inspect-swe 0.2.4__py3-none-any.whl → 0.2.6__py3-none-any.whl

inspect_swe/__init__.py

@@ -1,4 +1,4 @@
-from ._claude_code.claude_code import ClaudeCodeOptions, claude_code
+from ._claude_code.claude_code import claude_code
 from ._tools.download import download_agent_binary
 from ._util.sandbox import SandboxPlatform
 
@@ -10,7 +10,6 @@ except ImportError:
 
 __all__ = [
     "claude_code",
-    "ClaudeCodeOptions",
     "download_agent_binary",
     "SandboxPlatform",
     "__version__",

inspect_swe/_claude_code/claude_code.py

@@ -4,41 +4,20 @@ from typing import Any, Literal, Sequence
 
 from inspect_ai.agent import (
     Agent,
+    AgentAttempts,
     AgentState,
     agent,
     agent_with,
     sandbox_agent_bridge,
 )
 from inspect_ai.model import ChatMessageSystem, ChatMessageUser
+from inspect_ai.scorer import score
 from inspect_ai.tool import MCPServerConfig
 from inspect_ai.util import sandbox as sandbox_env
-from pydantic import BaseModel, Field
 from pydantic_core import to_json
 
-from inspect_swe._claude_code.install.install import ensure_claude_code_installed
-
-# TODO: AgentAttempts
-# TODO: AgentContinue
-# TODO: generate config merging (they are passing max_tokens=32000, temperature=1)
-
-
-class ClaudeCodeOptions(BaseModel):
-    """Claude Code options."""
-
-    system_prompt: str | None = Field(default=None)
-    """Additional system prompt to append to default system prompt."""
-
-    mcp_servers: Sequence[MCPServerConfig] | None = Field(default=None)
-    """MCP servers to make available to the agent."""
-
-    model: str | None = Field(default=None)
-    """ Model name to use for Opus and Sonnet calls (defaults to main model for task)."""
-
-    small_model: str | None = Field(default=None)
-    """Model to use for Haiku calls (defaults to main model for task)."""
-
-    env: dict[str, str] | None = Field(default=None)
-    """Environment variables to set for claude code."""
+from .._util._async import is_callable_coroutine
+from .install.install import ensure_claude_code_installed
 
 
 @agent
@@ -48,7 +27,12 @@ def claude_code(
        Autonomous coding agent capable of writing, testing, debugging,
        and iterating on code across multiple languages.
     """),
-    options: ClaudeCodeOptions | None = None,
+    system_prompt: str | None = None,
+    mcp_servers: Sequence[MCPServerConfig] | None = None,
+    attempts: int | AgentAttempts = 1,
+    model: str | None = None,
+    small_model: str | None = None,
+    env: dict[str, str] | None = None,
     version: Literal["auto", "sandbox", "stable", "latest"] | str = "auto",
     user: str | None = None,
     sandbox: str | None = None,
@@ -59,10 +43,18 @@ def claude_code(
 
     The agent can either use a version of Claude Code installed in the sandbox, or can download a version and install it in the sandbox (see docs on `version` option below for details).
 
+    Use the `attempts` option to enable additional submissions if the initial
+    submission(s) are incorrect (by default, no additional attempts are permitted).
+
     Args:
         name: Agent name (used in multi-agent systems with `as_tool()` and `handoff()`)
         description: Agent description (used in multi-agent systems with `as_tool()` and `handoff()`)
-        options: Claude code options.
+        system_prompt: Additional system prompt to append to default system prompt.
+        mcp_servers: MCP servers to make available to the agent.
+        attempts: Configure agent to make multiple attempts.
+        model: Model name to use for Opus and Sonnet calls (defaults to main model for task).
+        small_model: Model to use for Haiku calls (defaults to main model for task).
+        env: Environment variables to set for claude code.
         version: Version of claude code to use. One of:
             - "auto": Use any available version of claude code in the sandbox, otherwise download the current stable version.
             - "sandbox": Use the version of claude code in the sandbox (raises `RuntimeError` if claude is not available in the sandbox)
@@ -72,16 +64,12 @@ def claude_code(
         user: User to execute claude code with.
         sandbox: Optional sandbox environment name.
     """
-    # provide default options if none specified
-    options = options or ClaudeCodeOptions()
-
     # resolve models
-    model = f"inspect/{options.model}" if options.model is not None else "inspect"
-    small_model = (
-        f"inspect/{options.small_model}"
-        if options.small_model is not None
-        else "inspect"
-    )
+    model = f"inspect/{model}" if model is not None else "inspect"
+    small_model = f"inspect/{small_model}" if small_model is not None else "inspect"
+
+    # resolve attempts
+    attempts = AgentAttempts(attempts) if isinstance(attempts, int) else attempts
 
     async def execute(state: AgentState) -> AgentState:
         async with sandbox_agent_bridge(state) as bridge:
@@ -95,9 +83,6 @@ def claude_code(
 
             # base options
             cmd = [
-                claude_binary,
-                "--session-id",
-                session_id,
                 "--print",  # run without interactions
                 "--dangerously-skip-permissions",
                 "--model",
@@ -108,50 +93,82 @@ def claude_code(
             system_messages = [
                 m.text for m in state.messages if isinstance(m, ChatMessageSystem)
             ]
-            if options.system_prompt is not None:
-                system_messages.append(options.system_prompt)
+            if system_prompt is not None:
+                system_messages.append(system_prompt)
             if system_messages:
                 cmd.extend(["--append-system-prompt", "\n\n".join(system_messages)])
 
             # mcp servers
-            if options.mcp_servers:
-                cmd.extend(mcp_server_args(options.mcp_servers))
+            if mcp_servers:
+                cmd.extend(mcp_server_args(mcp_servers))
 
             # user prompt
             prompt = "\n\n".join(
                 [m.text for m in state.messages if isinstance(m, ChatMessageUser)]
             )
-            cmd.append("--")
-            cmd.append(prompt)
 
             # resolve sandbox
             sbox = sandbox_env(sandbox)
 
             # execute the agent
-            result = await sbox.exec(
-                cmd=cmd,
-                env={
-                    "ANTHROPIC_BASE_URL": f"http://localhost:{bridge.port}",
-                    "ANTHROPIC_API_KEY": "sk-ant-api03-DOq5tyLPrk9M4hPE",
-                    "ANTHROPIC_MODEL": model,
-                    "ANTHROPIC_DEFAULT_OPUS_MODEL": model,
-                    "ANTHROPIC_DEFAULT_SONNET_MODEL": model,
-                    "CLAUDE_CODE_SUBAGENT_MODEL": model,
-                    "ANTHROPIC_DEFAULT_HAIKU_MODEL": small_model,
-                    "ANTHROPIC_SMALL_FAST_MODEL": small_model,
-                    "CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC": "1",
-                    "IS_SANDBOX": "1",
-                }
-                | (options.env or {}),
-                user=user,
-            )
-
-        if result.success:
-            return bridge.state
-        else:
-            raise RuntimeError(
-                f"Error executing claude code agent: {result.stdout}\n{result.stderr}"
-            )
+            agent_prompt = prompt
+            attempt_count = 0
+            while True:
+                # either starting a new session or resuming one
+                id_param = "--session-id" if attempt_count == 0 else "--resume"
+                agent_cmd = (
+                    [claude_binary, id_param, session_id] + cmd + ["--", agent_prompt]
+                )
+
+                # run agent
+                result = await sbox.exec(
+                    cmd=agent_cmd,
+                    env={
+                        "ANTHROPIC_BASE_URL": f"http://localhost:{bridge.port}",
+                        "ANTHROPIC_API_KEY": "sk-ant-api03-DOq5tyLPrk9M4hPE",
+                        "ANTHROPIC_MODEL": model,
+                        "ANTHROPIC_DEFAULT_OPUS_MODEL": model,
+                        "ANTHROPIC_DEFAULT_SONNET_MODEL": model,
+                        "CLAUDE_CODE_SUBAGENT_MODEL": model,
+                        "ANTHROPIC_DEFAULT_HAIKU_MODEL": small_model,
+                        "ANTHROPIC_SMALL_FAST_MODEL": small_model,
+                        "CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC": "1",
+                        "IS_SANDBOX": "1",
+                    }
+                    | (env or {}),
+                    user=user,
+                )
+
+                # raise for error
+                if not result.success:
+                    f"Error executing claude code agent: {result.stdout}\n{result.stderr}"
+
+                # exit if we are at max_attempts
+                attempt_count += 1
+                if attempt_count >= attempts.attempts:
+                    break
+
+                # score this attempt
+                answer_scores = await score(state)
+
+                # break if we score 'correct'
+                if attempts.score_value(answer_scores[0].value) == 1.0:
+                    break
+
+                # otherwise update prompt with incorrect message and continue
+                else:
+                    if callable(attempts.incorrect_message):
+                        if not is_callable_coroutine(attempts.incorrect_message):
+                            raise ValueError(
+                                "The incorrect_message function must be async."
+                            )
+                        agent_prompt = await attempts.incorrect_message(
+                            state, answer_scores
+                        )
+                    else:
+                        agent_prompt = attempts.incorrect_message
+
+        return bridge.state
 
     # return agent with specified name and descritpion
     return agent_with(execute, name=name, description=description)

inspect_swe/_util/_async.py

@@ -1,5 +1,6 @@
 import asyncio
-from typing import Coroutine, Literal, TypeVar, cast
+import inspect
+from typing import Any, Coroutine, Literal, TypeVar, cast
 
 import nest_asyncio  # type: ignore
 import sniffio
@@ -9,6 +10,14 @@ from .platform import running_in_notebook
 T = TypeVar("T")
 
 
+def is_callable_coroutine(func_or_cls: Any) -> bool:
+    if inspect.iscoroutinefunction(func_or_cls):
+        return True
+    elif callable(func_or_cls):
+        return inspect.iscoroutinefunction(func_or_cls.__call__)
+    return False
+
+
 def run_coroutine(coroutine: Coroutine[None, None, T]) -> T:
     if current_async_backend() == "trio":
         raise RuntimeError("run_coroutine cannot be used with trio")

inspect_swe/_version.py

@@ -28,7 +28,7 @@ version_tuple: VERSION_TUPLE
 commit_id: COMMIT_ID
 __commit_id__: COMMIT_ID
 
-__version__ = version = '0.2.4'
-__version_tuple__ = version_tuple = (0, 2, 4)
+__version__ = version = '0.2.6'
+__version_tuple__ = version_tuple = (0, 2, 6)
 
 __commit_id__ = commit_id = None

{inspect_swe-0.2.4.dist-info → inspect_swe-0.2.6.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: inspect_swe
-Version: 0.2.4
+Version: 0.2.6
 Summary: Software engineering agents for Inspect AI.
 Project-URL: Documentation, https://meridianlabs-ai.github.io/inspect_swe/
 Project-URL: Source Code, https://github.com/meridianlabs-ai/inspect_swe
@@ -10,7 +10,7 @@ License: MIT License
 License-File: LICENSE
 Requires-Python: >=3.10
 Requires-Dist: httpx
-Requires-Dist: inspect-ai>=0.3.125
+Requires-Dist: inspect-ai>=0.3.126
 Requires-Dist: nest-asyncio
 Requires-Dist: platformdirs
 Requires-Dist: pydantic>=2.11.4

{inspect_swe-0.2.4.dist-info → inspect_swe-0.2.6.dist-info}/RECORD

@@ -1,9 +1,9 @@
-inspect_swe/__init__.py,sha256=aqHkY79cer0TXcw2dy7RRLkGWoTCmCPFprPuCtXR_6k,386
+inspect_swe/__init__.py,sha256=yJ9tBcF2Wy11mVmLh1fTYXgYcsSHv30GAW-tVwE-r3s,342
 inspect_swe/_registry.py,sha256=jM37ysrY39Ufd67GRKbiwfSViOLlm-82lm_JEaWKshw,97
-inspect_swe/_version.py,sha256=NRw4Jle4n9v_DD2wtplRqflGCvX8OU5eAjycYY0vY3Y,704
+inspect_swe/_version.py,sha256=2Q6v117QPuRsVsIEaHT3nJJVx7xxa47FYOkmuhVbGAI,704
 inspect_swe/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 inspect_swe/_claude_code/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-inspect_swe/_claude_code/claude_code.py,sha256=-E_Ibu_xwIuDGGYgSFKXeBitfny0kINjwu-n-2rQdj8,7114
+inspect_swe/_claude_code/claude_code.py,sha256=bRYPsSYK3G4J5mPbgcpZSSKqWtUnhAy4_3acWDyaORs,8359
 inspect_swe/_claude_code/install/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 inspect_swe/_claude_code/install/cache.py,sha256=k08bCxGq-iYVpO16LNQhPjxTM9p2iecpqMjqYd2WBss,1708
 inspect_swe/_claude_code/install/download.py,sha256=s1y4CDHVbJenfsR7OUwwxr5QFp-rDi4XnIxumDEvmws,3217
@@ -11,7 +11,7 @@ inspect_swe/_claude_code/install/install.py,sha256=nbf1SZJzr4DBPfUmBH64zWcdI4AnK
 inspect_swe/_tools/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 inspect_swe/_tools/download.py,sha256=Jn_gcFR5Kw2vTYA1dWOFYRpqFtoFnKFv2Kv-4xT8tz4,1283
 inspect_swe/_util/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-inspect_swe/_util/_async.py,sha256=cL8_Smmj2Es41TefceGDYLyVaO7gZ56VJcA4oByrWfQ,1520
+inspect_swe/_util/_async.py,sha256=foxHmEaZusCbK8HOBbThZKCnwaPFerwLhQXh7jIafVU,1778
 inspect_swe/_util/_yaml.py,sha256=sRgf0UryF9Bd7pEEyfzL1qZBCgrpYe0l3l3U7bYeU44,505
 inspect_swe/_util/appdirs.py,sha256=V3o1ERdSYLjKP-m4O1T_Hvkx0UsP2HdfvsshLSQgP6E,562
 inspect_swe/_util/checksum.py,sha256=i-_GhtgCFd5eFj3PPJiGSCHDhZdPcIPNwiqddX93Sls,186
@@ -20,8 +20,8 @@ inspect_swe/_util/download.py,sha256=cCUau4ZBOKezpotJV5-v3JY_5CuYDZ-VcWlLf_EyNL0
 inspect_swe/_util/platform.py,sha256=wm4efIFfdyTeaV2oxOXVvYl1u22MHX3jQMERHJMgv7A,339
 inspect_swe/_util/sandbox.py,sha256=2wYmVz5EGUDBhqbN3NgLAOsyKeU-KRI161MZMJ54n4M,1769
 inspect_swe/_util/trace.py,sha256=mFHmBKn2F8iJP9PpTHaCseMHnTMz3ErRx6RCKV83rZk,139
-inspect_swe-0.2.4.dist-info/METADATA,sha256=wxryGFAjtZarLk41tmkyAGPVuiIh5OWFQq1QylHw0VM,1724
-inspect_swe-0.2.4.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
-inspect_swe-0.2.4.dist-info/entry_points.txt,sha256=OzpvUhd7M3T2Rog4MjwJAxIKeX5ljiR0mVYM9GefBKg,49
-inspect_swe-0.2.4.dist-info/licenses/LICENSE,sha256=Hi3UDcbD6yCKZ1mcgt7pprzSG0rDEnSrbrm3XinyiDA,1070
-inspect_swe-0.2.4.dist-info/RECORD,,
+inspect_swe-0.2.6.dist-info/METADATA,sha256=SJwbJADMfKoOBkiQ0_RST4iR1hjoHNM0nrn-kN_kU_o,1724
+inspect_swe-0.2.6.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
+inspect_swe-0.2.6.dist-info/entry_points.txt,sha256=OzpvUhd7M3T2Rog4MjwJAxIKeX5ljiR0mVYM9GefBKg,49
+inspect_swe-0.2.6.dist-info/licenses/LICENSE,sha256=Hi3UDcbD6yCKZ1mcgt7pprzSG0rDEnSrbrm3XinyiDA,1070
+inspect_swe-0.2.6.dist-info/RECORD,,