feat: stage per-trial browser challenge workspaces

examples/agent_challenges/browser_click_challenge/README.md

@@ -12,15 +12,17 @@ examples/browser_click_workflow/
 ## Default Behavior
 
 By default the harness does not start a `wf-rpc-server`. It prompts agents to
-use the configured local CLI path:
+use a per-trial configured local CLI path:
 
 ```powershell
-uv run wf --config examples/browser_click_workflow/wf.config.json --local
+uv run wf --config examples/agent_challenges/browser_click_challenge/workspaces/<trial>/wf.config.json --local
 ```
 
+For each local-mode trial, the harness copies `workspace_template/` into
+`workspaces/<model>-trial-<n>/` and injects that config path into the prompt.
 This builds the configured workflow server in the CLI process for each command
-and reuses the configured durable store. It does not reuse in-memory source
+and uses the copied workspace's durable store. It does not reuse in-memory
-sessions across CLI invocations.
+source sessions across CLI invocations.
 
 Use `--start-server` when the trial should exercise the JSON-RPC server path.
 With `--start-server`, the harness starts:
@@ -60,7 +62,8 @@ examples/agent_challenges/browser_click_challenge/workspace_template/
 
 It contains a local workflow config and prompt template that point at the
 browser-click Python source without exposing a generated draft patch answer
-file. Use it as the starting context for an agent trial, or copy it under:
+file. The harness copies it automatically for normal local-mode trials. For
+manual experiments, copy it under:
 
 ```text
 examples/agent_challenges/browser_click_challenge/workspaces/

examples/agent_challenges/browser_click_challenge/run_opencode_trials.py

@@ -2,6 +2,7 @@ from __future__ import annotations
 
 import argparse
 import json
+import shutil
 import subprocess
 import sys
 import time
@@ -25,6 +26,8 @@ ROOT = Path(__file__).resolve().parents[3]
 CHALLENGE_DIR = Path(__file__).resolve().parent
 DEFAULT_PROMPT = CHALLENGE_DIR / "prompt.md"
 DEFAULT_RESULTS_DIR = CHALLENGE_DIR / "results"
+DEFAULT_WORKSPACES_DIR = CHALLENGE_DIR / "workspaces"
+DEFAULT_WORKSPACE_TEMPLATE = CHALLENGE_DIR / "workspace_template"
 DEFAULT_SERVER_PORT = 8772
 EXAMPLE_CONFIG = ROOT / "examples" / "browser_click_workflow" / "wf.config.json"
 EXAMPLE_CONFIG_ARG = "examples/browser_click_workflow/wf.config.json"
@@ -68,6 +71,15 @@ class ManagedServer:
     rpc_url: str
 
 
+@dataclass(frozen=True, slots=True)
+class TrialWorkspace:
+    """Per-trial scratch area copied from the challenge workspace template."""
+
+    root: Path
+    config_path: Path
+    prompt_path: Path
+
+
 def start_server(*, port: int, timeout_seconds: int = 30) -> ManagedServer:
     command = server_command(port=port)
     process = subprocess.Popen(
@@ -105,6 +117,42 @@ def wait_for_status(*, rpc_url: str, timeout_seconds: int) -> None:
         raise RuntimeError(f"wf status did not become ready: {last_stderr}")
 
 
+def _safe_model_name(model: str) -> str:
+    return model.replace("/", "_").replace(":", "_")
+
+
+def prepare_trial_workspace(
+    *,
+    model: str,
+    index: int,
+    workspaces_dir: Path = DEFAULT_WORKSPACES_DIR,
+    template_dir: Path = DEFAULT_WORKSPACE_TEMPLATE,
+) -> TrialWorkspace:
+    """Copy the authoring template into a clean ignored per-trial directory."""
+    root = workspaces_dir / f"{_safe_model_name(model)}-trial-{index:03d}"
+    if root.exists():
+        # Stale scratch files can leak answers between trials, so reset only
+        # the ignored per-trial directory before copying the template.
+        shutil.rmtree(root)
+    shutil.copytree(template_dir, root)
+    return TrialWorkspace(
+        root=root,
+        config_path=root / "wf.config.json",
+        prompt_path=root / "prompt.md",
+    )
+
+
+def wf_command_prefix_for_config(config_path: Path) -> str:
+    path = config_path
+    if not path.is_absolute():
+        path = ROOT / path
+    try:
+        path_arg = path.resolve().relative_to(ROOT.resolve()).as_posix()
+    except ValueError:
+        path_arg = str(path.resolve())
+    return f"uv run wf --config {path_arg} --local"
+
+
 def stop_server(process: subprocess.Popen[str]) -> None:
     if process.poll() is not None:
         return
@@ -334,8 +382,7 @@ def _contains_bool_marker(text: str, marker: str, value: str) -> bool:
 
 
 def trial_output_path(results_dir: Path, *, model: str, index: int) -> Path:
-    safe_model = model.replace("/", "_").replace(":", "_")
+    return results_dir / f"{_safe_model_name(model)}-trial-{index:03d}.json"
-    return results_dir / f"{safe_model}-trial-{index:03d}.json"
 
 
 def run_trial(config: TrialConfig, *, index: int, results_dir: Path) -> dict[str, Any]:
@@ -433,6 +480,7 @@ def main(argv: list[str] | None = None) -> int:
     )
     parser.add_argument("--prompt", type=Path, default=DEFAULT_PROMPT)
     parser.add_argument("--results-dir", type=Path, default=DEFAULT_RESULTS_DIR)
+    parser.add_argument("--workspaces-dir", type=Path, default=DEFAULT_WORKSPACES_DIR)
     parser.add_argument("--server-url", default=None)
     parser.add_argument("--start-server", action="store_true", default=False)
     parser.add_argument("--no-start-server", action="store_false", dest="start_server")
@@ -464,18 +512,37 @@ def main(argv: list[str] | None = None) -> int:
         )
 
     try:
+        use_trial_workspace = args.server_url is None and not args.start_server
+        summaries: list[dict[str, Any]] = []
+        for index in range(1, args.trials + 1):
+            prompt_path = args.prompt
+            trial_wf_command_prefix = wf_command_prefix
+            trial_server_context = server_context
+            if use_trial_workspace:
+                workspace = prepare_trial_workspace(
+                    model=args.model,
+                    index=index,
+                    workspaces_dir=args.workspaces_dir,
+                )
+                if args.prompt == DEFAULT_PROMPT:
+                    prompt_path = workspace.prompt_path
+                trial_wf_command_prefix = wf_command_prefix_for_config(
+                    workspace.config_path
+                )
+                trial_server_context = (
+                    "No external workflow RPC server is staged. Use the "
+                    "per-trial workspace config copied to "
+                    f"`{workspace.config_path.relative_to(ROOT).as_posix()}`."
+                )
             config = TrialConfig(
                 model=args.model,
                 variant=args.variant,
-            prompt_path=args.prompt,
+                prompt_path=prompt_path,
                 attach_url=args.attach_url,
                 timeout_seconds=args.timeout_seconds,
-            wf_command_prefix=wf_command_prefix,
+                wf_command_prefix=trial_wf_command_prefix,
-            server_context=server_context,
+                server_context=trial_server_context,
             )
-
-        summaries: list[dict[str, Any]] = []
-        for index in range(1, args.trials + 1):
             result = run_trial(config, index=index, results_dir=args.results_dir)
             summaries.append(
                 {

examples/agent_challenges/browser_click_challenge/workspace_template/prompt.md

@@ -10,9 +10,11 @@ Build and successfully run a workflow that:
 Use this command prefix for product-facing operations:
 
 ```powershell
-uv run wf --config examples/agent_challenges/browser_click_challenge/workspace_template/wf.config.json --local
+{{wf_command_prefix}}
 ```
 
+{{server_context}}
+
 Start from `skills/`, especially the workflow/CLI skill references, before doing
 broad docs or code search. Do not read outside this repository. Do not solve the
 challenge with only a standalone Playwright/Python script.
@@ -70,4 +72,3 @@ the `wf` CLI, either in local same-process mode or through `wf-rpc-server`. Set
 `WorkflowApi` directly.
 
 If something fails, report the exact command and error instead of hiding it.
-

tests/examples/test_opencode_browser_click_challenge.py

@@ -11,9 +11,11 @@ from examples.agent_challenges.browser_click_challenge.run_opencode_trials impor
     classify_output,
     extract_challenge_report,
     parse_opencode_output,
+    prepare_trial_workspace,
     render_prompt,
     server_command,
     trial_output_path,
+    wf_command_prefix_for_config,
 )
 
 
@@ -247,6 +249,71 @@ def test_trial_output_path_is_zero_padded(tmp_path: Path) -> None:
     assert path.name == "opencode_mimo-v2.5-free-trial-003.json"
 
 
+def test_prepare_trial_workspace_copies_template_to_model_trial_dir(
+    tmp_path: Path,
+) -> None:
+    template = tmp_path / "template"
+    template.mkdir()
+    (template / "wf.config.json").write_text('{"version": 1}', encoding="utf-8")
+    (template / "prompt.md").write_text("prompt", encoding="utf-8")
+    (template / ".gitignore").write_text(".wf_store/\n", encoding="utf-8")
+    workspaces = tmp_path / "workspaces"
+
+    prepared = prepare_trial_workspace(
+        model="opencode/mimo-v2.5-free",
+        index=7,
+        workspaces_dir=workspaces,
+        template_dir=template,
+    )
+
+    assert prepared.root == workspaces / "opencode_mimo-v2.5-free-trial-007"
+    assert prepared.config_path.read_text(encoding="utf-8") == '{"version": 1}'
+    assert prepared.prompt_path.read_text(encoding="utf-8") == "prompt"
+    assert (prepared.root / ".gitignore").read_text(encoding="utf-8") == ".wf_store/\n"
+
+
+def test_prepare_trial_workspace_removes_stale_previous_attempt(
+    tmp_path: Path,
+) -> None:
+    template = tmp_path / "template"
+    template.mkdir()
+    (template / "wf.config.json").write_text('{"version": 1}', encoding="utf-8")
+    (template / "prompt.md").write_text("prompt", encoding="utf-8")
+    workspaces = tmp_path / "workspaces"
+    stale = workspaces / "opencode_mimo-v2.5-free-trial-001" / "old-answer.json"
+    stale.parent.mkdir(parents=True)
+    stale.write_text("stale", encoding="utf-8")
+
+    prepared = prepare_trial_workspace(
+        model="opencode/mimo-v2.5-free",
+        index=1,
+        workspaces_dir=workspaces,
+        template_dir=template,
+    )
+
+    assert prepared.root.exists()
+    assert not stale.exists()
+
+
+def test_wf_command_prefix_for_config_uses_repo_relative_path() -> None:
+    config_path = (
+        Path("examples")
+        / "agent_challenges"
+        / "browser_click_challenge"
+        / "workspaces"
+        / "opencode_mimo-v2.5-free-trial-001"
+        / "wf.config.json"
+    )
+
+    prefix = wf_command_prefix_for_config(config_path)
+
+    assert prefix == (
+        "uv run wf --config "
+        "examples/agent_challenges/browser_click_challenge/workspaces/"
+        "opencode_mimo-v2.5-free-trial-001/wf.config.json --local"
+    )
+
+
 def test_render_prompt_injects_command_prefix_and_server_context(
     tmp_path: Path,
 ) -> None: