LangChain Agents 2 · Tools, routing, and the failure you have to handle

[akash-coded]

LangChain Agents · Exercise 2

Language: Python Topics: tools with @tool, the schema the model reads, tool selection by docstring, multi-tool loops, tool errors Level: Foundational to Applied

Answers are letters, pairs like 1-C, T/F, or a sequence. Still reading code, not writing it.

---

Q1 · (2 pts) When the model decides whether to call lookup_pnr, it reads:

• A) the function body
• B) the name, the docstring, and the argument schema
• C) the return value
• D) the whole file

---

Q2 · Dead on arrival (2 pts) Copied from a 2023 blog. Which import no longer belongs in LangChain 1.0?

• A) from langchain.agents import create_agent
• B) from langchain.agents import AgentExecutor
• C) from langchain.tools import tool
• D) from langgraph.graph import StateGraph

---

Q3 · Read the tool, match the schema (4 pts)

@tool
def disruption_reason(flight: str) -> str:
    '''Return why a flight was cancelled or delayed, given its flight number.'''
    ...

Field		Value the model sees
1. name		A. {"flight": {"type": "string"}}
2. description		B. disruption_reason
3. args		C. Return why a flight was cancelled or delayed, given its flight number.
		D. the function body

---

Q4 · Code walkthrough, match line to role (4 pts)

L1  @tool
L2  def lookup_pnr(pnr: str) -> str:
L3      '''Return the booking status for a PNR.'''
L4      return records.get(pnr, "PNR not found")
L5  agent = create_agent(model, tools=[lookup_pnr], system_prompt="You are TravelMind.")

Line		What it does
1. L3		A. registers the tool with the agent
2. L4		B. the description the model reads to pick this tool
3. L5		C. returns a safe result even on a bad PNR
		D. sets the model temperature

---

Q5 · Trace the two-tool flow (3 pts)

flowchart TD
    U[cancelled, JX48Q2, options?] --> M1[model]
    M1 -->|call lookup_pnr| T1[booking record]
    T1 --> M2[model]
    M2 -->|call search_flights| T2[two flights]
    T2 --> M3[model]
    M3 --> A[final answer]

Loading

How many times does the model node run before the final answer?

• A) 1 B) 2 C) 3 D) 4

---

Q6 · One step is out of order (3 pts) This flow searches before it knows the booking.

flowchart TD
    U[cancelled, JX48Q2, options?] -->|a| M1[model]
    M1 -->|b call search_flights| T2[two flights]
    T2 -->|c| M2[model]
    M2 -->|d call lookup_pnr| T1[booking record]

Loading

Which tagged edge is the problem?

• A) a B) b C) c D) d

---

Q7 · Order the trace (3 pts) The model calls lookup_pnr, then answers. Order the four messages.

• a) AIMessage requesting lookup_pnr
• b) HumanMessage with the question
• c) AIMessage with the grounded answer
• d) ToolMessage with the booking record

---

Q8 · Predict the output (3 pts)

model = ScriptedChatModel(responses=[
    AIMessage(content="", tool_calls=[{"name": "lookup_pnr", "args": {"pnr": "ZZZZZZ"}, "id": "e", "type": "tool_call"}]),
    AIMessage(content="I could not find booking ZZZZZZ. Please recheck the PNR."),
])
agent = create_agent(model, tools=[lookup_pnr], system_prompt="You are TravelMind.")
result = agent.invoke({"messages": [{"role": "user", "content": "check ZZZZZZ"}]})
print(result["messages"][-1].content)

Assume lookup_pnr returns "PNR not found" for unknown PNRs. The printout is:

• A) PNR not found
• B) I could not find booking ZZZZZZ. Please recheck the PNR.
• C) an exception
• D) an empty string

---

Q9 · Spot the risky line (3 pts)

L1  @tool
L2  def rebook(pnr, flight):
L3      '''handle it'''
L4      return f"{pnr} moved to {flight}"

Which line most weakens the model's ability to pick this tool correctly?

• A) L1 B) L2 (no type hints) C) L3 (empty description) D) L4

---

Q10 · Pick the flow that grounds its answer (2 pts)

Option A

flowchart LR
    U[is JX48Q2 cancelled?] --> M[model] --> A[answer from memory]

Loading

Option B

flowchart LR
    U[is JX48Q2 cancelled?] --> M[model] --> T[lookup_pnr] --> M2[model] --> A[answer from the record]

Loading

---

Q11 · Match across frameworks (4 pts)

Strands		LangChain
1. @tool on a function		A. create_agent(model, tools=[...], system_prompt=)
2. Agent(model=, tools=[...])		B. @tool on a function
3. str(agent("..."))		C. result["messages"][-1].content after invoke
		D. response_format=Model

---

Q12 · Pick all that apply (3 pts) A good tool description:

• A) says what the tool does in plain terms
• B) is left blank to keep the schema small
• C) names the arguments' meaning when it is not obvious
• D) warns if the action is irreversible
• E) uses the function name only, no sentence

---

Case study · Two tools, one wrong pick (5 pts)

Two tools ship together. A user types "sort out my booking."

flowchart TD
    U[sort out my booking] --> M{model picks a tool}
    M --> C[cancel_booking: permanently cancel, start a refund, irreversible]
    M --> H[handle: handle the booking]

Loading

Q13a (2 pts) Which tool is the model most likely to misfire on?

• A) cancel_booking, its description is too detailed
• B) handle, its description says nothing about what it does
• C) neither, the model always asks first
• D) both fail equally

Q13b (3 pts) The safest fix:

• A) delete cancel_booking
• B) give handle a precise description of exactly what it does
• C) rename handle to handle2
• D) raise the model temperature

---

Q14 · Pick the docstring that steers the model right (3 pts) For a search_flights(origin, dest, date) tool:

• A) '''search'''
• B) '''does flight stuff'''
• C) '''Find alternate flights between two airport codes on a given date.'''
• D) '''flights table'''

[akash-coded]

from typing import Any, List
from langchain_core.language_models.chat_models import BaseChatModel
from langchain_core.messages import AIMessage
from langchain_core.outputs import ChatResult, ChatGeneration
from langchain.agents import create_agent
from langchain.tools import tool


class ScriptedChatModel(BaseChatModel):
    '''Deterministic stand-in. Real framework, scripted replies, no credentials.'''
    responses: List[Any]
    idx: int = 0

    @property
    def _llm_type(self) -> str:
        return "scripted"

    def _generate(self, messages, stop=None, run_manager=None, **kwargs):
        reply = self.responses[min(self.idx, len(self.responses) - 1)]
        object.__setattr__(self, "idx", self.idx + 1)
        return ChatResult(generations=[ChatGeneration(message=reply)])

    def bind_tools(self, tools, **kwargs):
        return self


@tool
def lookup_pnr(pnr: str) -> str:
    '''Return the booking status for a PNR.'''
    return {"JX48Q2": "BLR-DEL cancelled, Rao, Gold tier"}.get(pnr, "PNR not found")


@tool
def search_flights(origin: str, dest: str) -> str:
    '''Find alternate flights between two airport codes.'''
    return "AI-506 09:40, AI-812 14:15"


# Grounded answer: lookup, then search, then reply. Three model turns.
agent = create_agent(
    ScriptedChatModel(responses=[
        AIMessage(content="", tool_calls=[{"name": "lookup_pnr", "args": {"pnr": "JX48Q2"}, "id": "a", "type": "tool_call"}]),
        AIMessage(content="", tool_calls=[{"name": "search_flights", "args": {"origin": "BLR", "dest": "DEL"}, "id": "b", "type": "tool_call"}]),
        AIMessage(content="Cancelled. Options: AI-506 09:40 or AI-812 14:15."),
    ]),
    tools=[lookup_pnr, search_flights],
    system_prompt="You are TravelMind.",
)
print(agent.invoke({"messages": [{"role": "user", "content": "JX48Q2 cancelled, options?"}]})["messages"][-1].content)

# Bad PNR: the tool returns a clean error, the model recovers instead of crashing.
recover = create_agent(
    ScriptedChatModel(responses=[
        AIMessage(content="", tool_calls=[{"name": "lookup_pnr", "args": {"pnr": "ZZZZZZ"}, "id": "e", "type": "tool_call"}]),
        AIMessage(content="I could not find booking ZZZZZZ. Please recheck the PNR."),
    ]),
    tools=[lookup_pnr],
    system_prompt="You are TravelMind.",
)
print(recover.invoke({"messages": [{"role": "user", "content": "check ZZZZZZ"}]})["messages"][-1].content)

[ksvivekanand86-source]

Q1. B
Q2. D
Q3. 1-B, 2-C, 3-A
Q4. L3-B, L4-C, L5-A
Q5. C

Q6. B
Q7. B, A, D, C
Q8. B
Q9. C
Q10. Option B
Q11. 1-B, 2-A, 3-C
Q12. A, C, D
Q13a. B
Q13b. B
Q14. C

[saneesh-ibs]

Q1 :B
Q2:B
Q3: 1-B, 2-C, 3-A
Q4: 1-B, 2-C, 3-A
Q5: C
Q6 : D
Q7: b-a-d-c
Q8 : B
Q9 : C
Q10:B
Q11: 1-B, 2-A, 3-C
Q12:A,C,D
Q13:B, B
Q14:C

[Vysakhan208853]

Q1 B
Q2 B
Q3 1-B, 2-C, 3-A
Q4 1-B, 2-C, 3-A
Q5 C
Q6 D
Q7 b -> a -> d -> c
Q8 B
Q9 C
Q10 B
Q11 1-B, 2-A, 3-C
Q12 A, C, D
Q13a B
Q13b B
Q14 C

[beenadinesh]

1.B
2.B
3. B / C / A
4. B / C / A
5. C
6. D
7. B,A,D,C
8. B
9. C
10. B
11. B/A/C
12. A
13. a) B b) B
14. C

[sudhin85]

Q1 -- B
Q2 -- B
Q3
1 --- B
2 --- C
3 --- A
Q4
1 --- B
2 --- C
3 --- A

Q5 -- C
Q6 -- B
Q7 -- b --- a --- d --- c
Q8 -- B
Q9 -- C
Q10 -- Option B
Q11
1 --- B
2 --- A
3 --- C

Q12 --- A, C, D
Q13a --- B
Q13b --- B
Q14 -- C

[NimeshNairIBS]

<style> </style>

Q1	B
Q2	D
Q3	1-B;2-C;3-A
Q4	L3-B; L4-C; L5-a
Q5	C-3
Q6	D
Q7	c
Q8	b
Q9	C
Q10	OptionB
Q11	1-B;2-A;3-C
Q12	A,C,E
Q13a	b
Q13b	B
q14	c

[krishhaan]

B
B
B C A
B C A
C
D
b a d c
B
L3
B
B A C
A C D
B B
C

[rishidaspm]

Q1 - B
Q2 - B
Q3 - 1-B, 2-C, 3-A
Q4 - 1-B, 2-C, 3-A
Q5 - C
Q6 - B
Q7 - b,a,d,c
Q8 - B
Q9 - C
Q10 - B
Q11 - 1-B, 2-A, 3-C
Q12 - A,C,D
Q13a - B
Q13b - B
Q14 - C

[vivek1506-ui]

1 :B
2:B
3: 1-B, 2-C, 3-A
4: 1-B, 2-C, 3-A
5: C
6 : D
7: b-a-d-c
8 : B
9 : C
10:B
11: 1-B, 2-A, 3-C
12:A,C,D
13:B, B
14:C