coder-eval 0.11.1__tar.gz → 0.11.2__tar.gz

{coder_eval-0.11.1 → coder_eval-0.11.2}/CHANGELOG.md

@@ -2,6 +2,25 @@
 
 <!-- version list -->
 
+## v0.11.2 (2026-08-24)
+
+### Bug Fixes
+
+- **docker**: Expand ~ and $VAR in extra_mounts destinations
+  ([#128](https://github.com/UiPath/coder_eval/pull/128),
+  [`02dbf69`](https://github.com/UiPath/coder_eval/commit/02dbf694705fd3f7de4ce164230a70945f08a1ce))
+
+- **docker**: Reject expansions that inject a ':' into a mount path
+  ([#128](https://github.com/UiPath/coder_eval/pull/128),
+  [`02dbf69`](https://github.com/UiPath/coder_eval/commit/02dbf694705fd3f7de4ce164230a70945f08a1ce))
+
+### Documentation
+
+- **docker**: Trim the extra_mounts destination comments to scope
+  ([#128](https://github.com/UiPath/coder_eval/pull/128),
+  [`02dbf69`](https://github.com/UiPath/coder_eval/commit/02dbf694705fd3f7de4ce164230a70945f08a1ce))
+
+
 ## v0.11.1 (2026-08-21)
 
 ### Bug Fixes

{coder_eval-0.11.1 → coder_eval-0.11.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: coder-eval
-Version: 0.11.1
+Version: 0.11.2
 Summary: Evaluate, benchmark, and A/B-test AI coding agents (Claude Code, Codex, Gemini/Antigravity) with sandboxed, reproducible YAML task suites.
 Project-URL: Homepage, https://coder-eval.com
 Project-URL: Repository, https://github.com/UiPath/coder_eval

{coder_eval-0.11.1 → coder_eval-0.11.2}/action.yml

@@ -41,7 +41,7 @@ inputs:
   version:
     description: coder-eval version to install from PyPI, or "local" to install from the action checkout
     required: false
-    default: "0.11.1"   # <-- kept in sync with releases by release.yml
+    default: "0.11.2"   # <-- kept in sync with releases by release.yml
   run-dir:
     description: Run directory (--run-dir)
     required: false

{coder_eval-0.11.1 → coder_eval-0.11.2}/plugins/coder-eval/.claude-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "coder-eval",
-  "version": "0.11.1",
+  "version": "0.11.2",
   "description": "Author, run, and analyze coder-eval suites — including whether your Claude Code skills actually trigger.",
   "author": { "name": "UiPath", "url": "https://github.com/UiPath/coder_eval" },
   "homepage": "https://coder-eval.com",

{coder_eval-0.11.1 → coder_eval-0.11.2}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "coder-eval"
-version = "0.11.1"
+version = "0.11.2"
 description = "Evaluate, benchmark, and A/B-test AI coding agents (Claude Code, Codex, Gemini/Antigravity) with sandboxed, reproducible YAML task suites."
 readme = "README.md"
 license = "Apache-2.0"

{coder_eval-0.11.1 → coder_eval-0.11.2}/src/coder_eval/__init__.py

@@ -1,3 +1,3 @@
 """coder_eval - A framework for evaluating AI coding agents."""
 
-__version__ = "0.11.1"
+__version__ = "0.11.2"

{coder_eval-0.11.1 → coder_eval-0.11.2}/src/coder_eval/isolation/docker_runner.py

@@ -290,11 +290,14 @@ def _validate_extra_mount(spec: str) -> str:
 
     Defends against typos that would silently expose the host fs to the
     container, and against mount specs that shadow framework-owned mounts.
-    Normalizes the source side by expanding ``~`` and ``$VAR`` so authors
-    can write portable specs. Returns the (possibly rewritten) spec to
-    feed back into argv.
+    Normalizes BOTH sides by expanding ``~`` and ``$VAR`` so authors can
+    write portable specs. Returns the (possibly rewritten) spec to feed
+    back into argv.
 
     Notes:
+      - Destinations are expanded too. A container path that has to match a
+        host-valued var (``$SKILLS_REPO_PATH``) would otherwise have to be
+        hardcoded per machine.
       - Mode is REQUIRED. Forgetting ``:ro`` is the single most common way
         to accidentally hand the container RW access to a host directory,
         so we make the author write it explicitly.
@@ -312,26 +315,35 @@ def _validate_extra_mount(spec: str) -> str:
     parts = body.split(":")
     if len(parts) < 2 or len(parts) > 3:
         raise ValueError(f"Invalid extra_mounts entry {spec!r}: expected `src:dst[:ro|rw]`.")
-    src, dst = head + parts[0], parts[1]
+    src, raw_dst = head + parts[0], parts[1]
     # Default to read-only when mode is omitted. Mounting host paths RW
     # by default is the wrong sandbox stance: the few RW use-cases are
     # better stated explicitly than implied by silence.
     mode = parts[2] if len(parts) == 3 else "ro"
     if not src:
         raise ValueError(f"Invalid extra_mounts entry {spec!r}: empty source path.")
-    if not dst:
+    if not raw_dst:
         raise ValueError(f"Invalid extra_mounts entry {spec!r}: empty destination path.")
+    # Expanded before the absolute-path check: that is the point.
+    expanded_src = os.path.expandvars(os.path.expanduser(src))
+    dst = os.path.expandvars(os.path.expanduser(raw_dst))
+    # A variable whose value carries a ':' would add fields to the spec rebuilt
+    # at the bottom, silently moving the destination or widening the mode.
+    # The drive prefix is excluded: its colon is legitimate and already split off.
+    if ":" in dst or ":" in expanded_src[len(head) :]:
+        raise ValueError(f"Invalid extra_mounts entry {spec!r}: expansion introduced a ':' into a path.")
     if not dst.startswith("/"):
-        raise ValueError(f"Invalid extra_mounts entry {spec!r}: destination must be an absolute path.")
+        # expandvars leaves an unset var verbatim, so typos land here.
+        detail = f"{raw_dst!r}" if dst == raw_dst else f"{raw_dst!r} (expanded to {dst!r})"
+        raise ValueError(f"Invalid extra_mounts entry {spec!r}: destination must be an absolute path, got {detail}.")
     if mode not in ("ro", "rw"):
         raise ValueError(f"Invalid extra_mounts entry {spec!r}: mode must be 'ro' or 'rw'.")
-    # Expand ~ and $VAR in the source so authors can write portable specs.
-    expanded_src = os.path.expandvars(os.path.expanduser(src))
     if not Path(expanded_src).exists():
         raise ValueError(f"Invalid extra_mounts entry {spec!r}: source path does not exist on host.")
     # Reject destinations that shadow framework-owned mounts inside the
     # container. ``/work`` substrings are caught too -- /work/foo would
     # land underneath our staging dir and shadow the input/output tree.
+    # Expanded form: a var could itself expand to a reserved path.
     dst_norm = dst.rstrip("/") or "/"
     if dst_norm in _RESERVED_MOUNT_DESTS or dst_norm.startswith(CONTAINER_WORK_DIR + "/"):
         raise ValueError(

{coder_eval-0.11.1 → coder_eval-0.11.2}/tests/test_docker_runner_mounts.py

@@ -108,6 +108,41 @@ class TestValidateExtraMount:
         result = _validate_extra_mount("$MYDIR:/mnt/x")
         assert result.startswith(real_dir + ":")
 
+    def test_var_expansion_in_destination(self, real_dir, monkeypatch):
+        """``$VAR`` in the destination expands too."""
+        monkeypatch.setenv("HOME", "/home/someuser")
+        result = _validate_extra_mount(f"{real_dir}:$HOME/.uipath:rw")
+        assert result == f"{real_dir}:/home/someuser/.uipath:rw"
+
+    def test_home_expansion_in_destination(self, real_dir, monkeypatch):
+        """``~`` in the destination expands the same way the source does."""
+        monkeypatch.setenv("HOME", "/home/someuser")
+        result = _validate_extra_mount(f"{real_dir}:~/.uipath:ro")
+        assert result == f"{real_dir}:/home/someuser/.uipath:ro"
+
+    def test_unset_var_destination_rejected_with_both_forms(self, real_dir):
+        """An unset var is left verbatim, so it must fail loudly."""
+        with pytest.raises(ValueError, match="destination must be an absolute path"):
+            _validate_extra_mount(f"{real_dir}:$NO_SUCH_VAR_HERE/x:ro")
+
+    def test_destination_var_expanding_to_reserved_is_rejected(self, real_dir, monkeypatch):
+        """The shadow check runs on the expanded destination."""
+        monkeypatch.setenv("SNEAKY", "/work")
+        with pytest.raises(ValueError, match="shadows a framework-owned mount"):
+            _validate_extra_mount(f"{real_dir}:$SNEAKY:ro")
+
+    def test_destination_var_carrying_a_colon_is_rejected(self, real_dir, monkeypatch):
+        """A ':' in an expanded value would add fields to the rebuilt spec."""
+        monkeypatch.setenv("SNEAKY", "/mnt/x:rw")
+        with pytest.raises(ValueError, match="introduced a ':'"):
+            _validate_extra_mount(f"{real_dir}:$SNEAKY:ro")
+
+    def test_source_var_carrying_a_colon_is_rejected(self, monkeypatch):
+        """Same guard on the source side, which is rebuilt the same way."""
+        monkeypatch.setenv("SNEAKY", "/mnt/x:rw")
+        with pytest.raises(ValueError, match="introduced a ':'"):
+            _validate_extra_mount("$SNEAKY:/mnt/y:ro")
+
     def test_malformed_no_colon(self):
         with pytest.raises(ValueError, match="expected `src:dst"):
             _validate_extra_mount("just-one-token")

{coder_eval-0.11.1 → coder_eval-0.11.2}/uv.lock

@@ -448,7 +448,7 @@ wheels = [
 
 [[package]]
 name = "coder-eval"
-version = "0.11.1"
+version = "0.11.2"
 source = { editable = "." }
 dependencies = [
     { name = "anthropic" },