diff --git a/packages/bcbench-core/src/bcbench_core/agent/harness.py b/packages/bcbench-core/src/bcbench_core/agent/harness.py
new file mode 100644
index 000000000..7f1b28428
--- /dev/null
+++ b/packages/bcbench-core/src/bcbench_core/agent/harness.py
@@ -0,0 +1,58 @@
+"""Harness-neutral contract between evaluation pipelines and the agents they run."""
+
+from dataclasses import dataclass
+from pathlib import Path
+from typing import Protocol
+
+from pydantic import BaseModel, ConfigDict
+
+from bcbench_core.agent.experiment import ExperimentConfiguration
+from bcbench_core.agent.metrics import AgentMetrics
+
+__all__ = ["AgentIdentity", "AgentRequest", "AgentResult", "Harness"]
+
+
+class AgentIdentity[N: str = str](BaseModel):
+    """What is being evaluated, resolved before any task runs.
+
+    ``N`` lets applications type the harness name, e.g. with their own ``StrEnum`` of harnesses.
+    """
+
+    model_config = ConfigDict(frozen=True, extra="forbid")
+
+    harness: N
+    # CLI version, engine commit, or hosted agent version
+    version: str
+    # Configured model; None when the harness pins or delegates model selection
+    model: str | None = None
+
+
+@dataclass(frozen=True)
+class AgentRequest[T]:
+    task: T
+    # The agent's workspace: a prepared repository, or an empty directory for answer-only tasks
+    work_dir: Path
+    # Receives logs and raw captures
+    output_dir: Path
+    # None uses the harness's default
+    timeout: int | None = None
+
+
+@dataclass(frozen=True)
+class AgentResult:
+    # Final answer; None when the output is left in the workspace
+    response: str | None = None
+    metrics: AgentMetrics | None = None
+    experiment: ExperimentConfiguration | None = None
+
+
+class Harness[T, N: str = str](Protocol):
+    """Runs an agent on tasks of type ``T``, identified by a harness name of type ``N``.
+
+    ``identity`` checks the harness is usable and raises ``AgentError`` otherwise. ``run`` raises
+    ``AgentTimeoutError`` when the timeout is exceeded and ``AgentError`` for any other failure.
+    """
+
+    def identity(self) -> AgentIdentity[N]: ...
+
+    def run(self, request: AgentRequest[T], /) -> AgentResult: ...
diff --git a/src/bcbench/agent/__init__.py b/src/bcbench/agent/__init__.py
index 4f2a4e0b6..9a08648e6 100644
--- a/src/bcbench/agent/__init__.py
+++ b/src/bcbench/agent/__init__.py
@@ -2,7 +2,7 @@
 
 from bcbench.agent.bcal import BCalBackendConfig, get_bcal_version, run_bcal_agent
 from bcbench.agent.claude import get_claude_version, run_claude_code
-from bcbench.agent.copilot import run_copilot_agent
+from bcbench.agent.copilot import CopilotHarness
 from bcbench.agent.pr_review import get_pr_review_version, run_pr_review_agent
 
-__all__ = ["BCalBackendConfig", "get_bcal_version", "get_claude_version", "get_pr_review_version", "run_bcal_agent", "run_claude_code", "run_copilot_agent", "run_pr_review_agent"]
+__all__ = ["BCalBackendConfig", "CopilotHarness", "get_bcal_version", "get_claude_version", "get_pr_review_version", "run_bcal_agent", "run_claude_code", "run_pr_review_agent"]
diff --git a/src/bcbench/agent/bcal/agent.py b/src/bcbench/agent/bcal/agent.py
index 24ff0005a..8764ddb46 100644
--- a/src/bcbench/agent/bcal/agent.py
+++ b/src/bcbench/agent/bcal/agent.py
@@ -7,6 +7,7 @@
 from pathlib import Path
 
 from bcbench_core.agent.experiment import ExperimentConfiguration
+from bcbench_core.agent.harness import AgentResult
 from bcbench_core.agent.metrics import AgentMetrics
 from bcbench_core.agent.version import get_cli_version
 from bcbench_core.artifacts import ALPACKAGES_DIRNAME
@@ -124,7 +125,7 @@ def run_bcal_agent(
     repo_path: Path,
     backend_config: BCalBackendConfig,
     timeout: int = 25 * 60,
-) -> tuple[AgentMetrics | None, ExperimentConfiguration]:
+) -> AgentResult:
     logger.info(f"Running bcal CLI on: {entry.instance_id}")
 
     # The .alpackages dir is created by the NL2AL pipeline setup step
@@ -151,7 +152,7 @@ def run_bcal_agent(
         execution_time = time.monotonic() - start
 
         logger.info(f"bcal CLI run complete for: {entry.instance_id}")
-        return AgentMetrics(execution_time=execution_time), ExperimentConfiguration()
+        return AgentResult(metrics=AgentMetrics(execution_time=execution_time), experiment=ExperimentConfiguration())
     except subprocess.TimeoutExpired:
         logger.exception(f"bcal CLI timed out after {timeout} seconds")
         metrics = AgentMetrics(execution_time=timeout)
diff --git a/src/bcbench/agent/claude/agent.py b/src/bcbench/agent/claude/agent.py
index 8c9f34d24..c04cd5888 100644
--- a/src/bcbench/agent/claude/agent.py
+++ b/src/bcbench/agent/claude/agent.py
@@ -5,6 +5,7 @@
 
 from bcbench_core.agent.customization import setup_custom_agent, setup_instructions, setup_skills
 from bcbench_core.agent.experiment import ExperimentConfiguration
+from bcbench_core.agent.harness import AgentResult
 from bcbench_core.agent.metrics import AgentMetrics
 from bcbench_core.agent.version import get_cli_version
 from bcbench_core.exceptions import AgentError, AgentTimeoutError
@@ -39,12 +40,8 @@ def run_claude_code(
     plugin_root: Path,
     runtime: AgentRuntimeConfig | None = None,
     timeout: int = 60 * 60,
-) -> tuple[AgentMetrics | None, ExperimentConfiguration]:
-    """Run Claude Code on a single dataset entry.
-
-    Returns:
-        Tuple of (AgentMetrics, ExperimentConfiguration) with metrics and configuration used
-    """
+) -> AgentResult:
+    """Run Claude Code on a single dataset entry."""
     claude_cmd = shutil.which("claude")
     if not claude_cmd:
         raise AgentError("Claude Code not found in PATH. Please ensure it is installed and available.")
@@ -145,7 +142,7 @@ def run_claude_code(
         stdout: str = result.stdout.decode("utf-8", errors="replace") if result.stdout else ""
         logger.debug(f"Claude Code raw output: {stdout}")
 
-        metrics, _ = parse_stream_output(stdout.splitlines(), log_transcript=True)
+        metrics, response = parse_stream_output(stdout.splitlines(), log_transcript=True)
     except subprocess.TimeoutExpired:
         logger.exception(f"Claude Code timed out after {timeout} seconds")
         metrics = AgentMetrics(execution_time=timeout)
@@ -157,7 +154,7 @@ def run_claude_code(
         logger.exception("Unexpected error running Claude Code")
         raise
     else:
-        return metrics, config
+        return AgentResult(response=response, metrics=metrics, experiment=config)
     finally:
         if bc_gateway is not None:
             bc_gateway.stop()
diff --git a/src/bcbench/agent/copilot/__init__.py b/src/bcbench/agent/copilot/__init__.py
index 40648ec39..340415886 100644
--- a/src/bcbench/agent/copilot/__init__.py
+++ b/src/bcbench/agent/copilot/__init__.py
@@ -1,5 +1,5 @@
 """GitHub Copilot CLI agent module."""
 
-from bcbench.agent.copilot.agent import run_copilot_agent
+from bcbench.agent.copilot.agent import CopilotHarness
 
-__all__ = ["run_copilot_agent"]
+__all__ = ["CopilotHarness"]
diff --git a/src/bcbench/agent/copilot/agent.py b/src/bcbench/agent/copilot/agent.py
index c95a530bb..2d23d72e0 100644
--- a/src/bcbench/agent/copilot/agent.py
+++ b/src/bcbench/agent/copilot/agent.py
@@ -1,12 +1,14 @@
 """GitHub Copilot CLI Agent implementation."""
 
 import logging
+from dataclasses import dataclass
 from pathlib import Path
+from typing import override
 
-from bcbench_core.agent.copilot import CopilotOptions, CopilotTimeoutError, invoke_copilot
+from bcbench_core.agent.copilot import CopilotOptions, CopilotTimeoutError, get_copilot_version, invoke_copilot
 from bcbench_core.agent.customization import setup_custom_agent, setup_instructions, setup_skills
 from bcbench_core.agent.experiment import ExperimentConfiguration
-from bcbench_core.agent.metrics import AgentMetrics
+from bcbench_core.agent.harness import AgentIdentity, AgentRequest, AgentResult, Harness
 from bcbench_core.exceptions import AgentTimeoutError
 
 from bcbench.agent.shared import (
@@ -24,87 +26,90 @@
 logger = logging.getLogger(__name__)
 
 
-def run_copilot_agent(
-    entry: BaseDatasetEntry,
-    model: str,
-    category: EvaluationCategory,
-    repo_path: Path,
-    output_dir: Path,
-    agent_config: AgentConfig,
-    plugin_root: Path,
-    runtime: AgentRuntimeConfig | None = None,
-    timeout: int = 60 * 60,
-) -> tuple[AgentMetrics | None, ExperimentConfiguration]:
-    """Run GitHub Copilot CLI agent on a single dataset entry.
+@dataclass(frozen=True)
+class CopilotHarness(Harness[BaseDatasetEntry, AgentHarness]):
+    """GitHub Copilot CLI on any category, customized through the agent config."""
 
-    Returns:
-        Tuple of (AgentMetrics, ExperimentConfiguration) with metrics and configuration used during the experiment
-    """
-    logger.info(f"Running GitHub Copilot CLI on: {entry.instance_id}")
+    model: str
+    category: EvaluationCategory
+    agent_config: AgentConfig
+    plugin_root: Path
+    runtime: AgentRuntimeConfig | None = None
+    timeout: int = 60 * 60
 
-    prompt: str = build_prompt(entry, repo_path, agent_config, category, al_mcp=bool(runtime and runtime.al_mcp))
-    bc_gateway = start_bc_mcp_gateway(runtime)
-    mcp_config_json, mcp_server_names = build_mcp_config(
-        agent_config,
-        entry,
-        repo_path,
-        runtime=runtime,
-        bc_mcp_gateway_url=bc_gateway.base_url if bc_gateway else None,
-    )
-    lsp_plugin_dir: Path | None = build_al_lsp_plugin(
-        entry,
-        category,
-        repo_path,
-        AgentHarness.COPILOT,
-        plugin_root,
-        runtime=runtime,
-    )
-    profile_dir: Path = instructions_profile_dir(entry.customization_profile)
-    target_dir: Path = AgentHarness.COPILOT.get_target_dir(repo_path)
-    instructions_enabled: bool = setup_instructions(agent_config.instructions, profile_dir, target_dir, AgentHarness.COPILOT.instruction_filename)
-    skills_enabled: bool = setup_skills(agent_config.skills, profile_dir, target_dir)
-    custom_agent: str | None = setup_custom_agent(agent_config.agents, profile_dir, target_dir)
-    plugins: list[tuple[PluginConfig, Path]] = resolve_config_plugins(agent_config, plugin_root, allow_copilot_manifest=True)
+    @override
+    def identity(self) -> AgentIdentity[AgentHarness]:
+        return AgentIdentity[AgentHarness](harness=AgentHarness.COPILOT, version=get_copilot_version(), model=self.model)
 
-    config = ExperimentConfiguration(
-        mcp_servers=mcp_server_names,
-        al_lsp_enabled=lsp_plugin_dir is not None,
-        custom_instructions=instructions_enabled,
-        skills_enabled=skills_enabled,
-        custom_agent=custom_agent,
-        plugins=[plugin.record for plugin, _ in plugins] or None,
-    )
+    @override
+    def run(self, request: AgentRequest[BaseDatasetEntry], /) -> AgentResult:
+        entry, repo_path, output_dir = request.task, request.work_dir, request.output_dir
+        agent_config, category, runtime, plugin_root = self.agent_config, self.category, self.runtime, self.plugin_root
+        logger.info(f"Running GitHub Copilot CLI on: {entry.instance_id}")
 
-    try:
-        lsp_plugin_dirs = (lsp_plugin_dir,) if lsp_plugin_dir is not None else ()
-        # --add-dir grants read+write (unlike --plugin-dir, which only registers a plugin), so hand it
-        # only to plugins that opt in via grant_dir_access - currently a temporary accommodation for
-        # BCQuality, whose skill reads its own knowledge files at runtime. Enabling a plugin must not
-        # silently widen the agent's sandbox access.
-        options = CopilotOptions(
-            allow_all_tools=True,
+        prompt: str = build_prompt(entry, repo_path, agent_config, category, al_mcp=bool(runtime and runtime.al_mcp))
+        bc_gateway = start_bc_mcp_gateway(runtime)
+        mcp_config_json, mcp_server_names = build_mcp_config(
+            agent_config,
+            entry,
+            repo_path,
+            runtime=runtime,
+            bc_mcp_gateway_url=bc_gateway.base_url if bc_gateway else None,
+        )
+        lsp_plugin_dir: Path | None = build_al_lsp_plugin(
+            entry,
+            category,
+            repo_path,
+            AgentHarness.COPILOT,
+            plugin_root,
+            runtime=runtime,
+        )
+        profile_dir: Path = instructions_profile_dir(entry.customization_profile)
+        target_dir: Path = AgentHarness.COPILOT.get_target_dir(repo_path)
+        instructions_enabled: bool = setup_instructions(agent_config.instructions, profile_dir, target_dir, AgentHarness.COPILOT.instruction_filename)
+        skills_enabled: bool = setup_skills(agent_config.skills, profile_dir, target_dir)
+        custom_agent: str | None = setup_custom_agent(agent_config.agents, profile_dir, target_dir)
+        plugins: list[tuple[PluginConfig, Path]] = resolve_config_plugins(agent_config, plugin_root, allow_copilot_manifest=True)
+
+        config = ExperimentConfiguration(
+            mcp_servers=mcp_server_names,
+            al_lsp_enabled=lsp_plugin_dir is not None,
             custom_instructions=instructions_enabled,
-            log_dir=output_dir,
-            mcp_config_json=mcp_config_json,
-            plugin_dirs=(*lsp_plugin_dirs, *(plugin_dir for _, plugin_dir in plugins)),
-            granted_dirs=tuple(plugin_dir for plugin, plugin_dir in plugins if plugin.grant_dir_access),
+            skills_enabled=skills_enabled,
             custom_agent=custom_agent,
-            workspace_mcp=True,
+            plugins=[plugin.record for plugin, _ in plugins] or None,
         )
 
-        metrics, _ = invoke_copilot(
-            prompt=prompt,
-            model=model,
-            work_dir=repo_path,
-            timeout=timeout,
-            env=agent_subprocess_env(pass_bc_credentials=category.pass_on_bc_container_credentials),
-            options=options,
-        )
-        logger.info(f"Copilot CLI run complete for: {entry.instance_id}")
-    except CopilotTimeoutError as exc:
-        raise AgentTimeoutError("Copilot CLI timed out", metrics=exc.metrics, config=config) from exc
-    else:
-        return metrics, config
-    finally:
-        if bc_gateway is not None:
-            bc_gateway.stop()
+        try:
+            lsp_plugin_dirs = (lsp_plugin_dir,) if lsp_plugin_dir is not None else ()
+            # --add-dir grants read+write (unlike --plugin-dir, which only registers a plugin), so hand it
+            # only to plugins that opt in via grant_dir_access - currently a temporary accommodation for
+            # BCQuality, whose skill reads its own knowledge files at runtime. Enabling a plugin must not
+            # silently widen the agent's sandbox access.
+            options = CopilotOptions(
+                allow_all_tools=True,
+                custom_instructions=instructions_enabled,
+                log_dir=output_dir,
+                mcp_config_json=mcp_config_json,
+                plugin_dirs=(*lsp_plugin_dirs, *(plugin_dir for _, plugin_dir in plugins)),
+                granted_dirs=tuple(plugin_dir for plugin, plugin_dir in plugins if plugin.grant_dir_access),
+                custom_agent=custom_agent,
+                workspace_mcp=True,
+            )
+
+            metrics, response = invoke_copilot(
+                prompt=prompt,
+                model=self.model,
+                work_dir=repo_path,
+                timeout=self.timeout if request.timeout is None else request.timeout,
+                env=agent_subprocess_env(pass_bc_credentials=category.pass_on_bc_container_credentials),
+                options=options,
+            )
+            logger.info(f"Copilot CLI run complete for: {entry.instance_id}")
+        except CopilotTimeoutError as exc:
+            raise AgentTimeoutError("Copilot CLI timed out", metrics=exc.metrics, config=config) from exc
+        else:
+            return AgentResult(response=response, metrics=metrics, experiment=config)
+        finally:
+            if bc_gateway is not None:
+                bc_gateway.stop()
diff --git a/src/bcbench/agent/pr_review/agent.py b/src/bcbench/agent/pr_review/agent.py
index a68fc2c1d..87f62f81d 100644
--- a/src/bcbench/agent/pr_review/agent.py
+++ b/src/bcbench/agent/pr_review/agent.py
@@ -22,6 +22,7 @@
 
 import yaml
 from bcbench_core.agent.experiment import ExperimentConfiguration
+from bcbench_core.agent.harness import AgentResult
 from bcbench_core.exceptions import AgentError, AgentTimeoutError
 from bcbench_core.git import commit_changes, has_changes, init_repo
 
@@ -161,16 +162,13 @@ def run_pr_review_agent(
     engine_path: Path | None = None,
     min_severity: str | None = None,
     timeout: int = 60 * 60,
-) -> tuple[PRReviewMetrics, ExperimentConfiguration]:
+) -> AgentResult:
     """Run the engine's complete local review pipeline and write review.json.
 
     Separate from run_copilot_agent by design: this spawns the PROD BC-ALAgents
     PowerShell orchestrator (Copilot is spawned inside the engine, not here), so it
     owns none of the copilot-harness prompt/MCP/LSP wiring and takes engine-specific
     inputs (BC-ALAgents source, min severity) for the code-review category only.
-
-    Returns:
-        Tuple of (PRReviewMetrics, ExperimentConfiguration).
     """
     if category is not EvaluationCategory.CODE_REVIEW:
         raise AgentError(f"The engine agent only supports the code-review category, got {category.value}.")
@@ -235,4 +233,4 @@ def run_pr_review_agent(
         logger.exception("Unexpected error running engine review")
         raise
     else:
-        return build_pr_review_metrics(output_dir, time.monotonic() - start), config
+        return AgentResult(metrics=build_pr_review_metrics(output_dir, time.monotonic() - start), experiment=config)
diff --git a/src/bcbench/commands/evaluate.py b/src/bcbench/commands/evaluate.py
index 963fbac77..925249d03 100644
--- a/src/bcbench/commands/evaluate.py
+++ b/src/bcbench/commands/evaluate.py
@@ -4,13 +4,13 @@
 from typing import Annotated, cast, override
 
 import typer
-from bcbench_core.agent.copilot import get_copilot_version
 from bcbench_core.agent.experiment import ExperimentConfiguration
+from bcbench_core.agent.harness import AgentIdentity, AgentResult
 from bcbench_core.agent.metrics import AgentMetrics
 from bcbench_core.exceptions import AgentError
 from bcbench_core.filesystem import prepare_run_dir
 
-from bcbench.agent import BCalBackendConfig, get_bcal_version, get_claude_version, get_pr_review_version, run_bcal_agent, run_claude_code, run_copilot_agent, run_pr_review_agent
+from bcbench.agent import BCalBackendConfig, CopilotHarness, get_bcal_version, get_claude_version, get_pr_review_version, run_bcal_agent, run_claude_code, run_pr_review_agent
 from bcbench.cli_options import (
     AgentConfigPath,
     ClaudeCodeModel,
@@ -33,7 +33,7 @@
 from bcbench.config import get_config
 from bcbench.dataset import BaseDatasetEntry, NL2ALEntry
 from bcbench.dataset.codereview import CodeReviewEntry
-from bcbench.evaluate import AgentRunner, CodeReviewPipeline, EvaluationPipeline, NL2ALPipeline
+from bcbench.evaluate import AgentRunner, CodeReviewPipeline, EvaluationPipeline, NL2ALPipeline, harness_runner
 from bcbench.evaluate.codereview_judge_calibration import run_calibration
 from bcbench.paths import SHARED_CONFIG_FILE
 from bcbench.results import BaseEvaluationResult, CodeReviewResult, ExecutionBasedEvaluationResult, JudgeBasedEvaluationResult
@@ -86,6 +86,7 @@ def evaluate_copilot(
     )
     entry = category.entry_class.load(category.dataset_path(tier), entry_id=entry_id)[0]
     agent_config = AgentConfig.from_file(config_path)
+    harness = CopilotHarness(model=model, category=category, agent_config=agent_config, plugin_root=_config.paths.plugin_root, runtime=runtime)
     run_dir = prepare_run_dir(output_dir, run_id)
 
     logger.info(f"Running evaluation on entry {entry_id} with GitHub Copilot CLI")
@@ -95,27 +96,13 @@ def evaluate_copilot(
         repo_path=repo_path,
         result_dir=run_dir,
         container=runtime.container if runtime else None,
-        model=model,
-        agent_name=AgentHarness.COPILOT,
-        agent_version=get_copilot_version(),
+        agent=harness.identity(),
         category=category,
         tier=tier,
     )
 
     pipeline = category.pipeline(agent_config)
-    pipeline.execute(
-        context,
-        lambda ctx: run_copilot_agent(
-            entry=ctx.entry,
-            repo_path=ctx.repo_path,
-            category=category,
-            model=ctx.model,
-            output_dir=ctx.result_dir,
-            agent_config=agent_config,
-            plugin_root=_config.paths.plugin_root,
-            runtime=runtime,
-        ),
-    )
+    pipeline.execute(context, harness_runner(harness))
 
     logger.info("Evaluation complete!")
     logger.info(f"Results saved to: {run_dir}")
@@ -171,9 +158,7 @@ def evaluate_claude_code(
         repo_path=repo_path,
         result_dir=run_dir,
         container=runtime.container if runtime else None,
-        model=model,
-        agent_name=AgentHarness.CLAUDE,
-        agent_version=get_claude_version(),
+        agent=AgentIdentity[AgentHarness](harness=AgentHarness.CLAUDE, version=get_claude_version(), model=model),
         category=category,
         tier=tier,
     )
@@ -185,7 +170,7 @@ def evaluate_claude_code(
             entry=ctx.entry,
             repo_path=ctx.repo_path,
             category=category,
-            model=ctx.model,
+            model=model,
             output_dir=ctx.result_dir,
             agent_config=agent_config,
             plugin_root=_config.paths.plugin_root,
@@ -228,9 +213,7 @@ def evaluate_pr_review(
         repo_path=repo_path,
         result_dir=run_dir,
         container=None,
-        model=model,
-        agent_name=AgentHarness.PR_REVIEW,
-        agent_version=get_pr_review_version(engine_path),
+        agent=AgentIdentity[AgentHarness](harness=AgentHarness.PR_REVIEW, version=get_pr_review_version(engine_path), model=model),
         category=category,
     )
 
@@ -240,7 +223,7 @@ def evaluate_pr_review(
             entry=ctx.entry,
             repo_path=ctx.repo_path,
             category=category,
-            model=ctx.model,
+            model=model,
             output_dir=ctx.result_dir,
             engine_path=engine_path,
         ),
@@ -280,9 +263,7 @@ def evaluate_bcal(
         repo_path=repo_path,
         result_dir=run_dir,
         container=None,
-        model=llm_model or "external-command",
-        agent_name=AgentHarness.BCAL,
-        agent_version=get_bcal_version(),
+        agent=AgentIdentity[AgentHarness](harness=AgentHarness.BCAL, version=get_bcal_version(), model=llm_model or "external-command"),
         category=category,
         tier=tier,
     )
@@ -352,14 +333,13 @@ def evaluate_mock(
         entry=entry,
         repo_path=Path(),
         result_dir=run_dir,
-        model="mock-model",
-        agent_name=AgentHarness.MOCK,
+        agent=AgentIdentity[AgentHarness](harness=AgentHarness.MOCK, version="mock", model="mock-model"),
         category=category,
         tier=tier,
     )
 
     pipeline = MockEvaluationPipeline()
-    pipeline.execute(context, lambda ctx: (None, None))
+    pipeline.execute(context, lambda _ctx: AgentResult())
 
     logger.info("Mock evaluation complete!")
     logger.info(f"Results saved to: {run_dir}")
diff --git a/src/bcbench/commands/run.py b/src/bcbench/commands/run.py
index 1f4e024fa..c259c52fe 100644
--- a/src/bcbench/commands/run.py
+++ b/src/bcbench/commands/run.py
@@ -4,8 +4,9 @@
 from typing import Annotated, cast
 
 import typer
+from bcbench_core.agent.harness import AgentRequest
 
-from bcbench.agent import BCalBackendConfig, run_bcal_agent, run_claude_code, run_copilot_agent, run_pr_review_agent
+from bcbench.agent import BCalBackendConfig, CopilotHarness, run_bcal_agent, run_claude_code, run_pr_review_agent
 from bcbench.cli_options import (
     AgentConfigPath,
     ClaudeCodeModel,
@@ -81,16 +82,8 @@ def run_copilot(
     agent_config = AgentConfig.from_file(config_path)
     category.pipeline(agent_config).setup_workspace(entry, repo_path)
 
-    run_copilot_agent(
-        entry=entry,
-        repo_path=repo_path,
-        model=model,
-        category=category,
-        output_dir=output_dir,
-        agent_config=agent_config,
-        plugin_root=_config.paths.plugin_root,
-        runtime=runtime,
-    )
+    harness = CopilotHarness(model=model, category=category, agent_config=agent_config, plugin_root=_config.paths.plugin_root, runtime=runtime)
+    harness.run(AgentRequest(task=entry, work_dir=repo_path, output_dir=output_dir))
 
 
 @run_app.command("claude")
diff --git a/src/bcbench/evaluate/__init__.py b/src/bcbench/evaluate/__init__.py
index a0e7f1fd4..49ee8eb15 100644
--- a/src/bcbench/evaluate/__init__.py
+++ b/src/bcbench/evaluate/__init__.py
@@ -1,6 +1,6 @@
 """Evaluation module for running pipelines and creating results."""
 
-from bcbench.evaluate.base import AgentRunner, EvaluationPipeline
+from bcbench.evaluate.base import AgentRunner, EvaluationPipeline, harness_runner
 from bcbench.evaluate.bugfix import BugFixPipeline
 from bcbench.evaluate.codereview import CodeReviewPipeline
 from bcbench.evaluate.dataquery import DataQueryPipeline
@@ -21,4 +21,5 @@
     "ExtRequestTriagePipeline",
     "NL2ALPipeline",
     "TestGenerationPipeline",
+    "harness_runner",
 ]
diff --git a/src/bcbench/evaluate/base.py b/src/bcbench/evaluate/base.py
index 8d2750619..8b037a6b2 100644
--- a/src/bcbench/evaluate/base.py
+++ b/src/bcbench/evaluate/base.py
@@ -5,22 +5,30 @@
 from pathlib import Path
 from typing import Protocol
 
-from bcbench_core.agent.experiment import ExperimentConfiguration
-from bcbench_core.agent.metrics import AgentMetrics
+from bcbench_core.agent.harness import AgentRequest, AgentResult, Harness
 from bcbench_core.exceptions import AgentTimeoutError
 
 from bcbench.dataset import BaseDatasetEntry
 from bcbench.results import BaseEvaluationResult
 from bcbench.results.base import RESULT_FILE_SUFFIX
-from bcbench.types import EvaluationContext
+from bcbench.types import AgentHarness, EvaluationContext
 
 logger = logging.getLogger(__name__)
 
-__all__ = ["AgentRunner", "EvaluationPipeline"]
+__all__ = ["AgentRunner", "EvaluationPipeline", "harness_runner"]
 
 
 class AgentRunner[E: BaseDatasetEntry](Protocol):
-    def __call__(self, context: EvaluationContext[E], /) -> tuple[AgentMetrics | None, ExperimentConfiguration | None]: ...
+    def __call__(self, context: EvaluationContext[E], /) -> AgentResult: ...
+
+
+def harness_runner[E: BaseDatasetEntry](harness: Harness[E, AgentHarness]) -> AgentRunner[E]:
+    """Adapt ``harness`` to the pipelines' ``AgentRunner`` until every harness implements ``Harness``."""
+
+    def run(context: EvaluationContext[E], /) -> AgentResult:
+        return harness.run(AgentRequest(task=context.entry, work_dir=context.repo_path, output_dir=context.result_dir))
+
+    return run
 
 
 class EvaluationPipeline[E: BaseDatasetEntry](ABC):
diff --git a/src/bcbench/evaluate/bugfix.py b/src/bcbench/evaluate/bugfix.py
index 4f513f40f..8dded4c63 100644
--- a/src/bcbench/evaluate/bugfix.py
+++ b/src/bcbench/evaluate/bugfix.py
@@ -44,8 +44,9 @@ def setup(self, context: EvaluationContext[BugFixEntry]) -> None:
 
     @override
     def run_agent(self, context: EvaluationContext[BugFixEntry], agent_runner: AgentRunner[BugFixEntry]) -> None:
-        with github_log_group(f"{context.agent_name} -- Entry: {context.entry.instance_id}"):
-            context.metrics, context.experiment = agent_runner(context)
+        with github_log_group(f"{context.agent.harness} -- Entry: {context.entry.instance_id}"):
+            result = agent_runner(context)
+            context.metrics, context.experiment = result.metrics, result.experiment
 
     @override
     def evaluate(self, context: EvaluationContext[BugFixEntry]) -> None:
diff --git a/src/bcbench/evaluate/codereview.py b/src/bcbench/evaluate/codereview.py
index 8ad821415..7fa87f7f8 100644
--- a/src/bcbench/evaluate/codereview.py
+++ b/src/bcbench/evaluate/codereview.py
@@ -56,8 +56,9 @@ def setup(self, context: EvaluationContext[CodeReviewEntry]) -> None:
 
     @override
     def run_agent(self, context: EvaluationContext[CodeReviewEntry], agent_runner: AgentRunner[CodeReviewEntry]) -> None:
-        with github_log_group(f"{context.agent_name} -- Entry: {context.entry.instance_id}"):
-            context.metrics, context.experiment = agent_runner(context)
+        with github_log_group(f"{context.agent.harness} -- Entry: {context.entry.instance_id}"):
+            result = agent_runner(context)
+            context.metrics, context.experiment = result.metrics, result.experiment
 
     @override
     def evaluate(self, context: EvaluationContext[CodeReviewEntry]) -> None:
diff --git a/src/bcbench/evaluate/dataquery.py b/src/bcbench/evaluate/dataquery.py
index 77ba43705..2665e1a5a 100644
--- a/src/bcbench/evaluate/dataquery.py
+++ b/src/bcbench/evaluate/dataquery.py
@@ -106,8 +106,9 @@ def setup(self, context: EvaluationContext[DataQueryEntry]) -> None:
 
     @override
     def run_agent(self, context: EvaluationContext[DataQueryEntry], agent_runner: AgentRunner[DataQueryEntry]) -> None:
-        with github_log_group(f"{context.agent_name} -- Entry: {context.entry.instance_id}"):
-            context.metrics, context.experiment = agent_runner(context)
+        with github_log_group(f"{context.agent.harness} -- Entry: {context.entry.instance_id}"):
+            result = agent_runner(context)
+            context.metrics, context.experiment = result.metrics, result.experiment
 
     @override
     def evaluate(self, context: EvaluationContext[DataQueryEntry]) -> None:
diff --git a/src/bcbench/evaluate/ext_request_advisor.py b/src/bcbench/evaluate/ext_request_advisor.py
index 7c98a12d0..58e0e23ad 100644
--- a/src/bcbench/evaluate/ext_request_advisor.py
+++ b/src/bcbench/evaluate/ext_request_advisor.py
@@ -30,8 +30,9 @@ def setup(self, context: EvaluationContext[ExtRequestAdvisorEntry]) -> None:
 
     @override
     def run_agent(self, context: EvaluationContext[ExtRequestAdvisorEntry], agent_runner: AgentRunner[ExtRequestAdvisorEntry]) -> None:
-        with github_log_group(f"{context.agent_name} -- Entry: {context.entry.instance_id}"):
-            context.metrics, context.experiment = agent_runner(context)
+        with github_log_group(f"{context.agent.harness} -- Entry: {context.entry.instance_id}"):
+            result = agent_runner(context)
+            context.metrics, context.experiment = result.metrics, result.experiment
 
     @override
     def evaluate(self, context: EvaluationContext[ExtRequestAdvisorEntry]) -> None:
diff --git a/src/bcbench/evaluate/ext_request_implement.py b/src/bcbench/evaluate/ext_request_implement.py
index 79d6ee20c..d439d9aee 100644
--- a/src/bcbench/evaluate/ext_request_implement.py
+++ b/src/bcbench/evaluate/ext_request_implement.py
@@ -37,8 +37,9 @@ def setup(self, context: EvaluationContext[ExtRequestImplementEntry]) -> None:
 
     @override
     def run_agent(self, context: EvaluationContext[ExtRequestImplementEntry], agent_runner: AgentRunner[ExtRequestImplementEntry]) -> None:
-        with github_log_group(f"{context.agent_name} -- Entry: {context.entry.instance_id}"):
-            context.metrics, context.experiment = agent_runner(context)
+        with github_log_group(f"{context.agent.harness} -- Entry: {context.entry.instance_id}"):
+            result = agent_runner(context)
+            context.metrics, context.experiment = result.metrics, result.experiment
 
     @override
     def evaluate(self, context: EvaluationContext[ExtRequestImplementEntry]) -> None:
diff --git a/src/bcbench/evaluate/ext_request_triage.py b/src/bcbench/evaluate/ext_request_triage.py
index 1743916e8..5366bef71 100644
--- a/src/bcbench/evaluate/ext_request_triage.py
+++ b/src/bcbench/evaluate/ext_request_triage.py
@@ -40,8 +40,9 @@ def setup(self, context: EvaluationContext[ExtRequestTriageEntry]) -> None:
 
     @override
     def run_agent(self, context: EvaluationContext[ExtRequestTriageEntry], agent_runner: AgentRunner[ExtRequestTriageEntry]) -> None:
-        with github_log_group(f"{context.agent_name} -- Entry: {context.entry.instance_id}"):
-            context.metrics, context.experiment = agent_runner(context)
+        with github_log_group(f"{context.agent.harness} -- Entry: {context.entry.instance_id}"):
+            result = agent_runner(context)
+            context.metrics, context.experiment = result.metrics, result.experiment
 
     @override
     def evaluate(self, context: EvaluationContext[ExtRequestTriageEntry]) -> None:
diff --git a/src/bcbench/evaluate/nl2al.py b/src/bcbench/evaluate/nl2al.py
index e04e4c65b..40cb7a99d 100644
--- a/src/bcbench/evaluate/nl2al.py
+++ b/src/bcbench/evaluate/nl2al.py
@@ -71,8 +71,9 @@ def setup(self, context: EvaluationContext[NL2ALEntry]) -> None:
     def run_agent(self, context: EvaluationContext[NL2ALEntry], agent_runner: AgentRunner[NL2ALEntry]) -> None:
         # Single attempt — retries are disabled. An empty diff (the agent asked for clarification
         # instead of editing) is scored as a failure in evaluate(), not re-run.
-        with github_log_group(f"{context.agent_name} -- Entry: {context.entry.instance_id}"):
-            context.metrics, context.experiment = agent_runner(context)
+        with github_log_group(f"{context.agent.harness} -- Entry: {context.entry.instance_id}"):
+            result = agent_runner(context)
+            context.metrics, context.experiment = result.metrics, result.experiment
 
     @override
     def evaluate(self, context: EvaluationContext[NL2ALEntry]) -> None:
diff --git a/src/bcbench/evaluate/testgeneration.py b/src/bcbench/evaluate/testgeneration.py
index a72b6f526..12e7704d9 100644
--- a/src/bcbench/evaluate/testgeneration.py
+++ b/src/bcbench/evaluate/testgeneration.py
@@ -66,8 +66,9 @@ def setup(self, context: EvaluationContext[TestGenEntry]) -> None:
 
     @override
     def run_agent(self, context: EvaluationContext[TestGenEntry], agent_runner: AgentRunner[TestGenEntry]) -> None:
-        with github_log_group(f"{context.agent_name} -- Entry: {context.entry.instance_id}"):
-            context.metrics, context.experiment = agent_runner(context)
+        with github_log_group(f"{context.agent.harness} -- Entry: {context.entry.instance_id}"):
+            result = agent_runner(context)
+            context.metrics, context.experiment = result.metrics, result.experiment
 
     @override
     def evaluate(self, context: EvaluationContext[TestGenEntry]) -> None:
diff --git a/src/bcbench/results/base.py b/src/bcbench/results/base.py
index 765617267..16b32237b 100644
--- a/src/bcbench/results/base.py
+++ b/src/bcbench/results/base.py
@@ -38,22 +38,22 @@ class BaseEvaluationResult(BaseModel):
 
     @classmethod
     def _base_fields[E: BaseDatasetEntry](cls, context: EvaluationContext[E]) -> dict[str, Any]:
-        metrics_contract = context.agent_name.metrics_contract
+        metrics_contract = context.agent.harness.metrics_contract
         if not context.metrics:
             logger.warning(f"Creating result for {context.entry.instance_id} with no agent metrics - performance data will be unavailable")
         elif type(context.metrics) is not metrics_contract.metrics_type:
-            raise TypeError(f"{context.agent_name} must use {metrics_contract.metrics_type.__name__}, got {type(context.metrics).__name__}")
+            raise TypeError(f"{context.agent.harness} must use {metrics_contract.metrics_type.__name__}, got {type(context.metrics).__name__}")
         elif missing_metrics := sorted(name for name in metrics_contract.required_fields if getattr(context.metrics, name) is None):
             logger.warning(f"Result for {context.entry.instance_id} missing metrics: {', '.join(missing_metrics)}")
 
         return {
             "instance_id": context.entry.instance_id,
             "project": context.entry.extract_project_name(),
-            "model": context.model.replace(".", "-"),
+            "model": (context.agent.model or "").replace(".", "-"),
             "category": context.category,
             "tier": context.tier,
-            "agent_name": context.agent_name,
-            "agent_version": context.agent_version,
+            "agent_name": context.agent.harness,
+            "agent_version": context.agent.version,
             "metrics": context.metrics,
             "experiment": context.experiment,
         }
diff --git a/src/bcbench/types.py b/src/bcbench/types.py
index c75a1e733..0500a2c08 100644
--- a/src/bcbench/types.py
+++ b/src/bcbench/types.py
@@ -10,6 +10,7 @@
 import yaml
 from bcbench_core.agent.customization import CustomAgentsConfig, ToggleConfig
 from bcbench_core.agent.experiment import ExperimentConfiguration
+from bcbench_core.agent.harness import AgentIdentity
 from bcbench_core.agent.metrics import AgentMetrics
 from bcbench_core.container import ContainerConfig
 from pydantic import BaseModel, ConfigDict, Field, StrictBool, StringConstraints, model_validator
@@ -629,9 +630,8 @@ class EvaluationContext[E: BaseDatasetEntry]:
     repo_path: Path
     result_dir: Path
 
-    # Agent metadata
-    agent_name: AgentHarness
-    model: str
+    # The agent under evaluation, resolved before running
+    agent: AgentIdentity[AgentHarness]
 
     # Evaluation category
     category: EvaluationCategory
@@ -640,8 +640,6 @@ class EvaluationContext[E: BaseDatasetEntry]:
     # BC Container configuration (optional — not all categories require a container)
     container: ContainerConfig | None = None
 
-    agent_version: str | None = None
-
     # Agent execution metrics
     metrics: AgentMetrics | None = None
 
diff --git a/tests/conftest.py b/tests/conftest.py
index 0aae5ea5f..1ac1d615e 100644
--- a/tests/conftest.py
+++ b/tests/conftest.py
@@ -13,6 +13,7 @@
 
 import pytest
 from bcbench_core.agent.experiment import ExperimentConfiguration
+from bcbench_core.agent.harness import AgentIdentity
 from bcbench_core.agent.metrics import AgentMetrics
 from bcbench_core.container import ContainerConfig
 from bcbench_core.dataset import TestEntry
@@ -85,6 +86,7 @@ def create_evaluation_context[EntryT: BaseDatasetEntry](
     entry: EntryT | None = None,
     agent_name: AgentHarness = AgentHarness.COPILOT,
     model: str = "test-model",
+    agent_version: str = "1.0.0",
     category: EvaluationCategory = EvaluationCategory.BUG_FIX,
     container_name: str = "test-container",
     password: str = "test-password",
@@ -97,8 +99,7 @@ def create_evaluation_context[EntryT: BaseDatasetEntry](
         repo_path=tmp_path / "repo",
         result_dir=tmp_path / "results",
         container=ContainerConfig(name=container_name, username=username, password=password, company="CRONUS"),
-        agent_name=agent_name,
-        model=model,
+        agent=AgentIdentity[AgentHarness](harness=agent_name, version=agent_version, model=model),
         category=category,
     )
 
@@ -218,8 +219,7 @@ def create_codereview_result(
         repo_path=Path(),
         result_dir=Path(),
         container=ContainerConfig(name="t", username="t", password="t", company="CRONUS"),
-        agent_name=agent_name,
-        model=model,
+        agent=AgentIdentity[AgentHarness](harness=agent_name, version="1.0.0", model=model),
         category=EvaluationCategory.CODE_REVIEW,
     )
     context.metrics = metrics
diff --git a/tests/test_agent_version_results.py b/tests/test_agent_version_results.py
index 1af54016f..e785746f2 100644
--- a/tests/test_agent_version_results.py
+++ b/tests/test_agent_version_results.py
@@ -17,9 +17,7 @@
 
 @pytest.mark.parametrize("version", ["1.2.3", "a" * 40, None])
 def test_agent_version_round_trips_through_results_and_summaries(tmp_path: Path, version: str | None) -> None:
-    context = create_evaluation_context(tmp_path)
-    context.agent_version = version
-    result = BugFixResult.create_success(context, "patch")
+    result = BugFixResult.create_success(create_evaluation_context(tmp_path), "patch").model_copy(update={"agent_version": version})
     result.save(tmp_path, "result.jsonl")
     restored = BaseEvaluationResult.from_json(json.loads((tmp_path / "result.jsonl").read_text()))
     summary = EvaluationResultSummary.from_results([restored], "run")
@@ -32,8 +30,7 @@ def test_agent_version_round_trips_through_results_and_summaries(tmp_path: Path,
 
 
 def test_timeout_result_keeps_version_resolved_before_execution(tmp_path: Path) -> None:
-    context = create_evaluation_context(tmp_path)
-    context.agent_version = "1.2.3"
+    context = create_evaluation_context(tmp_path, agent_version="1.2.3")
     pipeline = context.category.pipeline(AgentConfig())
     with (
         patch.object(type(pipeline), "setup"),
diff --git a/tests/test_bcal_agent_provider.py b/tests/test_bcal_agent_provider.py
index 062d9d31d..29a364b30 100644
--- a/tests/test_bcal_agent_provider.py
+++ b/tests/test_bcal_agent_provider.py
@@ -63,7 +63,7 @@ def fake_run(args: list[str], **_: object) -> MagicMock:
             patch.object(bcal_agent, "_resolve_bcal_executable", return_value="C:\\fake\\bcal.exe"),
             patch.object(subprocess, "run", side_effect=fake_run),
         ):
-            metrics, _ = bcal_agent.run_bcal_agent(
+            result = bcal_agent.run_bcal_agent(
                 entry=entry,
                 repo_path=workspace,
                 backend_config=BCalBackendConfig(
@@ -72,7 +72,7 @@ def fake_run(args: list[str], **_: object) -> MagicMock:
                 ),
             )
 
-        assert metrics is not None
+        assert result.metrics is not None
         args = captured["args"]
         assert "--deployment=gpt-5" in args
         assert "--llm-backend=external-command" in args
diff --git a/tests/test_cli_commands.py b/tests/test_cli_commands.py
index b1a4d004d..0060be941 100644
--- a/tests/test_cli_commands.py
+++ b/tests/test_cli_commands.py
@@ -142,8 +142,8 @@ def execute(self, context, _agent_runner):
             llm_model=llm_model,
         )
 
-    assert captured["context"].model == expected_model
-    assert captured["context"].agent_version == "18.0.1.2-beta"
+    assert captured["context"].agent.model == expected_model
+    assert captured["context"].agent.version == "18.0.1.2-beta"
 
 
 def _option_names(command: click.Command) -> set[str]:
@@ -190,9 +190,9 @@ def execute(self, context, agent_runner):
 @pytest.mark.parametrize(
     ("commands", "command_name", "agent_name", "extra_args"),
     [
-        (run_commands, "run_copilot", "run_copilot_agent", {}),
+        (run_commands, "run_copilot", "CopilotHarness", {}),
         (run_commands, "run_claude", "run_claude_code", {}),
-        (evaluate_commands, "evaluate_copilot", "run_copilot_agent", {"run_id": "test-run"}),
+        (evaluate_commands, "evaluate_copilot", "CopilotHarness", {"run_id": "test-run"}),
         (evaluate_commands, "evaluate_claude_code", "run_claude_code", {"run_id": "test-run"}),
     ],
 )
@@ -201,7 +201,6 @@ def test_agent_commands_preserve_requested_mcp_flags(tmp_path, agent_command_cat
 
     with (
         patch.object(commands, agent_name) as run_agent,
-        patch.object(evaluate_commands, "get_copilot_version", return_value="1.2.3"),
         patch.object(evaluate_commands, "get_claude_version", return_value="1.2.3"),
     ):
         getattr(commands, command_name)(
@@ -226,9 +225,9 @@ def test_agent_commands_preserve_requested_mcp_flags(tmp_path, agent_command_cat
 @pytest.mark.parametrize(
     ("commands", "command_name", "agent_name", "extra_args"),
     [
-        (run_commands, "run_copilot", "run_copilot_agent", {}),
+        (run_commands, "run_copilot", "CopilotHarness", {}),
         (run_commands, "run_claude", "run_claude_code", {}),
-        (evaluate_commands, "evaluate_copilot", "run_copilot_agent", {"run_id": "test-run"}),
+        (evaluate_commands, "evaluate_copilot", "CopilotHarness", {"run_id": "test-run"}),
         (evaluate_commands, "evaluate_claude_code", "run_claude_code", {"run_id": "test-run"}),
     ],
 )
@@ -239,7 +238,6 @@ def test_agent_commands_pass_the_given_agent_config(tmp_path, agent_command_cate
 
     with (
         patch.object(commands, agent_name) as run_agent,
-        patch.object(evaluate_commands, "get_copilot_version", return_value="1.2.3"),
         patch.object(evaluate_commands, "get_claude_version", return_value="1.2.3"),
     ):
         getattr(commands, command_name)(
@@ -324,7 +322,7 @@ def test_result_summarize_creates_all_outputs(sample_results_directory, problem_
     with (
         patch.object(_BugFixTestGenBase, "problem_statement_dir", property(lambda self: problem_statement_dir)),
         patch.object(EvaluationCategory, "dataset_path", return_value=dataset_path),
-        patch.object(evaluate_commands, "get_copilot_version", side_effect=AssertionError("Must use artifact version")),
+        patch("bcbench.agent.copilot.agent.get_copilot_version", side_effect=AssertionError("Must use artifact version")),
         patch.object(evaluate_commands, "get_claude_version", side_effect=AssertionError("Must use artifact version")),
         patch.object(evaluate_commands, "get_pr_review_version", side_effect=AssertionError("Must use artifact version")),
     ):
diff --git a/tests/test_copilot_agent.py b/tests/test_copilot_agent.py
index ab4bd4f74..4a47b8c55 100644
--- a/tests/test_copilot_agent.py
+++ b/tests/test_copilot_agent.py
@@ -5,14 +5,30 @@
 import pytest
 from bcbench_core.agent.copilot import CopilotProcessError, CopilotTimeoutError
 from bcbench_core.agent.experiment import ExperimentConfiguration
+from bcbench_core.agent.harness import AgentRequest, AgentResult
 from bcbench_core.exceptions import AgentTimeoutError
 
-from bcbench.agent.copilot.agent import run_copilot_agent
+from bcbench.agent.copilot.agent import CopilotHarness
+from bcbench.dataset import BaseDatasetEntry
 from bcbench.operations import instructions_profile_dir
 from bcbench.types import AgentConfig, EvaluationCategory, PluginConfig
 from tests.conftest import create_dataset_entry
 
 
+def run_copilot_harness(
+    entry: BaseDatasetEntry,
+    model: str,
+    category: EvaluationCategory,
+    repo_path: Path,
+    output_dir: Path,
+    agent_config: AgentConfig,
+    plugin_root: Path,
+    timeout: int | None = None,
+) -> AgentResult:
+    harness = CopilotHarness(model=model, category=category, agent_config=agent_config, plugin_root=plugin_root)
+    return harness.run(AgentRequest(task=entry, work_dir=repo_path, output_dir=output_dir, timeout=timeout))
+
+
 def test_copilot_does_not_enable_hooks_memory_or_unrestricted_urls(tmp_path: Path, monkeypatch):
     repo_path = tmp_path / "repo"
     output_dir = tmp_path / "output"
@@ -30,13 +46,13 @@ def test_copilot_does_not_enable_hooks_memory_or_unrestricted_urls(tmp_path: Pat
         patch("bcbench.agent.copilot.agent.setup_skills", return_value=False),
         patch("bcbench.agent.copilot.agent.setup_custom_agent", return_value=None),
         patch("bcbench.agent.copilot.agent.resolve_config_plugins", return_value=[]),
-        patch("bcbench_core.agent.copilot.agent.parse_output", return_value=(None, None)) as mock_parse_output,
+        patch("bcbench_core.agent.copilot.agent.parse_output", return_value=(None, "final answer")) as mock_parse_output,
         patch(
             "bcbench_core.agent.copilot.agent.subprocess.run",
             return_value=subprocess.CompletedProcess(args=[], returncode=0, stdout='{"type":"result"}\n', stderr=""),
         ) as mock_run,
     ):
-        run_copilot_agent(
+        result = run_copilot_harness(
             entry=create_dataset_entry(),
             model="copilot-test-model",
             category=EvaluationCategory.BUG_FIX,
@@ -46,6 +62,7 @@ def test_copilot_does_not_enable_hooks_memory_or_unrestricted_urls(tmp_path: Pat
             plugin_root=tmp_path / "plugins",
         )
 
+    assert result.response == "final answer"
     assert mock_run.call_args.args[0] == [
         "copilot",
         "--output-format=json",
@@ -95,7 +112,7 @@ def test_copilot_session_options_preserve_plugin_order_and_explicit_directory_gr
             return_value=subprocess.CompletedProcess(args=[], returncode=0, stdout='{"type":"result"}\n', stderr=""),
         ) as mock_run,
     ):
-        _, config = run_copilot_agent(
+        config = run_copilot_harness(
             entry=create_dataset_entry(),
             model="copilot-test-model",
             category=EvaluationCategory.BUG_FIX,
@@ -103,7 +120,7 @@ def test_copilot_session_options_preserve_plugin_order_and_explicit_directory_gr
             output_dir=output_dir,
             agent_config=AgentConfig(),
             plugin_root=tmp_path / "plugins",
-        )
+        ).experiment
 
     assert mock_run.call_args.args[0] == [
         "copilot",
@@ -121,6 +138,7 @@ def test_copilot_session_options_preserve_plugin_order_and_explicit_directory_gr
         "--agent=al-dev",
         "--prompt=do the task",
     ]
+    assert config is not None
     assert config.mcp_servers == ["probe"]
     assert config.al_lsp_enabled is lsp_enabled
     assert config.custom_instructions is True
@@ -161,7 +179,7 @@ def test_copilot_timeout_adapts_core_metrics_and_preserves_experiment_metadata(c
     run.side_effect = subprocess.TimeoutExpired(["copilot"], timeout)
 
     with pytest.raises(AgentTimeoutError, match="Copilot CLI timed out") as error:
-        run_copilot_agent(create_dataset_entry(), "test-model", EvaluationCategory.BUG_FIX, repo_path, output_dir, AgentConfig(), repo_path.parent / "plugins", timeout=timeout)
+        run_copilot_harness(create_dataset_entry(), "test-model", EvaluationCategory.BUG_FIX, repo_path, output_dir, AgentConfig(), repo_path.parent / "plugins", timeout=timeout)
 
     assert error.value.metrics is not None
     assert error.value.metrics.execution_time == timeout
@@ -177,7 +195,7 @@ def test_copilot_process_errors_propagate_and_stop_gateway(configured_copilot_ru
     run.side_effect = failure
 
     with pytest.raises(CopilotProcessError) as error:
-        run_copilot_agent(create_dataset_entry(), "test-model", EvaluationCategory.BUG_FIX, repo_path, output_dir, AgentConfig(), repo_path.parent / "plugins")
+        run_copilot_harness(create_dataset_entry(), "test-model", EvaluationCategory.BUG_FIX, repo_path, output_dir, AgentConfig(), repo_path.parent / "plugins")
 
     assert error.value.__cause__ is failure
     gateway.stop.assert_called_once_with()
@@ -187,7 +205,7 @@ def test_missing_copilot_stops_gateway_without_invoking_a_process(configured_cop
     repo_path, output_dir, gateway, run = configured_copilot_run
 
     with patch("bcbench_core.agent.copilot.agent._find_copilot", return_value=None), pytest.raises(CopilotProcessError, match="not found"):
-        run_copilot_agent(create_dataset_entry(), "test-model", EvaluationCategory.BUG_FIX, repo_path, output_dir, AgentConfig(), repo_path.parent / "plugins")
+        run_copilot_harness(create_dataset_entry(), "test-model", EvaluationCategory.BUG_FIX, repo_path, output_dir, AgentConfig(), repo_path.parent / "plugins")
 
     run.assert_not_called()
     gateway.stop.assert_called_once_with()
@@ -196,7 +214,7 @@ def test_missing_copilot_stops_gateway_without_invoking_a_process(configured_cop
 def test_successful_copilot_run_stops_gateway(configured_copilot_run):
     repo_path, output_dir, gateway, _ = configured_copilot_run
 
-    run_copilot_agent(create_dataset_entry(), "test-model", EvaluationCategory.BUG_FIX, repo_path, output_dir, AgentConfig(), repo_path.parent / "plugins")
+    run_copilot_harness(create_dataset_entry(), "test-model", EvaluationCategory.BUG_FIX, repo_path, output_dir, AgentConfig(), repo_path.parent / "plugins")
 
     gateway.stop.assert_called_once_with()
 
@@ -210,7 +228,7 @@ def test_copilot_sets_up_the_session_from_the_given_agent_config(configured_copi
         mocks["build_mcp_config"].return_value = (None, None)
         mocks["setup_custom_agent"].return_value = None
         mocks["resolve_config_plugins"].return_value = []
-        run_copilot_agent(create_dataset_entry(), "test-model", EvaluationCategory.BUG_FIX, repo_path, output_dir, agent_config, repo_path.parent / "plugins")
+        run_copilot_harness(create_dataset_entry(), "test-model", EvaluationCategory.BUG_FIX, repo_path, output_dir, agent_config, repo_path.parent / "plugins")
 
     for name in ("build_prompt", "build_mcp_config", "resolve_config_plugins"):
         assert agent_config in mocks[name].call_args.args, name
@@ -225,7 +243,7 @@ def test_copilot_sets_up_customizations_from_the_entry_profile_into_github(confi
 
     with patch.multiple("bcbench.agent.copilot.agent", **dict.fromkeys(helpers, DEFAULT)) as mocks:
         mocks["setup_custom_agent"].return_value = None
-        run_copilot_agent(create_dataset_entry(), "test-model", EvaluationCategory.BUG_FIX, repo_path, output_dir, AgentConfig(), repo_path.parent / "plugins")
+        run_copilot_harness(create_dataset_entry(), "test-model", EvaluationCategory.BUG_FIX, repo_path, output_dir, AgentConfig(), repo_path.parent / "plugins")
 
     profile_dir = instructions_profile_dir(create_dataset_entry().customization_profile)
     for name in helpers:
@@ -240,7 +258,7 @@ def test_copilot_keeps_plugins_in_the_given_plugin_root(configured_copilot_run):
     with patch.multiple("bcbench.agent.copilot.agent", build_al_lsp_plugin=DEFAULT, resolve_config_plugins=DEFAULT) as mocks:
         mocks["build_al_lsp_plugin"].return_value = None
         mocks["resolve_config_plugins"].return_value = []
-        run_copilot_agent(create_dataset_entry(), "test-model", EvaluationCategory.BUG_FIX, repo_path, output_dir, AgentConfig(), plugin_root)
+        run_copilot_harness(create_dataset_entry(), "test-model", EvaluationCategory.BUG_FIX, repo_path, output_dir, AgentConfig(), plugin_root)
 
     assert plugin_root in mocks["build_al_lsp_plugin"].call_args.args
     assert plugin_root in mocks["resolve_config_plugins"].call_args.args
diff --git a/tests/test_dataquery_evaluation.py b/tests/test_dataquery_evaluation.py
index 28783bb85..d2d2a89c7 100644
--- a/tests/test_dataquery_evaluation.py
+++ b/tests/test_dataquery_evaluation.py
@@ -2,6 +2,7 @@
 from pathlib import Path
 
 import pytest
+from bcbench_core.agent.harness import AgentIdentity
 from bcbench_core.container import ContainerConfig
 from bcbench_core.exceptions import BuildError
 
@@ -87,8 +88,7 @@ def _context(self, tmp_path: Path) -> EvaluationContext[DataQueryEntry]:
             repo_path=tmp_path / "repo",
             result_dir=tmp_path / "results",
             container=ContainerConfig("bcbench", "admin", "secret", company="CRONUS"),
-            model="test-model",
-            agent_name=AgentHarness.COPILOT,
+            agent=AgentIdentity[AgentHarness](harness=AgentHarness.COPILOT, version="1.0.0", model="test-model"),
             category=EvaluationCategory.DATA_QUERY,
         )
 
diff --git a/tests/test_evaluate_pipeline.py b/tests/test_evaluate_pipeline.py
index 17ea4e049..e3bd2bed5 100644
--- a/tests/test_evaluate_pipeline.py
+++ b/tests/test_evaluate_pipeline.py
@@ -7,6 +7,7 @@
 
 import pytest
 from bcbench_core.agent.experiment import ExperimentConfiguration
+from bcbench_core.agent.harness import AgentResult
 from bcbench_core.agent.metrics import AgentMetrics
 from bcbench_core.exceptions import AgentTimeoutError
 
@@ -39,7 +40,8 @@ def run_agent(self, context: EvaluationContext[E], agent_runner: AgentRunner[E])
         self.run_agent_called = True
         if self.raise_in_run_agent is not None:
             raise self.raise_in_run_agent
-        context.metrics, context.experiment = agent_runner(context)
+        result = agent_runner(context)
+        context.metrics, context.experiment = result.metrics, result.experiment
 
     @override
     def evaluate(self, context: EvaluationContext[E]) -> None:
@@ -48,8 +50,8 @@ def evaluate(self, context: EvaluationContext[E]) -> None:
             raise self.raise_in_evaluate
 
 
-def _noop_runner(_ctx: EvaluationContext[BugFixEntry]) -> tuple[AgentMetrics | None, ExperimentConfiguration | None]:
-    return AgentMetrics(execution_time=1.0), ExperimentConfiguration()
+def _noop_runner(_ctx: EvaluationContext[BugFixEntry]) -> AgentResult:
+    return AgentResult(metrics=AgentMetrics(execution_time=1.0), experiment=ExperimentConfiguration())
 
 
 def _read_only_result[E: BaseDatasetEntry](ctx: EvaluationContext[E]) -> BaseEvaluationResult:
@@ -93,7 +95,7 @@ def test_persists_category_specific_timeout_result(self, tmp_path):
         ctx = create_evaluation_context(tmp_path, entry=entry, category=EvaluationCategory.NL2AL)
         pipeline = _StubPipeline[NL2ALEntry](raise_in_run_agent=AgentTimeoutError("test timeout"))
 
-        pipeline.execute(ctx, lambda _: (None, None))
+        pipeline.execute(ctx, lambda _: AgentResult())
 
         result_file = ctx.result_dir / f"{entry.instance_id}{RESULT_FILE_SUFFIX}"
         payload = json.loads(result_file.read_text(encoding="utf-8"))
diff --git a/tests/test_nl2al_pipeline.py b/tests/test_nl2al_pipeline.py
index 06457d34e..ffc899dea 100644
--- a/tests/test_nl2al_pipeline.py
+++ b/tests/test_nl2al_pipeline.py
@@ -1,6 +1,7 @@
 """Tests for NL2ALPipeline.evaluate() — empty-diff handling is nl2al-specific."""
 
 import pytest
+from bcbench_core.agent.harness import AgentResult
 from bcbench_core.exceptions import EmptyDiffError
 
 from bcbench.evaluate.nl2al import NL2ALPipeline
@@ -85,7 +86,7 @@ def test_runs_agent_once_and_never_retries(self, tmp_path, monkeypatch):
 
         def agent_runner(_ctx):
             agent_calls["n"] += 1
-            return (None, None)
+            return AgentResult()
 
         reset_calls = {"n": 0}
         monkeypatch.setattr(pipeline, "setup_workspace", lambda *_args, **_kw: reset_calls.__setitem__("n", reset_calls["n"] + 1))
diff --git a/tests/test_pr_review_agent.py b/tests/test_pr_review_agent.py
index e294a47b4..11def48bc 100644
--- a/tests/test_pr_review_agent.py
+++ b/tests/test_pr_review_agent.py
@@ -163,7 +163,7 @@ def test_engine_environment_uses_target_repository_and_absolute_paths(tmp_path:
         patch("bcbench.agent.pr_review.agent.time.monotonic", side_effect=[10.0, 12.5]),
         patch("bcbench.agent.pr_review.agent.subprocess.run", return_value=completed) as run_process,
     ):
-        metrics, config = run_pr_review_agent(
+        result = run_pr_review_agent(
             entry=entry,
             model="gpt-5.6-luna",
             category=EvaluationCategory.CODE_REVIEW,
@@ -172,6 +172,7 @@ def test_engine_environment_uses_target_repository_and_absolute_paths(tmp_path:
             engine_path=tmp_path / "engine",
         )
 
+    metrics = result.metrics
     assert isinstance(metrics, PRReviewMetrics)
     assert metrics.execution_time == 2.5
     assert metrics.prompt_tokens == 100
@@ -180,7 +181,8 @@ def test_engine_environment_uses_target_repository_and_absolute_paths(tmp_path:
     assert metrics.ai_credits == 0.25
     assert metrics.api_calls == 2
     assert metrics.copilot_cli_version == "1.0.81-0"
-    assert config.is_empty()
+    assert result.experiment is not None
+    assert result.experiment.is_empty()
     resolve_engine.assert_called_once_with(tmp_path / "engine")
     prepare_bcquality.assert_called_once_with(tmp_path / "engine", "pwsh", (tmp_path / "output" / "bcquality").resolve())
     assert run_process.call_args.kwargs["encoding"] == "utf-8"
diff --git a/tests/test_review_runners.py b/tests/test_review_runners.py
index 9f2bba7f2..0d1ebe225 100644
--- a/tests/test_review_runners.py
+++ b/tests/test_review_runners.py
@@ -12,21 +12,27 @@
 from bcbench.types import AgentHarness, EvaluationCategory
 
 
-@pytest.mark.parametrize(
-    ("command", "runner_name"),
-    [
-        (run_commands.run_copilot, "run_copilot_agent"),
-        (run_commands.run_claude, "run_claude_code"),
-    ],
-)
-def test_generic_run_commands_accept_code_review(tmp_path: Path, command, runner_name: str) -> None:
+def test_copilot_run_command_accepts_code_review(tmp_path: Path) -> None:
+    entry = object()
+    with (
+        patch.object(CodeReviewEntry, "load", return_value=[entry]),
+        patch.object(CodeReviewPipeline, "setup_workspace"),
+        patch.object(run_commands, "CopilotHarness") as harness,
+    ):
+        run_commands.run_copilot("synthetic__style-018", EvaluationCategory.CODE_REVIEW, repo_path=tmp_path, output_dir=tmp_path / "out")
+
+    assert harness.call_args.kwargs["category"] is EvaluationCategory.CODE_REVIEW
+    assert harness.return_value.run.call_args.args[0].task is entry
+
+
+def test_claude_run_command_accepts_code_review(tmp_path: Path) -> None:
     entry = object()
     with (
         patch.object(CodeReviewEntry, "load", return_value=[entry]),
         patch.object(CodeReviewPipeline, "setup_workspace"),
-        patch.object(run_commands, runner_name) as agent_runner,
+        patch.object(run_commands, "run_claude_code") as agent_runner,
     ):
-        command("synthetic__style-018", EvaluationCategory.CODE_REVIEW, repo_path=tmp_path, output_dir=tmp_path / "out")
+        run_commands.run_claude("synthetic__style-018", EvaluationCategory.CODE_REVIEW, repo_path=tmp_path, output_dir=tmp_path / "out")
 
     assert agent_runner.call_args.kwargs["entry"] is entry
     assert agent_runner.call_args.kwargs["category"] is EvaluationCategory.CODE_REVIEW
@@ -44,7 +50,7 @@ def test_generic_evaluate_commands_use_code_review_pipeline(tmp_path: Path, comm
     with (
         patch.object(CodeReviewEntry, "load", return_value=[object()]),
         patch.object(CodeReviewPipeline, "execute", side_effect=lambda context, runner: contexts.append(context)),
-        patch.object(evaluate_commands, "get_copilot_version", return_value="1.2.3"),
+        patch("bcbench.agent.copilot.agent.get_copilot_version", return_value="1.2.3"),
         patch.object(evaluate_commands, "get_claude_version", return_value="1.2.3"),
     ):
         command(
@@ -56,8 +62,8 @@ def test_generic_evaluate_commands_use_code_review_pipeline(tmp_path: Path, comm
         )
 
     assert len(contexts) == 1
-    assert contexts[0].agent_name is agent_name
-    assert contexts[0].agent_version == "1.2.3"
+    assert contexts[0].agent.harness is agent_name
+    assert contexts[0].agent.version == "1.2.3"
     assert contexts[0].category is EvaluationCategory.CODE_REVIEW
 
 
@@ -88,10 +94,10 @@ def test_pr_review_evaluation_is_fixed_to_runner_and_category(tmp_path: Path) ->
 
     assert result.exit_code == 0, result.exception
     assert len(contexts) == 1
-    assert contexts[0].agent_name is AgentHarness.PR_REVIEW
-    assert contexts[0].agent_version == "a" * 40
+    assert contexts[0].agent.harness is AgentHarness.PR_REVIEW
+    assert contexts[0].agent.version == "a" * 40
     assert contexts[0].category is EvaluationCategory.CODE_REVIEW
-    assert contexts[0].model == "gpt-5.6-luna"
+    assert contexts[0].agent.model == "gpt-5.6-luna"
     assert agent_runner.call_args.kwargs["engine_path"] == tmp_path
     get_version.assert_called_once_with(tmp_path)