[4.D.5] Refactor backend_worker_agent, frontend_worker_agent, test_worker_agent to use LLMProvider

[frankbria]

Parent

Part of #542 — Multi-provider LLM support

Depends on

#546 — worker_agent.py pattern established first

What

Three additional worker agents hardcode AsyncAnthropic directly. Apply the same refactor pattern from #546.

Files to modify

• codeframe/agents/backend_worker_agent.py
• codeframe/agents/frontend_worker_agent.py
• codeframe/agents/test_worker_agent.py
• Corresponding test files

Pattern to follow

Same as #546: replace AsyncAnthropic(api_key=...) instantiation with self.llm_provider.async_complete(). Accept llm_provider: LLMProvider | None = None in constructor with get_provider() as default.

Notes

• backend_worker_agent.py has a fallback import (from anthropic import AsyncAnthropic) — remove entirely once abstraction is in place
• Each agent may have agent-specific system prompts; keep those, only replace the LLM call layer
• If any agent uses Anthropic-specific features (extended thinking, computer use) — note them explicitly and leave a # TODO: provider-specific comment rather than breaking functionality

Acceptance criteria

• All three agents have no import anthropic anywhere
• Each accepts llm_provider constructor param
• Existing tests pass
• No agent-specific functionality regressed

[traycerai]

Warning

Ticket Assist will be discontinued on May 1, 2026. After that date, the feature will no longer be available. We suggest installing the Traycer IDE Extension for continued AI-powered, spec-driven development. Thank you for choosing Traycer!

Plan

Observations

• FrontendWorkerAgent and TestWorkerAgent both import from anthropic import AsyncAnthropic at the top level and store self.client = AsyncAnthropic(api_key=...) in __init__. They call await self.client.messages.create(...) in their async generation methods.
• BackendWorkerAgent is a standalone class (not inheriting WorkerAgent) with a fallback from anthropic import AsyncAnthropic inside generate_code() for the non-SDK path.
• LLMProvider (in file:codeframe/adapters/llm/base.py) exposes a synchronous complete() method. The async bridge pattern used elsewhere in the codebase is asyncio.to_thread(provider.complete, ...).
• get_provider() in file:codeframe/adapters/llm/__init__.py accepts "anthropic" or "mock" and returns a configured LLMProvider.
• Test files (file:tests/agents/test_frontend_worker_agent.py, file:tests/agents/test_test_worker_agent.py) currently patch anthropic.types.Message and self.client directly.

Approach

Replace all direct AsyncAnthropic usage in the three agents with self.llm_provider: LLMProvider, injected via constructor with get_provider() as default. Async LLM calls become asyncio.to_thread(self.llm_provider.complete, ...). The BackendWorkerAgent's fallback Anthropic path is replaced with the same provider call. Test files are updated to inject a MockProvider instead of patching the Anthropic client.

---

Implementation Steps

1. Refactor FrontendWorkerAgent (file:codeframe/agents/frontend_worker_agent.py)

Imports:

• Remove from anthropic import AsyncAnthropic
• Add from codeframe.adapters.llm import LLMProvider, get_provider, Purpose

Constructor (__init__):

• Add parameter llm_provider: LLMProvider | None = None (after api_key)
• Remove self.api_key = api_key or os.getenv("ANTHROPIC_API_KEY")
• Remove self.client = AsyncAnthropic(api_key=self.api_key) if self.api_key else None
• Add self.llm_provider = llm_provider or get_provider()
• Keep all other constructor logic unchanged

_generate_react_component method:

• Replace the if not self.client: guard with if self.llm_provider is None: (though it will always be set now; keep the fallback path for safety)
• Replace the await self.client.messages.create(...) block with:
  • Build messages = [{"role": "user", "content": prompt}]
  • Call response = await asyncio.to_thread(self.llm_provider.complete, messages, Purpose.EXECUTION, None, 2000, 0.0, self.system_prompt)
  • Extract text via response.content (an LLMResponse object, not Anthropic SDK response)
  • Keep the markdown stripping logic unchanged

2. Refactor TestWorkerAgent (file:codeframe/agents/test_worker_agent.py)

Imports:

• Remove from anthropic import AsyncAnthropic
• Add from codeframe.adapters.llm import LLMProvider, get_provider, Purpose

Constructor (__init__):

• Add parameter llm_provider: LLMProvider | None = None
• Remove self.api_key = api_key or os.getenv("ANTHROPIC_API_KEY")
• Remove self.client = AsyncAnthropic(api_key=self.api_key) if self.api_key else None
• Add self.llm_provider = llm_provider or get_provider()

_generate_pytest_tests method:

• Replace if not self.client: guard with if self.llm_provider is None:
• Replace await self.client.messages.create(...) with asyncio.to_thread(self.llm_provider.complete, messages, Purpose.EXECUTION, None, 3000, 0.0, self.system_prompt)
• Extract text from response.content (the LLMResponse.content string field)

_correct_failing_tests method:

• Same replacement pattern as _generate_pytest_tests
• Replace if not self.client: guard
• Replace await self.client.messages.create(...) with asyncio.to_thread(self.llm_provider.complete, ...)

3. Refactor BackendWorkerAgent (file:codeframe/agents/backend_worker_agent.py)

Imports (top-level):

• No top-level from anthropic import AsyncAnthropic exists — the import is inside generate_code(). Remove it from there.
• Add from codeframe.adapters.llm import LLMProvider, get_provider, Purpose at the top

Constructor (__init__):

• Add parameter llm_provider: LLMProvider | None = None
• Add self.llm_provider = llm_provider or get_provider()
• Keep self.api_key only if it's still needed by SDKClientWrapper (it is — leave that untouched)

generate_code method — non-SDK fallback path (lines ~356–370):

• Remove the from anthropic import AsyncAnthropic local import
• Remove client = AsyncAnthropic(api_key=self.api_key)
• Remove await client.messages.create(...) call
• Replace with:
  • Build messages = [{"role": "user", "content": user_prompt}]
  • response = await asyncio.to_thread(self.llm_provider.complete, messages, Purpose.EXECUTION, None, 4096, 0.0, self._build_system_prompt())
  • response_text = response.content
  • result = json.loads(response_text)

4. Update test files

file:tests/agents/test_frontend_worker_agent.py:

• Remove from anthropic.types import Message, TextBlock import
• Add from codeframe.adapters.llm import MockProvider, LLMResponse
• In the frontend_agent fixture: pass llm_provider=MockProvider() instead of api_key="test-key"
• In tests that assert agent.client is not None (e.g., test_initialization_with_api_key): update to assert agent.llm_provider is not None
• In tests that mock self.client.messages.create: replace with mock_provider.add_text_response(...) using MockProvider

file:tests/agents/test_test_worker_agent.py:

• Remove from anthropic.types import Message, TextBlock
• Add from codeframe.adapters.llm import MockProvider, LLMResponse
• In the test_agent fixture: pass llm_provider=MockProvider() instead of api_key="test-key"
• Replace any patch("anthropic.AsyncAnthropic") or self.client patches with mock_provider.add_text_response(...)

file:tests/agents/test_backend_worker_agent.py:

• Add from codeframe.adapters.llm import MockProvider
• In tests that exercise the non-SDK generate_code path: pass llm_provider=MockProvider(default_response='{"files": [], "explanation": "mock"}') to BackendWorkerAgent
• Remove any patches targeting anthropic.AsyncAnthropic

5. Verify no remaining import anthropic in the three agent files

After the above changes, confirm with a search that none of backend_worker_agent.py, frontend_worker_agent.py, or test_worker_agent.py contain any import anthropic or from anthropic statement.

---

Data flow after refactor

sequenceDiagram
    participant Agent as FrontendWorkerAgent / TestWorkerAgent / BackendWorkerAgent
    participant Provider as LLMProvider (injected)
    participant Thread as asyncio.to_thread
    participant Anthropic as Anthropic API

    Agent->>Thread: asyncio.to_thread(llm_provider.complete, messages, purpose, ...)
    Thread->>Provider: complete(messages, purpose, ...)
    Provider->>Anthropic: client.messages.create(...)
    Anthropic-->>Provider: raw response
    Provider-->>Thread: LLMResponse(content=..., ...)
    Thread-->>Agent: LLMResponse
    Agent->>Agent: extract response.content (str)

Loading

Import In IDE

🤖 Prompt for AI Agents

## Observations

- `FrontendWorkerAgent` and `TestWorkerAgent` both import `from anthropic import AsyncAnthropic` at the top level and store `self.client = AsyncAnthropic(api_key=...)` in `__init__`. They call `await self.client.messages.create(...)` in their async generation methods.
- `BackendWorkerAgent` is a standalone class (not inheriting `WorkerAgent`) with a fallback `from anthropic import AsyncAnthropic` inside `generate_code()` for the non-SDK path.
- `LLMProvider` (in `file:codeframe/adapters/llm/base.py`) exposes a **synchronous** `complete()` method. The async bridge pattern used elsewhere in the codebase is `asyncio.to_thread(provider.complete, ...)`.
- `get_provider()` in `file:codeframe/adapters/llm/__init__.py` accepts `"anthropic"` or `"mock"` and returns a configured `LLMProvider`.
- Test files (`file:tests/agents/test_frontend_worker_agent.py`, `file:tests/agents/test_test_worker_agent.py`) currently patch `anthropic.types.Message` and `self.client` directly.

## Approach

Replace all direct `AsyncAnthropic` usage in the three agents with `self.llm_provider: LLMProvider`, injected via constructor with `get_provider()` as default. Async LLM calls become `asyncio.to_thread(self.llm_provider.complete, ...)`. The `BackendWorkerAgent`'s fallback Anthropic path is replaced with the same provider call. Test files are updated to inject a `MockProvider` instead of patching the Anthropic client.

---

## Implementation Steps

### 1. Refactor `FrontendWorkerAgent` (`file:codeframe/agents/frontend_worker_agent.py`)

**Imports:**
- Remove `from anthropic import AsyncAnthropic`
- Add `from codeframe.adapters.llm import LLMProvider, get_provider, Purpose`

**Constructor (`__init__`):**
- Add parameter `llm_provider: LLMProvider | None = None` (after `api_key`)
- Remove `self.api_key = api_key or os.getenv("ANTHROPIC_API_KEY")`
- Remove `self.client = AsyncAnthropic(api_key=self.api_key) if self.api_key else None`
- Add `self.llm_provider = llm_provider or get_provider()`
- Keep all other constructor logic unchanged

**`_generate_react_component` method:**
- Replace the `if not self.client:` guard with `if self.llm_provider is None:` (though it will always be set now; keep the fallback path for safety)
- Replace the `await self.client.messages.create(...)` block with:
  - Build `messages = [{"role": "user", "content": prompt}]`
  - Call `response = await asyncio.to_thread(self.llm_provider.complete, messages, Purpose.EXECUTION, None, 2000, 0.0, self.system_prompt)`
  - Extract text via `response.content` (an `LLMResponse` object, not Anthropic SDK response)
  - Keep the markdown stripping logic unchanged

### 2. Refactor `TestWorkerAgent` (`file:codeframe/agents/test_worker_agent.py`)

**Imports:**
- Remove `from anthropic import AsyncAnthropic`
- Add `from codeframe.adapters.llm import LLMProvider, get_provider, Purpose`

**Constructor (`__init__`):**
- Add parameter `llm_provider: LLMProvider | None = None`
- Remove `self.api_key = api_key or os.getenv("ANTHROPIC_API_KEY")`
- Remove `self.client = AsyncAnthropic(api_key=self.api_key) if self.api_key else None`
- Add `self.llm_provider = llm_provider or get_provider()`

**`_generate_pytest_tests` method:**
- Replace `if not self.client:` guard with `if self.llm_provider is None:`
- Replace `await self.client.messages.create(...)` with `asyncio.to_thread(self.llm_provider.complete, messages, Purpose.EXECUTION, None, 3000, 0.0, self.system_prompt)`
- Extract text from `response.content` (the `LLMResponse.content` string field)

**`_correct_failing_tests` method:**
- Same replacement pattern as `_generate_pytest_tests`
- Replace `if not self.client:` guard
- Replace `await self.client.messages.create(...)` with `asyncio.to_thread(self.llm_provider.complete, ...)`

### 3. Refactor `BackendWorkerAgent` (`file:codeframe/agents/backend_worker_agent.py`)

**Imports (top-level):**
- No top-level `from anthropic import AsyncAnthropic` exists — the import is inside `generate_code()`. Remove it from there.
- Add `from codeframe.adapters.llm import LLMProvider, get_provider, Purpose` at the top

**Constructor (`__init__`):**
- Add parameter `llm_provider: LLMProvider | None = None`
- Add `self.llm_provider = llm_provider or get_provider()`
- Keep `self.api_key` only if it's still needed by `SDKClientWrapper` (it is — leave that untouched)

**`generate_code` method — non-SDK fallback path (lines ~356–370):**
- Remove the `from anthropic import AsyncAnthropic` local import
- Remove `client = AsyncAnthropic(api_key=self.api_key)`
- Remove `await client.messages.create(...)` call
- Replace with:
  - Build `messages = [{"role": "user", "content": user_prompt}]`
  - `response = await asyncio.to_thread(self.llm_provider.complete, messages, Purpose.EXECUTION, None, 4096, 0.0, self._build_system_prompt())`
  - `response_text = response.content`
  - `result = json.loads(response_text)`

### 4. Update test files

**`file:tests/agents/test_frontend_worker_agent.py`:**
- Remove `from anthropic.types import Message, TextBlock` import
- Add `from codeframe.adapters.llm import MockProvider, LLMResponse`
- In the `frontend_agent` fixture: pass `llm_provider=MockProvider()` instead of `api_key="test-key"`
- In tests that assert `agent.client is not None` (e.g., `test_initialization_with_api_key`): update to assert `agent.llm_provider is not None`
- In tests that mock `self.client.messages.create`: replace with `mock_provider.add_text_response(...)` using `MockProvider`

**`file:tests/agents/test_test_worker_agent.py`:**
- Remove `from anthropic.types import Message, TextBlock`
- Add `from codeframe.adapters.llm import MockProvider, LLMResponse`
- In the `test_agent` fixture: pass `llm_provider=MockProvider()` instead of `api_key="test-key"`
- Replace any `patch("anthropic.AsyncAnthropic")` or `self.client` patches with `mock_provider.add_text_response(...)`

**`file:tests/agents/test_backend_worker_agent.py`:**
- Add `from codeframe.adapters.llm import MockProvider`
- In tests that exercise the non-SDK `generate_code` path: pass `llm_provider=MockProvider(default_response='{"files": [], "explanation": "mock"}')` to `BackendWorkerAgent`
- Remove any patches targeting `anthropic.AsyncAnthropic`

### 5. Verify no remaining `import anthropic` in the three agent files

After the above changes, confirm with a search that none of `backend_worker_agent.py`, `frontend_worker_agent.py`, or `test_worker_agent.py` contain any `import anthropic` or `from anthropic` statement.

---

## Data flow after refactor

```mermaid
sequenceDiagram
    participant Agent as FrontendWorkerAgent / TestWorkerAgent / BackendWorkerAgent
    participant Provider as LLMProvider (injected)
    participant Thread as asyncio.to_thread
    participant Anthropic as Anthropic API

    Agent->>Thread: asyncio.to_thread(llm_provider.complete, messages, purpose, ...)
    Thread->>Provider: complete(messages, purpose, ...)
    Provider->>Anthropic: client.messages.create(...)
    Anthropic-->>Provider: raw response
    Provider-->>Thread: LLMResponse(content=..., ...)
    Thread-->>Agent: LLMResponse
    Agent->>Agent: extract response.content (str)
```

---

Execution Information

Branch: main
Commit: f61db8d

---

💡 Tips

Supported Commands (Inside Comments)

• Use @traycerai generate to iterate on the previous version of the implementation plan.

Supported Commands (Inside Description)

• Add @traycerai ignore anywhere in the ticket description to prevent this ticket from being processed.
• Add @traycerai branch:<branch-name> anywhere in the ticket description to specify the target branch for the implementation plan.

Community

• Join our Discord Community to get help, request features, and share feedback.
• Follow us on X/Twitter for updates and announcements.

[coderabbitai]

🔗 Related PRs

#11 - feat: Convert worker agents to async/await pattern (cf-48) [merged]
#37 - feat: Multi-agent per project architecture (Phases 1-5) [merged]
#125 - Implement WorkerAgent execute_task method [merged]
#126 - Add token tracking to WorkerAgent execute_task method [merged]
#246 - feat(agents): trigger multi-agent execution after task approval [merged]

---

📝 Issue Planner

_{Check the box below or use the @coderabbitai plan command to generate an implementation plan and prompts that you can use with your favorite coding assistant.}

• Create Plan

---

🧪 Issue enrichment is currently in open beta.

You can configure auto-planning by selecting labels in the issue_enrichment configuration.

To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

issue_enrichment:
  auto_enrich:
    enabled: false

💬 Have feedback or questions? Drop into our discord!