Task 3.3: Implement Tool Result Status Auto-Detection for BedrockModelProvider Implementation by afarntrog · Pull Request #89 · strands-agents/sdk-typescript

afarntrog · 2025-10-29T16:43:46Z

Overview

Implements dynamic logic to automatically determine whether to include the status field in tool results based on the Bedrock model ID, matching the Python SDK's auto-detection behavior.

Resolves: #24

Changes

Interface and Configuration

Added includeToolResultStatus?: 'auto' | boolean to BedrockModelConfig interface
Added MODELS_INCLUDE_STATUS constant array with Claude model pattern
Default behavior is 'auto' mode which automatically detects based on model ID

Implementation

Implemented _shouldIncludeToolResultStatus() private method with:
- Explicit true/false handling
- Auto-detection logic checking model ID against patterns
- Debug logging for auto-detection decisions
Updated _formatContentBlock() method to conditionally include status field based on detection result

Testing

Added 8 comprehensive unit tests covering:
- Explicit true configuration (always includes status)
- Explicit false configuration (never includes status)
- 'auto' mode with Claude models (includes status)
- 'auto' mode with non-Claude models (excludes status)
- undefined (default) with Claude models (includes status)
- undefined (default) with non-Claude models (excludes status)
- Debug logging verification for auto-detection
All tests passing (45 total in bedrock.test.ts)

Implementation Notes

Model Pattern Matching

Uses substring matching with .includes() to detect Claude models
Pattern: 'anthropic.claude' matches all Claude model variants
Default model (global.anthropic.claude-sonnet-4-5-20250929-v1:0) correctly matches and includes status

Type Safety

Implemented conditional return statements to satisfy TypeScript's strict type checking
Avoids type assertion by having separate return paths for with/without status

Integration Tests

Integration tests for tool result formatting would require complex multi-turn tool invocation cycles. The comprehensive unit tests verify the formatting logic for all configuration modes and validate that the output matches the Bedrock API specification.

Exit Criteria

✅ includeToolResultStatus config option supports 'auto' | boolean
✅ Model pattern constants defined with TSDoc
✅ Auto-detection logic implemented with proper typing
✅ Tool result formatting uses auto-detection conditionally
✅ Debug logging for auto-detection decisions
✅ Unit tests cover all configuration modes (true, false, 'auto', undefined)
✅ TSDoc documentation explains auto-detection behavior
✅ Default value is 'auto' (via ?? 'auto' in the method)
✅ All tests pass (169 passing, pre-existing failures unrelated to changes)
✅ Linting and type checking pass for modified files

- Add includeToolResultStatus config option supporting 'auto' | boolean - Implement auto-detection logic to check model ID patterns - Add MODELS_INCLUDE_STATUS constant with Claude model pattern - Update _formatContentBlock to conditionally include status field - Add _shouldIncludeToolResultStatus private method with debug logging - Add 8 comprehensive unit tests covering all configuration modes - Default behavior is 'auto' mode which detects based on model ID Resolves: #24

afarntrog · 2025-10-29T16:53:26Z

/strands address the comment #89 (comment)

Use cleaner single-return pattern with conditional property spreading instead of separate if/else return statements for better readability.

afarntrog · 2025-10-29T16:59:31Z

/strands address the comment #89 (comment)

Removed duplicate test case that was testing non-Claude model behavior in 'auto' mode. This is already covered by the undefined/default tests.

afarntrog · 2025-10-29T18:14:46Z

/strands address the following 2 comments:

Task 3.3: Implement Tool Result Status Auto-Detection for BedrockModelProvider Implementation #89 (comment) and
Task 3.3: Implement Tool Result Status Auto-Detection for BedrockModelProvider Implementation #89 (comment)
For the second comment you need to do research and be sure to have support backing up your research - such as a link to the documentation

@see

- Remove generic @see link from includeToolResultStatus config - Add specific AWS API reference to MODELS_INCLUDE_STATUS constant - Document that status field is only supported by Claude models per AWS docs Reference: https://docs.aws.amazon.com/bedrock/latest/APIReference/API_runtime_ToolResultBlock.html

afarntrog · 2025-10-29T19:18:14Z

/strands address the following two comments:

Task 3.3: Implement Tool Result Status Auto-Detection for BedrockModelProvider Implementation #89 (comment)
Task 3.3: Implement Tool Result Status Auto-Detection for BedrockModelProvider Implementation #89 (comment)

- Change console.log to console.debug for auto-detection logging - Remove redundant 'follows auto logic for Claude models' test in undefined section - Update debug logging tests to spy on console.debug instead of console.log - Test count reduced from 44 to 43 tests (all passing)

afarntrog · 2025-10-29T19:29:53Z

/strands address this comment #89 (comment)

Remove debug logging test cases as they are overkill. The console.debug calls still exist in the implementation for debugging purposes. Test count reduced from 43 to 41 tests (all passing).

This fixes strands-agents/harness-sdk#88

afarntrog had a problem deploying to auto-approve October 29, 2025 16:44 — with GitHub Actions Failure

afarntrog commented Oct 29, 2025

View reviewed changes

Comment thread src/models/bedrock.ts Outdated

github-actions Bot added the strands-running <strands-managed> Whether or not an agent is currently running label Oct 29, 2025

refactor: use conditional property spreading for tool result status

77aa238

Use cleaner single-return pattern with conditional property spreading instead of separate if/else return statements for better readability.