inspect-swe 0.1.5__tar.gz → 0.2.0__tar.gz

{inspect_swe-0.1.5 → inspect_swe-0.2.0}/.gitignore

@@ -208,3 +208,9 @@ __marimo__/
 
 # Dynamic version
 src/inspect_swe/_version.py
+
+# Inspect logs
+logs/
+
+# MacOS
+.DS_Store

{inspect_swe-0.1.5 → inspect_swe-0.2.0}/PKG-INFO

@@ -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-0.1.5 → inspect_swe-0.2.0}/README.md

@@ -1,5 +1,6 @@
 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-0.1.5 → inspect_swe-0.2.0}/pyproject.toml

@@ -10,7 +10,7 @@ authors = [{ name = "Meridian Labs" }]
 readme = "README.md"
 requires-python = ">=3.10"
 license = { text = "MIT License" }
-dependencies = ["inspect_ai", "typing_extensions>=4.9.0"]
+dependencies = ["inspect_ai>=0.3.125", "typing_extensions>=4.9.0"]
 
 [project.urls]
 Documentation = "https://meridianlabs-ai.github.io/inspect_swe/"
@@ -19,7 +19,7 @@ Documentation = "https://meridianlabs-ai.github.io/inspect_swe/"
 
 
 [project.optional-dependencies]
-dev = ["ruff", "mypy", "pytest"]
+dev = ["ruff", "mypy", "pytest", "anthropic", "openai", "pytest-dotenv"]
 doc = ["quarto-cli==1.7.31"]
 
 [tool.hatch.build]

inspect_swe-0.2.0/src/inspect_swe/_claude_code/claude_code.py

@@ -0,0 +1,56 @@
+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:
+        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-0.1.5 → inspect_swe-0.2.0}/src/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/src/inspect_swe/_claude_code/claude_code.py

@@ -1,9 +0,0 @@
-from inspect_ai.agent import Agent, AgentState, agent
-
-
-@agent
-def claude_code() -> Agent:
-    async def execute(state: AgentState) -> AgentState:
-        raise RuntimeError("claude_code() not yet implemented.")
-
-    return execute