PR #4750: fix: Bedrock tool calls extract empty arguments - uses wrong field name (#4748)

• Repository: crewAIInc/crewAI
• Author: devin-ai-integration[bot]
• State: open | merged: False
• Link: https://github.com/crewAIInc/crewAI/pull/4750

Description (problem / solution / changelog)

Summary

Fixes #4748 — AWS Bedrock tool calls always received empty arguments {} regardless of what the LLM provided.

Root cause: In _parse_native_tool_call (line 850 of crew_agent_executor.py), the expression:

func_args = func_info.get("arguments", "{}") or tool_call.get("input", {})

The default value "{}" is a truthy string, so Python's or short-circuits and never evaluates the Bedrock input field.

Fix: Remove the default values so None (falsy) is returned when the key is absent, allowing the or chain to fall through correctly:

func_args = func_info.get("arguments") or tool_call.get("input") or {}

This matches the already-correct pattern in extract_tool_call_info in agent_utils.py (line 1155).

Review & Testing Checklist for Human

• Verify the one-line fix doesn't regress OpenAI-format tool calls — when func_info["arguments"] is a valid JSON string, it should still be returned directly (truthy string passes through or correctly)
• Consider edge case: if func_info["arguments"] is an empty string "", the old code returned "{}" while the new code falls through to input. Confirm this is acceptable (empty string arguments are arguably malformed)
• Ideally validate with an actual Bedrock model (e.g. us.amazon.nova-pro-v1:0) that tool arguments now arrive correctly

Notes

• The companion function extract_tool_call_info in agent_utils.py already used the correct pattern — only the _parse_native_tool_call method in CrewAgentExecutor had the bug. Tests were added for both.
• 15 new unit tests added across two test files covering OpenAI, Bedrock, Anthropic, and edge-case formats.
• Requested by: João
• Link to Devin session

[!NOTE] Medium Risk Touches native tool-call parsing in CrewAgentExecutor, which can affect tool execution inputs across providers; however the change is small and covered by new regression/unit tests.

Overview Fixes a bug in CrewAgentExecutor._parse_native_tool_call where missing function.arguments defaulted to the truthy string "{}", preventing fallback to Bedrock-style input and causing empty tool arguments.

Adds regression/unit coverage for _parse_native_tool_call and extract_tool_call_info across OpenAI/Bedrock/Anthropic-style tool call shapes, plus minor test file cleanup.

^{Written by Cursor Bugbot for commit 9900a7954699bff30fc46c0ace457428702e64ca. This will update automatically on new commits. Configure here.}

Changed files

• lib/crewai/src/crewai/agents/crew_agent_executor.py (modified, +1/-1)
• lib/crewai/tests/agents/test_async_agent_executor.py (modified, +92/-1)
• lib/crewai/tests/utilities/test_agent_utils.py (modified, +130/-0)

PR #4764: fix(bedrock): extract tool arguments from Bedrock's input field

• Repository: crewAIInc/crewAI
• Author: giulio-leone
• State: open | merged: False
• Link: https://github.com/crewAIInc/crewAI/pull/4764

Description (problem / solution / changelog)

Problem

AWS Bedrock tool calls always receive empty arguments {} regardless of what the LLM provides. This affects all Bedrock models (Nova Pro, Nova Lite, Claude via Bedrock, etc.).

Bedrock tool call format:

{"toolUseId": "tooluse_abc", "name": "search_tool", "input": {"query": "AWS Bedrock features"}}

OpenAI tool call format:

{"id": "call_xyz", "function": {"name": "search_tool", "arguments": '{"query": "test"}'}}

Root Cause

In _parse_native_tool_call (line 850):

func_args = func_info.get("arguments", "{}") or tool_call.get("input", {})

When func_info (from tool_call.get("function", {})) is an empty dict (Bedrock has no function key), .get("arguments", "{}") returns the default string "{}". Since "{}" is truthy, the or operator never evaluates tool_call.get("input").

Result: Bedrock's actual arguments in input are silently dropped.

Fix

Use None as the sentinel so or falls through correctly:

func_args = (
    func_info.get("arguments")
    or tool_call.get("input")
    or "{}"
)

Verified Locally

=== Bedrock tool call ===
OLD CODE: '{}'       <-- BUG: returns empty string instead of args
NEW CODE: {'query': 'AWS Bedrock features'}  <-- FIXED: returns actual args

=== OpenAI tool call (regression) ===
OLD CODE: '{"query": "test"}'
NEW CODE: '{"query": "test"}'
MATCH: True          <-- No regression

Tests

Added 7 unit tests covering:

• Bedrock dict with input field (the reported bug)
• OpenAI dict with function.arguments (regression)
• Bedrock dict without function key
• Empty input dict
• Neither function nor input (fallback)
• OpenAI-style object with .function attribute
• Bedrock-style object with .name and .input attributes

All pass.

Fixes #4748

[!NOTE] Low Risk Low risk bugfix limited to native tool-call argument extraction; main risk is subtle behavior changes for dict-style tool calls when function.arguments is missing/empty, but new tests cover expected formats.

Overview Fixes _parse_native_tool_call to correctly extract tool arguments from AWS Bedrock-style tool calls by falling back to the dict’s input field when function.arguments is absent.

Adds focused unit tests covering Bedrock dict/object tool calls, OpenAI regression cases, and fallback behavior when no arguments are provided.

^{Written by Cursor Bugbot for commit 310f000b4f34699934e5352be4b06ded899fa7bc. This will update automatically on new commits. Configure here.}

Changed files

• lib/crewai/src/crewai/agents/crew_agent_executor.py (modified, +5/-1)
• lib/crewai/tests/agents/test_bedrock_tool_args.py (added, +168/-0)

PR #4765: fix: extract Bedrock tool arguments from 'input' field correctly

• Repository: crewAIInc/crewAI
• Author: atian8179
• State: open | merged: False
• Link: https://github.com/crewAIInc/crewAI/pull/4765

Description (problem / solution / changelog)

Problem

When using AWS Bedrock models, all tool calls receive empty arguments {} regardless of what the LLM provides.

The bug is in crew_agent_executor.py:

func_args = func_info.get('arguments', '{}') or tool_call.get('input', {})

.get('arguments', '{}') returns the string '{}' when the key is absent. Since '{}' is truthy, the or operator never evaluates tool_call.get('input', {}), so Bedrock's input field is always ignored.

Solution

Use a sentinel object to distinguish "key absent" from "key is empty string". Only fall back to Bedrock's input field when no valid OpenAI-style arguments are present.

Fixes #4748

[!NOTE] Low Risk Low risk, small change isolated to native tool-call parsing; behavior only differs when OpenAI-style function.arguments is absent/empty, enabling Bedrock input to be used instead.

Overview Fixes native tool-call argument extraction for Bedrock-style responses so tools receive the actual payload from the input field when OpenAI-style function.arguments is missing or effectively empty.

Updates _parse_native_tool_call to use a sentinel to distinguish “key absent” from default '{}', preventing the previous truthy default from masking Bedrock arguments.

^{Written by Cursor Bugbot for commit 08a1c00f221ac7d4fd89befbf31ae39c74346855. This will update automatically on new commits. Configure here.}

Changed files

• lib/crewai/src/crewai/agents/crew_agent_executor.py (modified, +9/-1)

PR #4769: fix: extract Bedrock tool arguments from input field

• Repository: crewAIInc/crewAI
• Author: MaxwellCalkin
• State: open | merged: False
• Link: https://github.com/crewAIInc/crewAI/pull/4769

Description (problem / solution / changelog)

Summary

Fixes #4748.

When using AWS Bedrock models, tool calls always receive empty arguments {} because the fallback to Bedrock's input field never executes.

Root cause: func_info.get("arguments", "{}") returns the string "{}" (truthy) when no arguments key exists, so the or branch for tool_call.get("input", {}) is never evaluated.

Fix

# Before (broken):
func_args = func_info.get("arguments", "{}") or tool_call.get("input", {})

# After (fixed):
func_args = func_info.get("arguments") or tool_call.get("input", "{}")

Using None as the default for get("arguments") makes the or fallback work correctly. The "{}" default is moved to the Bedrock input fallback as the final default.

lib/crewai/src/crewai/agents/crew_agent_executor.py line 850 — one-line fix.

[!NOTE] Low Risk Very small, isolated change to native tool-call parsing; risk is limited to how tool arguments are extracted for dict-style (Bedrock) tool calls.

Overview Fixes native tool-call argument extraction for Bedrock-style responses by changing _parse_native_tool_call to only use function.arguments when it exists, otherwise falling back to the tool call’s input field (defaulting to the JSON string {}). This prevents missing arguments from incorrectly short-circuiting the fallback and producing empty tool args.

^{Written by Cursor Bugbot for commit fd59c902c1270f57e3c0ed220b52733280f2f77e. This will update automatically on new commits. Configure here.}

Changed files

• lib/crewai/src/crewai/agents/crew_agent_executor.py (modified, +1/-1)

PR #4803: fix: extract Bedrock tool arguments from input field correctly

• Repository: crewAIInc/crewAI
• Author: alvinttang
• State: closed | merged: False
• Link: https://github.com/crewAIInc/crewAI/pull/4803

Description (problem / solution / changelog)

Summary

Fixes #4748 - AWS Bedrock tool calls extract empty arguments due to wrong default value in get().

All AWS Bedrock models (Nova Pro, Nova Lite, Claude via Bedrock, etc.) receive empty {} arguments for every tool call, regardless of what the LLM actually provides. This causes tools to fail or return incorrect results.

Root cause

In CrewAgentExecutor._parse_native_tool_call() (line 850):

# BEFORE (buggy)
func_args = func_info.get("arguments", "{}") or tool_call.get("input", {})

func_info.get("arguments", "{}") returns the string "{}" when "arguments" is absent (Bedrock format). Since "{}" is truthy, Python's or operator never evaluates tool_call.get("input", {}), which is where Bedrock stores its arguments.

The correct pattern already exists in utilities/agent_utils.py line 1155:

func_args = func_info.get("arguments") or tool_call.get("input") or {}

Fix

One-line change: remove the default value "{}" from get("arguments") so it returns None when absent, allowing the or-chain to fall through to Bedrock's "input" field:

# AFTER (fixed)
func_args = func_info.get("arguments") or tool_call.get("input") or {}

Test plan

• Verify Bedrock tool call format {"name": "search_tool", "input": {"query": "test"}} correctly extracts arguments
• Verify OpenAI tool call format {"function": {"name": "search_tool", "arguments": {"query": "test"}}} still works
• Verify empty tool calls (no arguments in either format) default to {}
• Run existing test suite: pytest lib/crewai/tests/agents/

🤖 Generated with Claude Code

[!NOTE] Medium Risk Touches core tool-call argument extraction used during agent execution; while the change is small and targeted, it can affect tool invocation behavior across providers when arguments is absent or empty.

Overview Fixes native tool-call parsing for dict-based providers (notably AWS Bedrock) by no longer defaulting missing function.arguments to the truthy string "{}", allowing _parse_native_tool_call() to fall through to tool_call["input"] and correctly pass real tool arguments.

This changes the default behavior for tool calls with no provided args to return an empty dict ({}) rather than a stringified empty object, improving consistency across provider formats.

^{Written by Cursor Bugbot for commit 4d8f15f4284bcfa659f486cb404bca8b716f9be0. This will update automatically on new commits. Configure here.}

Changed files

• lib/crewai/src/crewai/agents/crew_agent_executor.py (modified, +1/-1)