Merge pull request #26 from spacedock-dev/fix/swe-bench-pro-default-run

fix(swe-bench-pro): make the default `rk run` work out of the box

Author: kentwelcome

examples/specs/swe-bench-pro-spacedock-codex.yaml

@@ -14,9 +14,12 @@ agent:
     top_p: null
     seed: null
   solver_workflow: ./examples/solver_workflows/codex-benchmark-solver
-  # SWE-tuned budget — above the 1200s codex default: large repos + long
-  # test suites need more turns and a longer per-attempt/overall timeout.
-  max_turns: 400
+  # SWE-tuned budget. The codex runtime does NOT honor a custom max_turns —
+  # it accepts only the default (200); any other value raises
+  # SpacedockSolverAgentError. So keep max_turns at 200 and buy the
+  # "large repo + long test suite" headroom via the TIMEOUTS below, which
+  # govern wall-clock (harbor trial config), not codex turn limits.
+  max_turns: 200
   override_timeout_sec: 5400
   max_timeout_sec: 7200
   reasoning_effort: xhigh

src/razorback/agents/_runtime/codex.py

@@ -209,9 +209,16 @@ def _codex_kwargs(harbor_agent_kwargs: dict[str, Any]) -> dict[str, Any]:
         if _is_empty_noop(name, value):
             continue
         if name not in _CODEX_SUPPORTED_KWARGS:
+            hint = ""
+            if name == "max_turns":
+                hint = (
+                    " The codex runtime accepts only the default max_turns (200); "
+                    "keep it at 200 and budget wall-clock via "
+                    "override_timeout_sec / max_timeout_sec instead."
+                )
             raise SpacedockSolverAgentError(
                 "codex runtime adapter cannot honor unsupported harbor_agent_kwargs "
-                f"field {name!r}; refusing to silently drop it."
+                f"field {name!r}; refusing to silently drop it.{hint}"
             )
         kw[name] = value
     return kw

src/razorback/harbor_tasks/materialize.py

@@ -113,12 +113,41 @@ def _reflect_allowed_files(
             target.mkdir(parents=True, exist_ok=True)
             continue
         target.parent.mkdir(parents=True, exist_ok=True)
-        if view_mode == "copy":
+        # The `environment/` subtree is the Docker build context (current Harbor
+        # convention): `docker compose build` runs with it as the project
+        # directory, and BuildKit cannot read a Dockerfile (or any build input)
+        # that symlinks outside the context ("failed to read dockerfile: no such
+        # file or directory"). So always materialize the build context as real
+        # files, even in link mode — mirroring how `_patch_task_toml` keeps
+        # task.toml view-owned. Bulk task files still symlink in link mode (the
+        # whole point of bind: no eager duplication).
+        copy_real = view_mode == "copy" or _is_build_context_path(rel)
+        if path.is_symlink():
+            # The name filter above only saw the LINK's own path. A symlink with
+            # an innocuous name can still point at a denied answer artifact (e.g.
+            # `environment/leak.patch -> ../gold.patch`); copying follows the link
+            # and embeds the target's bytes under an allowed view path, and a view
+            # symlink chases through to it too — bypassing the leakage deny
+            # boundary. Resolve the target and re-apply the deny check (and
+            # source containment) so a symlink cannot smuggle denied or
+            # out-of-tree content into the view.
+            try:
+                resolved_rel = path.resolve().relative_to(source).as_posix()
+            except ValueError:
+                continue  # target escapes the source tree — refuse it
+            if matches_denied_path(resolved_rel, exclude_globs):
+                continue
+        if copy_real:
             shutil.copy2(path, target)
         else:
             os.symlink(path, target)
 
 
+def _is_build_context_path(rel: Path) -> bool:
+    """True for files under the `environment/` Docker build context."""
+    return rel.parts[:1] == ("environment",)
+
+
 def _patch_task_toml(
     task_toml: Path,
     *,

tests/unit/test_harbor_task_view_materializer.py

@@ -109,6 +109,65 @@ def test_link_mode_symlinks_files_but_never_mutates_source_task_toml(tmp_path):
     assert "RAZORBACK_BENCHMARK_TASK_ID" not in (source / "task.toml").read_text()
 
 
+def test_link_mode_copies_environment_build_context_as_real_files(tmp_path):
+    """`view_mode="link"` must keep the `environment/` Docker build context as
+    REAL files, not symlinks. `docker compose build` runs with the view's
+    `environment/` dir as the build context; BuildKit cannot read a Dockerfile
+    that symlinks outside the context (`failed to read dockerfile: no such file
+    or directory`), so a symlinked build context breaks every build-from-source
+    benchmark (e.g. swe-bench-pro) under the default bind/link mode. Bulk task
+    files still symlink — only the build context is forced real.
+    """
+    source = _write_source_task(tmp_path)
+    dockerfile_text = (source / "environment" / "Dockerfile").read_text()
+
+    view = materialize_harbor_task_view(
+        source_task_dir=source,
+        view_root=tmp_path / "views",
+        benchmark_kind="fixture-bench",
+        benchmark_task_id="task-001",
+        transform_name="fixture-transform",
+        view_mode="link",
+    )
+
+    # the Docker build context is a real, view-owned file (readable by BuildKit)
+    assert (view / "environment" / "Dockerfile").is_file()
+    assert not (view / "environment" / "Dockerfile").is_symlink()
+    assert (view / "environment" / "Dockerfile").read_text() == dockerfile_text
+    # bulk files outside environment/ still symlink — the bind/link contract holds
+    assert (view / "instruction.md").is_symlink()
+    assert (view / "data" / "input.csv").is_symlink()
+
+
+def test_environment_symlink_to_denied_target_is_not_smuggled(tmp_path):
+    """A symlink under `environment/` with an innocuous name must not smuggle a
+    DENIED target's bytes into the view. Copying the build context follows
+    symlinks (shutil.copy2), and the name-based deny filter only sees the link's
+    own path — so `environment/leak.patch -> ../gold_patch.diff` would otherwise
+    embed gold-patch content under an allowed view path. The materializer must
+    resolve the target and re-apply the deny check, dropping it.
+    """
+    source = _write_source_task(tmp_path)
+    (source / "gold_patch.diff").write_text("--- GOLD ANSWER PATCH ---\n")
+    (source / "environment" / "leak.patch").symlink_to(source / "gold_patch.diff")
+
+    view = materialize_harbor_task_view(
+        source_task_dir=source,
+        view_root=tmp_path / "views",
+        benchmark_kind="fixture-bench",
+        benchmark_task_id="task-001",
+        transform_name="fixture-transform",
+        exclude_globs=("gold_patch*", "gold.patch"),
+        view_mode="link",
+    )
+
+    # the legit build context is still materialized as a real file …
+    assert (view / "environment" / "Dockerfile").is_file()
+    assert not (view / "environment" / "Dockerfile").is_symlink()
+    # … but the disguised symlink to a denied target is dropped, not embedded
+    assert not (view / "environment" / "leak.patch").exists()
+
+
 def test_materialized_view_is_harbor_taskconfig_path_ready(tmp_path):
     source = _write_source_task(tmp_path)
 

tests/unit/test_runtime_adapters.py

@@ -501,6 +501,23 @@ def test_codex_rejects_unsupported_contract_kwargs(tmp_path, kwarg):
         )
 
 
+def test_codex_max_turns_rejection_is_actionable(tmp_path):
+    """A custom max_turns is the most common codex trip (claude honors it, codex
+    does not). The error must tell the user the fix: keep the default (200) and
+    budget wall-clock via timeouts — not leave them to reverse-engineer it.
+    """
+    with pytest.raises(SpacedockSolverAgentError) as excinfo:
+        codex_adapter.build_inner_agent(
+            logs_dir=tmp_path,
+            model="gpt-5.1-codex",
+            harbor_agent_kwargs={"max_turns": 400},
+            extra_env={"OPENAI_API_KEY": "sk-fake"},
+        )
+    message = str(excinfo.value)
+    assert "200" in message
+    assert "timeout" in message.lower()
+
+
 def test_pi_raises_not_implemented(tmp_path):
     with pytest.raises(NotImplementedError, match="pi"):
         pi_adapter.build_inner_agent(