langchain-ai/deepagents · error · ValueError

trusted_compaction_tool must be named compact_conversation

Error message

trusted_compaction_tool must be named compact_conversation

What it means

Like the ask_user check, the optional `trusted_compaction_tool` passed to `AutoModeHITLMiddleware` must be named exactly `compact_conversation`, because the middleware dispatches on that name to treat compaction as a trusted operation. `__init__` raises this ValueError if a tool with a different name is supplied.

Source

Thrown at libs/code/deepagents_code/auto_mode.py:2168

            trusted_ask_user_tool: Built-in tool allowed to create consent receipts.
            trusted_compaction_tool: Built-in tool that performs conversation
                compaction.

        Raises:
            ValueError: If a trusted tool has an unexpected name.
        """
        if (
            trusted_ask_user_tool is not None
            and trusted_ask_user_tool.name != "ask_user"
        ):
            msg = "trusted_ask_user_tool must be named ask_user"
            raise ValueError(msg)
        if (
            trusted_compaction_tool is not None
            and trusted_compaction_tool.name != "compact_conversation"
        ):
            msg = "trusted_compaction_tool must be named compact_conversation"
            raise ValueError(msg)
        # The review deadline is a security control's budget, so reject a
        # nonsensical one at the boundary rather than trusting every caller:
        # a zero, negative, or NaN timeout expires immediately, silently turning
        # Auto into "deny every gated batch, then escalate". Callers that read
        # user config go through `resolve_auto_classifier_timeout`, which bounds
        # the value; this guards programmatic construction.
        for name, budget in (
            ("classifier_timeout_seconds", classifier_timeout_seconds),
            (
                "classifier_construction_timeout_seconds",
                classifier_construction_timeout_seconds,
            ),
        ):
            if not math.isfinite(budget) or budget <= 0:
                msg = f"{name} must be a positive finite number, got {budget!r}"
                raise ValueError(msg)
        interrupt_map = dict(interrupt_on)
        interrupt_map["create_temp_artifact"] = {

View on GitHub (pinned to a1af029e6e)

Solutions

1. Set the tool's `.name` to exactly 'compact_conversation' before passing it.
2. Leave `trusted_compaction_tool` unset (None) to use the default built-in compaction tool.
3. Check argument order: ensure the ask_user tool is bound to `trusted_ask_user_tool` and the compaction tool to `trusted_compaction_tool`.

Example fix

// before
mw = AutoModeHITLMiddleware(trusted_compaction_tool=compaction_tool.with_config(name='summarize_thread'))
// after
mw = AutoModeHITLMiddleware(trusted_compaction_tool=compaction_tool)  # name must be 'compact_conversation'

tool = get_trusted_compaction_tool()
if tool is not None and tool.name != 'compact_conversation':
    raise ValueError(f"fix: trusted_compaction_tool name is {tool.name!r}")

def is_trusted_compaction_tool(tool: BaseTool | None) -> bool:
    return tool is None or tool.name == 'compact_conversation'

try:
    mw = AutoModeHITLMiddleware(trusted_compaction_tool=tool)
except ValueError as e:
    if 'trusted_compaction_tool must be named compact_conversation' in str(e):
        tool.name = 'compact_conversation'
        mw = AutoModeHITLMiddleware(trusted_compaction_tool=tool)
    else:
        raise

When it happens

Trigger: Constructing `AutoModeHITLMiddleware(trusted_compaction_tool=<tool>)` where `<tool>.name != 'compact_conversation'` — e.g. a renamed compaction tool or a wrapper object with a different `.name`.

Common situations: Customizing the compaction tool's name for clarity; wrapping the standard compaction tool so `.name` changes; mixing up the order of the two trusted-tool constructor arguments.

Related errors

• trusted_ask_user_tool must be named ask_user
• Invalid agent name: {agent_name!r} is reserved for dcode's o
• tool_call_id must not be empty
• deny decisions require a reason
• trusted thread, turn, and tool-call identity are required

AI-assisted analysis of langchain-ai/deepagents@a1af029e6e (2026-08-29). Data as JSON: /api/errors/fc91271e7cccb597. Report an issue: GitHub.