inspect-swe 0.1.5__py3-none-any.whl → 0.2.0__py3-none-any.whl

inspect_swe/_claude_code/claude_code.py

@@ -1,9 +1,56 @@
-from inspect_ai.agent import Agent, AgentState, agent
+from inspect_ai.agent import (
+    Agent,
+    AgentState,
+    agent,
+    sandbox_agent_bridge,
+)
+from inspect_ai.model import ChatMessageSystem, ChatMessageUser
+from inspect_ai.util import sandbox
 
 
 @agent
 def claude_code() -> Agent:
     async def execute(state: AgentState) -> AgentState:
-        raise RuntimeError("claude_code() not yet implemented.")
+        async with sandbox_agent_bridge(state) as bridge:
+            # base options
+            cmd = [
+                "claude",
+                "--print",  # run without interactions
+                "--dangerously-skip-permissions",
+                "--model",  # use current inspect model
+                "inspect",
+            ]
+
+            # system message
+            system_message = "\n\n".join(
+                [m.text for m in state.messages if isinstance(m, ChatMessageSystem)]
+            )
+            if system_message:
+                cmd.extend(["--append-system-prompt", system_message])
+
+            # user prompt
+            prompt = "\n\n".join(
+                [m.text for m in state.messages if isinstance(m, ChatMessageUser)]
+            )
+            cmd.append(prompt)
+
+            # execute the agent
+            result = await sandbox().exec(
+                cmd=cmd,
+                env={
+                    "ANTHROPIC_BASE_URL": f"http://localhost:{bridge.port}",
+                    "ANTHROPIC_API_KEY": "sk-ant-api03-DOq5tyLPrk9M4hPE",
+                    "ANTHROPIC_SMALL_FAST_MODEL": "inspect",
+                    "CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC": "1",
+                    "IS_SANDBOX": "1",
+                },
+            )
+
+        if result.success:
+            return bridge.state
+        else:
+            raise RuntimeError(
+                f"Error executing claude code agent: {result.stdout}\n{result.stderr}"
+            )
 
     return execute

inspect_swe/_version.py

@@ -28,7 +28,7 @@ version_tuple: VERSION_TUPLE
 commit_id: COMMIT_ID
 __commit_id__: COMMIT_ID
 
-__version__ = version = '0.1.5'
-__version_tuple__ = version_tuple = (0, 1, 5)
+__version__ = version = '0.2.0'
+__version_tuple__ = version_tuple = (0, 2, 0)
 
 __commit_id__ = commit_id = None

{inspect_swe-0.1.5.dist-info → inspect_swe-0.2.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: inspect_swe
-Version: 0.1.5
+Version: 0.2.0
 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
@@ -9,11 +9,14 @@ Author: Meridian Labs
 License: MIT License
 License-File: LICENSE
 Requires-Python: >=3.10
-Requires-Dist: inspect-ai
+Requires-Dist: inspect-ai>=0.3.125
 Requires-Dist: typing-extensions>=4.9.0
 Provides-Extra: dev
+Requires-Dist: anthropic; extra == 'dev'
 Requires-Dist: mypy; extra == 'dev'
+Requires-Dist: openai; extra == 'dev'
 Requires-Dist: pytest; extra == 'dev'
+Requires-Dist: pytest-dotenv; extra == 'dev'
 Requires-Dist: ruff; extra == 'dev'
 Provides-Extra: doc
 Requires-Dist: quarto-cli==1.7.31; extra == 'doc'
@@ -21,6 +24,7 @@ Description-Content-Type: text/markdown
 
 Welcome to Inspect SWE, a suite of software engineering agents for [Inspect AI](https://inspect.aisi.org.uk/).
 
+For details on using Inspect SWE, please visit <https://meridianlabs-ai.github.io/inspect_swe/>.
 
 ## Installation
 

inspect_swe-0.2.0.dist-info/RECORD

@@ -0,0 +1,10 @@
+inspect_swe/__init__.py,sha256=6F52dddUoPvJA7RugtyDaswUqliDGgeaTK_OXWplvI0,185
+inspect_swe/_version.py,sha256=Dg8AmJomLVpjKL6prJylOONZAPRtB86LOce7dorQS_A,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=iE_-6Wv0m7hO1Tj-d21K8iZHgBIcTcfSqLHPVS1whMM,1788
+inspect_swe/_claude_code/install_claude.py,sha256=g5nHIY-JVKDQFgm0IIhpCsCX5B6MadYj8-CtKpKU4YE,11796
+inspect_swe-0.2.0.dist-info/METADATA,sha256=BspOyEN0yJbUDd_MK8Zn4A86osKL1tOGHq1axRM9iH8,1526
+inspect_swe-0.2.0.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
+inspect_swe-0.2.0.dist-info/licenses/LICENSE,sha256=Hi3UDcbD6yCKZ1mcgt7pprzSG0rDEnSrbrm3XinyiDA,1070
+inspect_swe-0.2.0.dist-info/RECORD,,

inspect_swe-0.1.5.dist-info/RECORD

@@ -1,10 +0,0 @@
-inspect_swe/__init__.py,sha256=6F52dddUoPvJA7RugtyDaswUqliDGgeaTK_OXWplvI0,185
-inspect_swe/_version.py,sha256=rdxBMYpwzYxiWk08QbPLHSAxHoDfeKWwyaJIAM0lSic,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=-WVHfA-kRDyEwCMeUlDR6A5fWSBcJGGV5o_np0Y6MnY,232
-inspect_swe/_claude_code/install_claude.py,sha256=g5nHIY-JVKDQFgm0IIhpCsCX5B6MadYj8-CtKpKU4YE,11796
-inspect_swe-0.1.5.dist-info/METADATA,sha256=5JqMYTi_E31VZmxOHqHICbc5Nhwq9tkhEZ8l2NtYQ14,1296
-inspect_swe-0.1.5.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
-inspect_swe-0.1.5.dist-info/licenses/LICENSE,sha256=Hi3UDcbD6yCKZ1mcgt7pprzSG0rDEnSrbrm3XinyiDA,1070
-inspect_swe-0.1.5.dist-info/RECORD,,