Do Right Now:`

Code for deep agent creation
agent = create_deep_agent( model=llm, tools=all_tools, system_prompt=system_prompt, subagents=a2a_subagents, checkpointer=checkpointer, backend=lambda rt: StateBackend(rt), name="sre_orchestrator", )

I'm using StateBackend only, still it creating documents and sharing it to the user.

[Harrison Chase (hwchase17)]

prompt it to not return them in documents! (im assuming you are talking about INDEX.md, EXECUTIVE_SUMMARY.md, etc)

[Yashwanth (yash025)]

yes right, i was checking if there are any other better options than prompting. Thanks!

[Yokubjon (yakubka)]

The "remote agent for Slack" producing document files (INDEX.md, EXECUTIVE_SUMMARY.md, etc.) is a model behavior issue, not a framework constraint. The agent is generating those files because its system prompt or skill instructions encourage it to produce structured outputs as artifacts.

The fix is prompt-level: add an explicit instruction to the system prompt that prohibits file creation and requires inline responses only:

IMPORTANT: Do not create, write, or reference any files or documents.
All responses must be inline text only. Never use the file creation tool.
Never produce INDEX.md, EXECUTIVE_SUMMARY.md, or any artifact files.
If you would normally create a file, return the content as plain text in the chat instead.

If the agent still creates files despite this, you may need to remove the file creation tools entirely from the tool set it has access to. In deepagents you can control which tools are available by filtering the tool list before passing it to create_deep_agent.

If you want to keep file creation for other use cases but prevent it for Slack specifically, wrap the file creation tool in a tool that checks the source channel before allowing the write:

@tool
def write_file_guarded(path: str, content: str, channel: str = "") -> str:
    if channel.startswith("slack"):
        return f"[File creation disabled in Slack. Content would have been written to {path}]"
    return write_file_actual(path, content)