If the `stdout` "done marker" arrives before the `stderr` output is
enqueued, the method returns early without capturing the `stderr` line.
The two reader threads run independently with no synchronization
guaranteeing `stderr` arrives before the done marker.
In environments with Python 3.10, timing differences can cause the
`stdout` marker to win the race, resulting in `<no output>` instead of
`[stderr]` error.
Observed as a flaky test on `test_stderr_output_labeling` in CI:
```shell
FAILED tests/unit_tests/agents/middleware/implementations/test_shell_tool.py::test_stderr_output_labeling - AssertionError: assert '[stderr] error' in '<no output>'
```
Use of the fixture `_base_vcr_config` is deprecated with alternative
function `base_vcr_config()`
This way:
* we don't need to import `_base_vcr_config` seen as unused (which leads
to ruff violations PLC0414 and F811)
* we don't need to make a copy since a new dict is created at each
function invocation
Co-authored-by: Mason Daugherty <mason@langchain.dev>
With this we get the correct types for `_runnable_support` annotated
functions.
* return list[BaseMessage] when messages is not None
* return Runnable when messages is None
* typing of function args
This PR changes how we find the cutoff for summarization, summarizing
content more eagerly if the initial cutoff point isn't safe (ie, would
break apart AI + tool message pairs)
This new algorithm is quite simple - it looks at the initial cutoff
point, if it's not safe, moves forward through the message list until it
finds the first non tool message.
For example:
```
H
AI
TM
--- theoretical cutoff based keep=('messages', 3)
TM
AI
TM
```
```
H
AI
TM
TM
--- actual cutoff, more aggressive summarization
AI
TM
```
The `HumanInTheLoopMiddleware` is missing a type annotation for the
context schema. Without the fix in this PR, the following code does not
type check:
```
graph = create_agent(
"gpt-5",
tools=[send_email_tool, read_email_tool],
middleware=[
HumanInTheLoopMiddleware(
interrupt_on={
# Require approval or rejection for sending emails
"send_email_tool": {
"allowed_decisions": ["approve", "reject"],
},
# Auto-approve reading emails
"read_email_tool": False,
}
),
],
context_schema=ContextSchema,
)
```
```
Argument of type "list[HumanInTheLoopMiddleware]" cannot be assigned to parameter "middleware" of type "Sequence[AgentMiddleware[StateT_co@create_agent, ContextT@create_agent]]" in function "create_agent"
"HumanInTheLoopMiddleware" is not assignable to "AgentMiddleware[AgentState[Unknown], ContextSchema | None]"
Type parameter "ContextT@AgentMiddleware" is invariant, but "None" is not the same as "ContextSchema | None"
```
* `create_agent`'s `system_prompt` allows `str | SystemMessage`
* added `system_message: SystemMessage` on `ModelRequest`
* `ModelRequest.system_prompt` is a function of `system_message.text`,
now deprecated
* disallow setting `system_prompt` and `system_message`
* `ModelRequest.system_prompt` can still be set (w/ custom setattr) for
custom backwards compat, but the updates just get propogated to the
`ModelRequest.system_message`
---------
Co-authored-by: Chester Curme <chester.curme@gmail.com>
Closes https://github.com/langchain-ai/langchain/issues/33983
* Adds `ModelRetryMiddleware` modeled after `ToolRetryMiddleware`
* Uses `on_failure` modes of `error` and `continue` to match the
`exit_behavior` modes of model + tool call limit middleware
* In a backwards compatible manner, aligns the API of
`ToolRetryMiddleware`'s `on_failure` with the above
* Centralize common "retry" utils across these middlewares
* use `override` instead of directly patching things on `ModelRequest`
* rely on `ToolNode` for execution of tools related to said middleware,
using `wrap_model_call` to inject the relevant claude tool specs +
allowing tool node to forward them along to corresponding langchain tool
implementations
* making the same change for the native shell tool middleware
* allowing shell tool middleware to specify a name for the shell tool
(negative diff then for claude bash middleware)
long term I think the solution might be to attach metadata to a tool to
map the provider spec to a langchain implementation, which we could also
take some lessons from on the MCP front.
middleware tests have gotten quite unwieldy, major restructuring, sets
the stage for coverage increase
this is super hard to review -- as a proof that we've retained important
tests, I ran coverage on `master` and this branch and confirmed
identical coverage.
* moving all middleware related tests to `agents/middleware` folder
* consolidating related test files
* adding coverage utility to makefile
