[devin-ai-integration]

Summary

Removes copy-pasted boilerplate that was duplicated across the chat-based agents in src/agentlab/agents/. Behavior is unchanged — this is a pure refactor to centralize reusable logic, validated against the existing agent test suite.

Two sources of duplication were consolidated:

1. The get_action epilogue — GenericAgent and VisualAgent had byte-for-byte identical code to build retry stats + AgentInfo, and to produce the parse-failure fallback dict. Extracted into agents/agent_utils.py:

def busted_retry_ans_dict(max_retry) -> dict:
    return dict(action=None, n_retry=max_retry + 1, busted_retry=1)

def make_agent_info(chat_llm, ans_dict, chat_messages, chat_model_args) -> AgentInfo:
    stats = chat_llm.get_stats()
    stats["n_retry"] = ans_dict["n_retry"]
    stats["busted_retry"] = ans_dict["busted_retry"]
    return AgentInfo(think=ans_dict.get("think"), chat_messages=chat_messages,
                     stats=stats, extra_info={"chat_model_args": asdict(chat_model_args)})

So each agent's get_action collapses from ~15 lines to:

except ParseError:
    ans_dict = busted_retry_ans_dict(self.max_retry)
...
agent_info = make_agent_info(self.chat_llm, ans_dict, chat_messages, self.chat_model_args)

2. prepare() / close() — these 4×-identical methods (return self.chat_model_args.{prepare,close}_server()) moved into a new ChatModelAgentArgs(AgentArgs) base in agents/agent_args.py. GenericAgentArgs, VisualAgentArgs, MostBasicAgentArgs, and VisualWebArenaAgentArgs now inherit it and drop their own copies.

Scope notes

Deliberately left out to avoid behavior changes:

• ToolUseAgentArgs uses model_args (not chat_model_args), so it keeps its own prepare/close.
• TapeAgentArgs / DebugAgentArgs have no LLM server and rely on the no-op prepare/close from the base — the new base is opt-in, so they are unaffected. The shared methods were intentionally not added to AgentArgs itself (which would have broken these two).
• ChatModelAgentArgs declares no new dataclass fields, so subclass field ordering / positional construction is preserved.

Testing

• New tests/agents/test_agent_utils.py covers make_agent_info, busted_retry_ans_dict, and ChatModelAgentArgs.prepare/close delegation.
• Full tests/agents/ suite passes (24 passed, 1 skipped) against the MiniWoB server, confirming GenericAgent end-to-end behavior is preserved.
• black src/ (pinned 25.1.0) and darglint clean on changed files.

Link to Devin session: https://app.devin.ai/sessions/5bea0615dcb648abb0e44224fac5c507
Requested by: @dr-phil

---

Devin Review

Status	Commit
⚪ Not started	—

Run Devin Review

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no bugs or issues to report.