Add Claude Code and Codex harnesses

assets/lab/environments/AGENTS.md

@@ -712,8 +712,8 @@ by the framework; do not accept `None` or write `config = config or MyEnvConfig(
 knobs on `TasksetConfig` or `HarnessConfig`, not on `EnvConfig` itself.
 Environment packages should not subclass `Env`.
 Reusable taskset environments can type `harness` as `vf.HarnessConfig`; TOML can
-then select a registered harness with names like `type = "terminus2"` or
-`type = "pi"` inside the harness table.
+then select a registered harness with names like `type = "codex"`,
+`type = "claude-code"`, or `type = "pi"` inside the harness table.
 
 The taskset-only shape is:
 

docs/byo-harness.md

@@ -429,9 +429,9 @@ config surface; do not subclass `Env` just to bypass inference.
 
 Packaged CLI harnesses should use the same boundary. These implementations live
 under `verifiers.v1.packages` while the v1 surface stabilizes, and are
-re-exported through `verifiers.v1`. `OpenCode`, `Pi`, `MiniSWEAgent`,
-`Terminus2`, and `RLM` are bundled `Harness` leaf wrappers for common
-command-line agents:
+re-exported through `verifiers.v1`. `OpenCode`, `ClaudeCode`, `Codex`, `Pi`,
+`MiniSWEAgent`, `Terminus2`, and `RLM` are bundled `Harness` leaf wrappers for
+common command-line agents:
 
 ```python
 class HarborEnvConfig(vf.EnvConfig):
@@ -543,8 +543,8 @@ and harness config types for the loader.
 
 Reusable taskset environments can keep `harness` typed as `vf.HarnessConfig`.
 Then TOML may select a registered harness config with `type`, for example
-`type = "terminus2"` or `type = "pi"`, and pass that config's ordinary fields
-beside it. Use `harness = "pi"` when the selected harness needs no field
+`type = "codex"` or `type = "claude-code"`, and pass that config's ordinary
+fields beside it. Use `harness = "pi"` when the selected harness needs no field
 overrides.
 
 ```python

docs/environments.md

@@ -705,8 +705,8 @@ by the framework; do not accept `None` or write `config = config or MyEnvConfig(
 knobs on `TasksetConfig` or `HarnessConfig`, not on `EnvConfig` itself.
 Environment packages should not subclass `Env`.
 Reusable taskset environments can type `harness` as `vf.HarnessConfig`; TOML can
-then select a registered harness with names like `type = "terminus2"` or
-`type = "pi"` inside the harness table.
+then select a registered harness with names like `type = "codex"`,
+`type = "claude-code"`, or `type = "pi"` inside the harness table.
 
 The taskset-only shape is:
 

docs/evaluation.md

@@ -383,15 +383,15 @@ optional:
 | `endpoint_id` | string | Endpoint registry id (requires TOML `endpoints_path`) |
 
 Use `harness.type` to choose a registered v1 harness config for reusable taskset
-environments. Bundled names include `opencode`, `mini-swe-agent`, `pi`, `rlm`,
-and `terminus2`:
+environments. Bundled names include `opencode`, `claude-code`, `codex`,
+`mini-swe-agent`, `pi`, `rlm`, and `terminus2`:
 
 ```toml
 [[eval]]
 id = "openthoughts-tblite"
 
 [eval.harness]
-type = "terminus2"
+type = "codex"
 max_turns = 4
 ```
 

docs/reference.md

@@ -1008,9 +1008,9 @@ Nested config defaults should be explicit config objects, e.g.
 `taskset: MyTasksetConfig = MyTasksetConfig()`.
 
 When `harness` is typed as `HarnessConfig`, TOML can select a registered
-harness config with `type`, such as `type = "terminus2"` or `type = "pi"`, then
-pass the normal fields for that config in the same table. The shorthand form
-`harness = "pi"` is also accepted when no fields need to be overridden.
+harness config with `type`, such as `type = "codex"` or `type = "claude-code"`,
+then pass the normal fields for that config in the same table. The shorthand
+form `harness = "pi"` is also accepted when no fields need to be overridden.
 
 `Config` subclasses are strict Pydantic config models. Validate raw mappings
 with `MyConfig.model_validate(...)` or use the typed object directly.

environments/AGENTS.md

@@ -711,8 +711,8 @@ by the framework; do not accept `None` or write `config = config or MyEnvConfig(
 knobs on `TasksetConfig` or `HarnessConfig`, not on `EnvConfig` itself.
 Environment packages should not subclass `Env`.
 Reusable taskset environments can type `harness` as `vf.HarnessConfig`; TOML can
-then select a registered harness with names like `type = "terminus2"` or
-`type = "pi"` inside the harness table.
+then select a registered harness with names like `type = "codex"`,
+`type = "claude-code"`, or `type = "pi"` inside the harness table.
 
 The taskset-only shape is:
 

tests/test_v1_config_extension.py

@@ -1565,32 +1565,36 @@ class LocalEnvConfig(EnvConfig):
 
 
 @pytest.mark.parametrize(
-    ("alias", "config_cls", "harness_cls"),
+    ("alias", "config_cls", "harness_cls", "config_fields"),
     [
-        ("opencode", vf.OpenCodeConfig, vf.OpenCode),
-        ("open-code", vf.OpenCodeConfig, vf.OpenCode),
-        ("mini-swe-agent", vf.MiniSWEAgentConfig, vf.MiniSWEAgent),
-        ("pi", vf.PiConfig, vf.Pi),
-        ("rlm", vf.RLMConfig, vf.RLM),
-        ("terminus2", vf.Terminus2Config, vf.Terminus2),
-        ("terminus-2", vf.Terminus2Config, vf.Terminus2),
+        ("opencode", vf.OpenCodeConfig, vf.OpenCode, {"max_turns": 4}),
+        ("open-code", vf.OpenCodeConfig, vf.OpenCode, {"max_turns": 4}),
+        ("claude", vf.ClaudeCodeConfig, vf.ClaudeCode, {"max_turns": 4}),
+        ("claude-code", vf.ClaudeCodeConfig, vf.ClaudeCode, {"max_turns": 4}),
+        ("codex", vf.CodexConfig, vf.Codex, {}),
+        ("mini-swe-agent", vf.MiniSWEAgentConfig, vf.MiniSWEAgent, {"max_turns": 4}),
+        ("pi", vf.PiConfig, vf.Pi, {"max_turns": 4}),
+        ("rlm", vf.RLMConfig, vf.RLM, {"max_turns": 4}),
+        ("terminus2", vf.Terminus2Config, vf.Terminus2, {"max_turns": 4}),
+        ("terminus-2", vf.Terminus2Config, vf.Terminus2, {"max_turns": 4}),
     ],
 )
 def test_env_config_harness_type_selects_packaged_harness_config(
-    alias, config_cls, harness_cls
+    alias, config_cls, harness_cls, config_fields
 ) -> None:
     class GenericEnvConfig(EnvConfig):
         taskset: TasksetConfig = TasksetConfig(source=[])
         harness: HarnessConfig = vf.OpenCodeConfig()
 
     config = coerce_config(
         GenericEnvConfig,
-        {"harness": {"type": alias, "max_turns": 4}},
+        {"harness": {"type": alias, **config_fields}},
     )
     env = Env(config=config)
 
     assert isinstance(config.harness, config_cls)
-    assert config.harness.max_turns == 4
+    if "max_turns" in config_fields:
+        assert config.harness.max_turns == 4
     assert isinstance(env.harness, harness_cls)
 
 

tests/test_v1_harbor_cli.py

@@ -12,6 +12,8 @@
 
 import verifiers as root_vf
 import verifiers.v1 as vf
+from verifiers.v1.packages.harnesses.claude_code import claude_code_mcp_json
+from verifiers.v1.packages.harnesses.codex import codex_mcp_toml
 from verifiers.v1.packages.harnesses.configs import (
     TERMINUS_2_DEFAULT_API_BASE_URL,
     TERMINUS_2_DEFAULT_HARBOR_PACKAGE,
@@ -212,9 +214,19 @@ async def test_harbor_reward_uses_background_job_for_tests(
 
 
 def test_packaged_harbor_and_opencode_imports_are_reexported() -> None:
-    from verifiers.v1.packages.harnesses import OpenCode, OpenCodeConfig, Pi
+    from verifiers.v1.packages.harnesses import (
+        ClaudeCode,
+        Codex,
+        OpenCode,
+        OpenCodeConfig,
+        Pi,
+    )
     from verifiers.v1.packages.tasksets import HarborTaskset
 
+    assert vf.ClaudeCode is ClaudeCode
+    assert root_vf.ClaudeCode is ClaudeCode
+    assert vf.Codex is Codex
+    assert root_vf.Codex is Codex
     assert vf.OpenCode is OpenCode
     assert vf.OpenCodeConfig is OpenCodeConfig
     assert vf.Pi is Pi
@@ -255,6 +267,8 @@ def test_opencode_config_owns_opencode_harness_fields() -> None:
     ("harness_cls", "config_cls"),
     [
         (vf.OpenCode, vf.OpenCodeConfig),
+        (vf.ClaudeCode, vf.ClaudeCodeConfig),
+        (vf.Codex, vf.CodexConfig),
         (vf.MiniSWEAgent, vf.MiniSWEAgentConfig),
         (vf.Pi, vf.PiConfig),
         (vf.RLM, vf.RLMConfig),
@@ -312,6 +326,73 @@ def test_pi_harness_writes_intercepted_model_and_mcp_config() -> None:
     assert mcp["mcpServers"]["verifiers-tools"]["command"] == "python3"
 
 
+def test_claude_code_harness_builds_sandbox_program() -> None:
+    harness = vf.ClaudeCode(
+        config=vf.ClaudeCodeConfig(
+            system_prompt="extra system prompt",
+            agent_workdir="/workspace",
+            max_turns=7,
+        )
+    )
+    program = cast(dict[str, object], harness.program)
+    command = cast(list[object], program["command"])
+    setup = cast(str, program["setup"])
+    files = cast(dict[str, object], program["files"])
+    env = cast(dict[str, object], program["env"])
+    mcp = json.loads(claude_code_mcp_json())
+
+    assert "npm install -g @anthropic-ai/claude-code" in setup
+    assert "/claude-code/instruction.txt" in files
+    assert "/claude-code/system.txt" in files
+    assert program["channels"] == "mcp"
+    assert env["ANTHROPIC_MODEL"] == "runtime.model"
+    assert env["CLAUDE_CODE_ENABLE_GATEWAY_MODEL_DISCOVERY"] == "1"
+    assert "cat /claude-code/instruction.txt | claude -p" in cast(str, command[2])
+    assert '"$(cat /claude-code/instruction.txt)"' not in cast(str, command[2])
+    assert "--max-turns 7" in cast(str, command[2])
+    assert "--permission-mode bypassPermissions" in cast(str, command[2])
+    assert "--mcp-config /tmp/claude-code-mcp.json" in cast(str, command[2])
+    assert mcp["mcpServers"]["verifiers-tools"]["command"] == "python3"
+
+
+def test_codex_harness_builds_sandbox_program() -> None:
+    harness = vf.Codex(
+        config=vf.CodexConfig(
+            system_prompt="extra system prompt",
+            agent_workdir="/workspace",
+            codex_sandbox="workspace-write",
+            model_reasoning_effort="high",
+        )
+    )
+    program = cast(dict[str, object], harness.program)
+    command = cast(list[object], program["command"])
+    setup = cast(str, program["setup"])
+    files = cast(dict[str, object], program["files"])
+    env = cast(dict[str, object], program["env"])
+    mcp_toml = codex_mcp_toml()
+
+    assert "npm install -g @openai/codex" in setup
+    assert "/codex/instruction.txt" in files
+    assert "/codex/system.txt" in files
+    assert program["channels"] == "mcp"
+    assert env["OPENAI_MODEL"] == "runtime.model"
+    assert callable(env["CODEX_API_KEY"])
+    assert 'model_provider = "verifiers"' in cast(str, command[2])
+    assert 'approval_policy = "never"' in cast(str, command[2])
+    assert 'sandbox_mode = "workspace-write"' in cast(str, command[2])
+    assert 'model_reasoning_effort = "high"' in cast(str, command[2])
+    assert "--sandbox workspace-write" in cast(str, command[2])
+    assert "--output-last-message /logs/agent/codex.txt.final" in cast(str, command[2])
+    assert "- < /logs/agent/codex.txt.prompt" in cast(str, command[2])
+    assert '"$(cat /logs/agent/codex.txt.prompt)"' not in cast(str, command[2])
+    assert 'command = "python3"' in mcp_toml
+
+
+def test_codex_config_rejects_max_turns() -> None:
+    with pytest.raises(ValueError, match="CodexConfig.max_turns is not supported"):
+        vf.CodexConfig(max_turns=7)
+
+
 def test_terminus_2_harness_builds_sandbox_program() -> None:
     harness = vf.Terminus2(
         config=vf.Terminus2Config(

verifiers/__init__.py

@@ -88,6 +88,10 @@
     "UserConfig",
     "HarborTaskset",
     "HarborTasksetConfig",
+    "ClaudeCode",
+    "ClaudeCodeConfig",
+    "Codex",
+    "CodexConfig",
     "MiniSWEAgent",
     "MiniSWEAgentConfig",
     "OpenCode",
@@ -219,6 +223,10 @@
     "UserConfig": "verifiers.v1:UserConfig",
     "HarborTaskset": "verifiers.v1:HarborTaskset",
     "HarborTasksetConfig": "verifiers.v1:HarborTasksetConfig",
+    "ClaudeCode": "verifiers.v1:ClaudeCode",
+    "ClaudeCodeConfig": "verifiers.v1:ClaudeCodeConfig",
+    "Codex": "verifiers.v1:Codex",
+    "CodexConfig": "verifiers.v1:CodexConfig",
     "MiniSWEAgent": "verifiers.v1:MiniSWEAgent",
     "MiniSWEAgentConfig": "verifiers.v1:MiniSWEAgentConfig",
     "OpenCode": "verifiers.v1:OpenCode",
@@ -315,6 +323,10 @@ def __getattr__(name: str):
         HarborTasksetConfig,
         MCPTool,
         MCPToolConfig,
+        ClaudeCode,
+        ClaudeCodeConfig,
+        Codex,
+        CodexConfig,
         MiniSWEAgent,
         MiniSWEAgentConfig,
         MutableConfigMap,

verifiers/v1/ENVIRONMENT_BEST_PRACTICES.md

@@ -42,8 +42,8 @@ your loader runs. The type annotation is not cosmetic.
    your loader can convert or forward them.
 3. If your environment has custom harness fields, the same rule applies to the
    `harness` annotation unless TOML selects a registered harness config with
-   `[env.harness] type = "terminus2"`, `[env.harness] type = "pi"`, or another
-   owner/config alias.
+   `[env.harness] type = "codex"`, `[env.harness] type = "claude-code"`, or
+   another owner/config alias.
 4. The config object that reaches `load_environment` is already validated and
    typed. Do not reconstruct child config objects just to recover their type.
 5. `vf.Env(taskset=MyTaskset(config=config.taskset), harness=MyHarness(config=config.harness))`

verifiers/v1/README.md

@@ -574,9 +574,9 @@ signature.
 
 Reusable CLI programs should be packaged as `Harness` subclasses. Package
 implementations live under `verifiers.v1.packages` while the v1 API stabilizes,
-and are re-exported from `verifiers.v1` for normal use. `OpenCode`, `Pi`,
-`MiniSWEAgent`, `Terminus2`, and `RLM` are bundled `Harness` leaf wrappers for
-common coding-agent CLIs.
+and are re-exported from `verifiers.v1` for normal use. `OpenCode`,
+`ClaudeCode`, `Codex`, `Pi`, `MiniSWEAgent`, `Terminus2`, and `RLM` are
+bundled `Harness` leaf wrappers for common coding-agent CLIs.
 
 ```python
 import verifiers as vf
@@ -603,6 +603,8 @@ endpoint and, when tools are enabled, installs the Pi MCP adapter and writes a
 project `.mcp.json`. Neither side needs to know the other's private fields.
 `MiniSWEAgent` owns mini-swe-agent installation, config layering, endpoint env,
 and log/trajectory artifacts.
+`ClaudeCode` and `Codex` package the Claude Code and Codex CLI non-interactive
+modes with endpoint, MCP proxy, and log artifact wiring.
 `Terminus2` owns Harbor Terminus agent installation, endpoint env, and log
 artifacts.
 `RLM` follows the same boundary for recursive LLM runs: `HarborTaskset` owns
@@ -1243,8 +1245,8 @@ and harness config types for the loader.
 
 Reusable taskset environments can keep `harness` typed as `vf.HarnessConfig`.
 Then TOML may select a registered harness config with `type`, for example
-`type = "terminus2"` or `type = "pi"`, and pass that config's ordinary fields
-beside it. Use `harness = "pi"` when the selected harness needs no field
+`type = "codex"` or `type = "claude-code"`, and pass that config's ordinary
+fields beside it. Use `harness = "pi"` when the selected harness needs no field
 overrides.
 
 ```python

verifiers/v1/__init__.py

@@ -37,6 +37,10 @@
 from .env import Env
 from .harness import Harness
 from .packages.harnesses import (
+    ClaudeCode,
+    ClaudeCodeConfig,
+    Codex,
+    CodexConfig,
     MiniSWEAgent,
     MiniSWEAgentConfig,
     OpenCode,
@@ -80,6 +84,10 @@
 __all__ = [
     "ConfigData",
     "CallableConfig",
+    "ClaudeCode",
+    "ClaudeCodeConfig",
+    "Codex",
+    "CodexConfig",
     "Config",
     "ConfigMap",
     "Env",

verifiers/v1/packages/harnesses/__init__.py

@@ -1,17 +1,25 @@
 from .configs import (
+    ClaudeCodeConfig,
+    CodexConfig,
     MiniSWEAgentConfig,
     OpenCodeConfig,
     PiConfig,
     RLMConfig,
     Terminus2Config,
 )
+from .claude_code import ClaudeCode
+from .codex import Codex
 from .mini_swe_agent import MiniSWEAgent
 from .opencode import OpenCode
 from .pi import Pi
 from .rlm import RLM
 from .terminus_2 import Terminus2
 
 __all__ = [
+    "ClaudeCode",
+    "ClaudeCodeConfig",
+    "Codex",
+    "CodexConfig",
     "MiniSWEAgent",
     "MiniSWEAgentConfig",
     "OpenCode",