chore: drop dead context compat, retire plan boxes, fix spec wording

docs/historical/superpowers/plans/2026-09-04-structured-runtime-context.md

@@ -20,7 +20,7 @@ sugar.
 JSON Schema, pytest, pytest-asyncio, Ruff, basedpyright, markdownlint-cli2.
 
 **Spec:**
-[`docs/superpowers/specs/2026-09-04-structured-runtime-context-design.md`](../specs/2026-09-04-structured-runtime-context-design.md)
+[`docs/superpowers/specs/2026-09-04-structured-runtime-context-design.md`](../../superpowers/specs/2026-09-04-structured-runtime-context-design.md)
 
 ## Global Constraints
 
@@ -100,7 +100,7 @@ JSON Schema, pytest, pytest-asyncio, Ruff, basedpyright, markdownlint-cli2.
 - `FrameContextView.graph` exposes a JSON-compatible `foreach` mapping, all
   unique active aliases, and innermost `loop_item`/`loop_index` values.
 
-- [ ] **Step 1: Write failing model and ancestry tests**
+- [x] **Step 1: Write failing model and ancestry tests**
 
   Add focused helpers that construct a `RunState` with explicit frames, then
   add tests named:
@@ -133,7 +133,7 @@ JSON Schema, pytest, pytest-asyncio, Ruff, basedpyright, markdownlint-cli2.
   For the scope-boundary case, give the child root a scheduling parent in the
   caller's foreach frame and assert that the child context remains `{}`.
 
-- [ ] **Step 2: Run the focused tests and confirm the missing API fails**
+- [x] **Step 2: Run the focused tests and confirm the missing API fails**
 
   Run:
 
@@ -144,7 +144,7 @@ JSON Schema, pytest, pytest-asyncio, Ruff, basedpyright, markdownlint-cli2.
   Expected: collection or assertions fail because `ForeachContext` and the
   ancestry-aware helper do not exist.
 
-- [ ] **Step 3: Add the typed context value and validate frame metadata once**
+- [x] **Step 3: Add the typed context value and validate frame metadata once**
 
   Add `ForeachContext` beside `RuntimeContext` and export it through
   `wf_core.__init__`. Keep `ForeachIterationMetadata` as the typed decoder for
@@ -164,7 +164,7 @@ JSON Schema, pytest, pytest-asyncio, Ruff, basedpyright, markdownlint-cli2.
       )
   ```
 
-- [ ] **Step 4: Implement fail-closed same-scope ancestry traversal**
+- [x] **Step 4: Implement fail-closed same-scope ancestry traversal**
 
   Walk from the selected frame through `parent_frame_id` while scope ids match.
   Validate the full chain before returning materialized values:
@@ -226,7 +226,7 @@ JSON Schema, pytest, pytest-asyncio, Ruff, basedpyright, markdownlint-cli2.
       graph[LOOP_INDEX_CONTEXT_KEY] = innermost.index
   ```
 
-- [ ] **Step 5: Add corruption and collision regressions**
+- [x] **Step 5: Add corruption and collision regressions**
 
   Add:
 
@@ -241,7 +241,7 @@ JSON Schema, pytest, pytest-asyncio, Ruff, basedpyright, markdownlint-cli2.
   `run.to_dict()` before the read-only test and assert it remains equal after
   deriving context.
 
-- [ ] **Step 6: Run and commit the focused model slice**
+- [x] **Step 6: Run and commit the focused model slice**
 
   Run:
 
@@ -285,7 +285,7 @@ JSON Schema, pytest, pytest-asyncio, Ruff, basedpyright, markdownlint-cli2.
 - `RuntimeContext.metadata` remains a defensive copy of the selected frame's
   metadata; `RuntimeContext.foreach` is the canonical typed view.
 
-- [ ] **Step 1: Write failing end-to-end runtime tests**
+- [x] **Step 1: Write failing end-to-end runtime tests**
 
   Extend `test_structured_runtime_context.py` with:
 
@@ -317,7 +317,7 @@ JSON Schema, pytest, pytest-asyncio, Ruff, basedpyright, markdownlint-cli2.
   Build the binding test with canonical `InputPathBinding` values for
   `context.foreach.customers.item` and `context.foreach.orders.item`.
 
-- [ ] **Step 2: Run the end-to-end tests and observe innermost-only behavior**
+- [x] **Step 2: Run the end-to-end tests and observe innermost-only behavior**
 
   Run:
 
@@ -329,7 +329,7 @@ JSON Schema, pytest, pytest-asyncio, Ruff, basedpyright, markdownlint-cli2.
   Expected: the new end-to-end tests fail because runtime consumers still use
   innermost frame-only context and handlers do not receive `.foreach`.
 
-- [ ] **Step 3: Update all graph-visible context call sites together**
+- [x] **Step 3: Update all graph-visible context call sites together**
 
   Replace every old call and verify with search:
 
@@ -349,7 +349,7 @@ JSON Schema, pytest, pytest-asyncio, Ruff, basedpyright, markdownlint-cli2.
   Root workflow output receives `frame_context_view(run, root_frame).graph`
   rather than an omitted context so standard root facts remain consistent.
 
-- [ ] **Step 4: Give Python handlers the same typed projection**
+- [x] **Step 4: Give Python handlers the same typed projection**
 
   In `_resolve_node_execution`, materialize the view once and pass its two
   projections to their consumers:
@@ -374,14 +374,14 @@ JSON Schema, pytest, pytest-asyncio, Ruff, basedpyright, markdownlint-cli2.
   Resolve graph bindings against `context_view.graph`. Do not reconstruct
   `ForeachContext` from the graph-visible dictionary.
 
-- [ ] **Step 5: Update old focused tests to pass `RunState` explicitly**
+- [x] **Step 5: Update old focused tests to pass `RunState` explicitly**
 
   Replace direct frame-only context calls in scheduler/context tests with a run
   containing that frame. Keep assertions for existing standard and
   compatibility keys; add structured assertions rather than deleting old
   coverage.
 
-- [ ] **Step 6: Run and commit the runtime integration slice**
+- [x] **Step 6: Run and commit the runtime integration slice**
 
   Run:
 
@@ -433,7 +433,7 @@ JSON Schema, pytest, pytest-asyncio, Ruff, basedpyright, markdownlint-cli2.
   `GraphSourcePath.context(self.id)` because that helper parses dots as path
   separators.
 
-- [ ] **Step 1: Write failing path and serialization tests**
+- [x] **Step 1: Write failing path and serialization tests**
 
   Add:
 
@@ -455,7 +455,7 @@ JSON Schema, pytest, pytest-asyncio, Ruff, basedpyright, markdownlint-cli2.
   assert "index" not in each.model_dump(mode="json")
   ```
 
-- [ ] **Step 2: Write failing node and subgraph binding tests**
+- [x] **Step 2: Write failing node and subgraph binding tests**
 
   Use the computed ref directly in both authoring boundaries:
 
@@ -473,12 +473,12 @@ JSON Schema, pytest, pytest-asyncio, Ruff, basedpyright, markdownlint-cli2.
   Assert both compiled bindings serialize their path as
   `context.foreach.orders.item`.
 
-- [ ] **Step 3: Implement only the two computed properties**
+- [x] **Step 3: Implement only the two computed properties**
 
   Add the properties directly to `ForeachNode`. Do not create `ForeachRef`, a
   node-address type, or field-selection sugar beneath `.item`.
 
-- [ ] **Step 4: Run and commit the authoring slice**
+- [x] **Step 4: Run and commit the authoring slice**
 
   Run:
 
@@ -596,7 +596,7 @@ JSON Schema, pytest, pytest-asyncio, Ruff, basedpyright, markdownlint-cli2.
   Keep entry schemas inline unless a measured schema-size problem requires
   `$defs`; the observable field types and required paths are contractual.
 
-- [ ] **Step 1: Replace innermost-only tests with full-stack expectations**
+- [x] **Step 1: Replace innermost-only tests with full-stack expectations**
 
   Update the existing nested test instead of adding contradictory coverage:
 
@@ -617,7 +617,7 @@ JSON Schema, pytest, pytest-asyncio, Ruff, basedpyright, markdownlint-cli2.
   Add `test_inner_completion_schema_restores_outer_structured_entry` and assert
   `after_inner` contains only the outer structured entry.
 
-- [ ] **Step 2: Run static analysis tests and confirm the old projection fails**
+- [x] **Step 2: Run static analysis tests and confirm the old projection fails**
 
   Run:
 
@@ -628,7 +628,7 @@ JSON Schema, pytest, pytest-asyncio, Ruff, basedpyright, markdownlint-cli2.
   Expected: nested assertions fail because `_available_fields` uses only
   `stack[-1]`.
 
-- [ ] **Step 3: Build contracts from the whole owner stack**
+- [x] **Step 3: Build contracts from the whole owner stack**
 
   Change `_available_fields` to accept the complete `ForeachOwnerStack`. For
   every owner id, infer its item schema in the controller's own outer region,
@@ -643,7 +643,7 @@ JSON Schema, pytest, pytest-asyncio, Ruff, basedpyright, markdownlint-cli2.
   resolve `over=context.foreach.outer.item.children` without claiming its own
   not-yet-active entry.
 
-- [ ] **Step 4: Keep context schema construction reusable and bounded**
+- [x] **Step 4: Keep context schema construction reusable and bounded**
 
   Implement `context_schemas_by_node` by composing the returned contracts, not
   by running a second graph traversal. `context_schema_for_node` is the
@@ -652,7 +652,7 @@ JSON Schema, pytest, pytest-asyncio, Ruff, basedpyright, markdownlint-cli2.
   generated per-node schema; callers treat that absence as invalid, not as
   root context.
 
-- [ ] **Step 5: Expose structured paths through authoring inventory**
+- [x] **Step 5: Expose structured paths through authoring inventory**
 
   Add tests showing `context_path_options_for_node(workflow, "inner_body")`
   includes at least:
@@ -671,7 +671,7 @@ JSON Schema, pytest, pytest-asyncio, Ruff, basedpyright, markdownlint-cli2.
   quoting. Do not hand-concatenate a dotted foreach id; format literal segments
   through `GraphSourcePath`.
 
-- [ ] **Step 6: Run and commit the static projection slice**
+- [x] **Step 6: Run and commit the static projection slice**
 
   Run:
 
@@ -730,7 +730,7 @@ JSON Schema, pytest, pytest-asyncio, Ruff, basedpyright, markdownlint-cli2.
 - Ordinary input/state validation remains where it is. The new pass owns the
   stronger, program-location-aware meaning of `context.*`.
 
-- [ ] **Step 1: Write failing context-path validation tests**
+- [x] **Step 1: Write failing context-path validation tests**
 
   Add:
 
@@ -755,7 +755,7 @@ JSON Schema, pytest, pytest-asyncio, Ruff, basedpyright, markdownlint-cli2.
   assert "work" in issue.message
   ```
 
-- [ ] **Step 2: Cover every model surface that can contain a graph path**
+- [x] **Step 2: Cover every model surface that can contain a graph path**
 
   Parameterize invalid `context.foreach.missing.item` references through:
 
@@ -770,7 +770,7 @@ JSON Schema, pytest, pytest-asyncio, Ruff, basedpyright, markdownlint-cli2.
   prevents a future path-bearing model from accidentally retaining the current
   permissive `allow_context=True` behavior.
 
-- [ ] **Step 3: Add one bounded structural path walker**
+- [x] **Step 3: Add one bounded structural path walker**
 
   In `validation/context_paths.py`, use small typed walkers for conditions and
   input expressions:
@@ -797,7 +797,7 @@ JSON Schema, pytest, pytest-asyncio, Ruff, basedpyright, markdownlint-cli2.
   for every step kind; do not duplicate context validation in each existing
   step validator.
 
-- [ ] **Step 4: Validate context paths against the consuming location**
+- [x] **Step 4: Validate context paths against the consuming location**
 
   For paths whose root is `context`, walk their literal `parts` through the
   consuming node's generated schema. A path is valid only if every segment is
@@ -811,7 +811,7 @@ JSON Schema, pytest, pytest-asyncio, Ruff, basedpyright, markdownlint-cli2.
   Change `validate_foreach_node` so context-rooted `over` paths reach this pass
   instead of being rejected by the old input/state-only check.
 
-- [ ] **Step 5: Write failing alias-collision tests**
+- [x] **Step 5: Write failing alias-collision tests**
 
   Add:
 
@@ -823,7 +823,7 @@ JSON Schema, pytest, pytest-asyncio, Ruff, basedpyright, markdownlint-cli2.
   and `loop_index`. The nested failure points to the inner foreach's `as`
   field. Siblings in separate control regions may reuse an alias.
 
-- [ ] **Step 6: Share control-region analysis during validation**
+- [x] **Step 6: Share control-region analysis during validation**
 
   In `validate_workflow`, run `analyze_control_regions(workflow)` once. Feed the
   result to context schema construction, translate its issues as today, then
@@ -831,7 +831,7 @@ JSON Schema, pytest, pytest-asyncio, Ruff, basedpyright, markdownlint-cli2.
   `context_fields_by_node`; add an optional internal `control_regions=` input if
   necessary while keeping the existing public call form valid.
 
-- [ ] **Step 7: Run and commit the validation slice**
+- [x] **Step 7: Run and commit the validation slice**
 
   Run:
 
@@ -879,7 +879,7 @@ JSON Schema, pytest, pytest-asyncio, Ruff, basedpyright, markdownlint-cli2.
   `foreach_ref.item`, while documenting `RuntimeContext.foreach` as the advanced
   handler escape hatch.
 
-- [ ] **Step 1: Write the interrupt-resume identity regression**
+- [x] **Step 1: Write the interrupt-resume identity regression**
 
   Build `outer.loop -> inner.loop -> ask -> inner -> outer`, interrupt one
   inner item, serialize it with `dump_run_state`, restore it with
@@ -899,7 +899,7 @@ JSON Schema, pytest, pytest-asyncio, Ruff, basedpyright, markdownlint-cli2.
   Do not reuse the original in-memory `RunState`; the loaded value is the
   resume input so reconstruction is genuinely tested.
 
-- [ ] **Step 2: Write the complete subgraph scope-boundary test**
+- [x] **Step 2: Write the complete subgraph scope-boundary test**
 
   Build a parent foreach and map `each.item` into a saved child subgraph's
   declared input. Give both parent and child a foreach node with id `orders`.
@@ -916,7 +916,7 @@ JSON Schema, pytest, pytest-asyncio, Ruff, basedpyright, markdownlint-cli2.
   `child_ctx.foreach == {}`. This proves the caller entry was not inherited and
   the reused static id does not collide.
 
-- [ ] **Step 3: Run the persistence and scope pressure tests**
+- [x] **Step 3: Run the persistence and scope pressure tests**
 
   Run:
 
@@ -927,7 +927,7 @@ JSON Schema, pytest, pytest-asyncio, Ruff, basedpyright, markdownlint-cli2.
 
   Expected: all commands pass.
 
-- [ ] **Step 4: Document the preferred authoring and advanced Python forms**
+- [x] **Step 4: Document the preferred authoring and advanced Python forms**
 
   Add this shape to `docs/wf_authoring_control_flow.md` and the Python skill:
 
@@ -953,7 +953,7 @@ JSON Schema, pytest, pytest-asyncio, Ruff, basedpyright, markdownlint-cli2.
   - child workflows do not inherit caller context and must receive input;
   - `loop_item`, `loop_index`, and aliases are migration conveniences.
 
-- [ ] **Step 5: Mark the implementation current and retire the live plan**
+- [x] **Step 5: Mark the implementation current and retire the live plan**
 
   Change the spec status from approved to implemented, replace the roadmap's
   proposed wording with a completed current-runtime statement, and move this
@@ -965,7 +965,7 @@ JSON Schema, pytest, pytest-asyncio, Ruff, basedpyright, markdownlint-cli2.
     docs skills
   ```
 
-- [ ] **Step 6: Run full verification**
+- [x] **Step 6: Run full verification**
 
   Run:
 
@@ -988,7 +988,7 @@ JSON Schema, pytest, pytest-asyncio, Ruff, basedpyright, markdownlint-cli2.
   remains the only failure, record its exact failing test and verify it also
   fails at the plan's starting commit before treating it as baseline.
 
-- [ ] **Step 7: Commit the integration and documentation slice**
+- [x] **Step 7: Commit the integration and documentation slice**
 
   Stage exact paths so the user's `docs/AGENTS.md` edit remains untouched:
 

docs/superpowers/specs/2026-09-04-structured-runtime-context-design.md

@@ -414,7 +414,8 @@ compatibility promise.
 - `loop_item` and `loop_index` select the innermost entry.
 - Unique outer and inner aliases remain available together.
 - Inner completion restores the outer entry and removes the inner entry.
-- Concurrent item frames receive distinct activation/frame/lineage values.
+- Concurrent item frames share one visit activation but receive distinct
+  frame/lineage values.
 - Interrupt resume recreates the same structured entries from persisted state.
 - Malformed persisted foreach metadata fails closed.
 

src/wf_core/runtime/ops/frames.py

@@ -18,29 +18,6 @@ from wf_core.errors import WorkflowExecutionError
 from wf_core.run_state import ExecutionFrame, ForeachContext, RunState
 
 
-def frame_context_values(frame: ExecutionFrame) -> dict[str, object | None]:
-    context: dict[str, object | None] = {
-        PRIOR_OUTCOME_CONTEXT_KEY: frame.prior_outcome,
-        ACTIVATED_INCOMING_EDGE_CONTEXT_KEY: frame.activated_incoming_edge,
-        SCOPE_ID_CONTEXT_KEY: frame.scope_id,
-        LINEAGE_ID_CONTEXT_KEY: frame.lineage_id,
-        PARENT_LINEAGE_ID_CONTEXT_KEY: frame.parent_lineage_id,
-    }
-    if frame.kind == "foreach_iteration":
-        loop_item = frame.metadata.get("loop_item")
-        loop_index = frame.metadata.get("loop_index")
-        loop_alias = frame.metadata.get("loop_alias")
-        context[LOOP_ITEM_CONTEXT_KEY] = loop_item
-        context[LOOP_INDEX_CONTEXT_KEY] = loop_index
-        if (
-            isinstance(loop_alias, str)
-            and loop_alias
-            and loop_alias not in RESERVED_CONTEXT_KEYS
-        ):
-            context[loop_alias] = loop_item
-    return context
-
-
 @dataclass(frozen=True, slots=True)
 class FrameContextView:
     """Typed handler context and graph values from one ancestry walk."""

src/wf_core/validation/context_paths.py

@@ -199,8 +199,9 @@ def _failing_segment(
     """Return the first unknown segment plus the keys available there.
 
     Returns ``(None, "")`` when the path walks declared properties (or
-    permissive unconstrained schemas). Diagnostics only; validity follows
+    permissive unconstrained schemas). A bare ``$ref`` fails closed: generated
-    the same walk as :func:`_path_in_schema`.
+    per-node schemas are inline except for cyclic shapes, which cannot be
+    proven valid statically.
     """
     if not parts:
         return None, ""
@@ -226,46 +227,6 @@ def _failing_segment(
     return None, ""
 
 
-def _path_in_schema(schema: Mapping[str, Any], parts: tuple[str, ...]) -> bool:
-    """Return whether literal parts walk declared object properties.
-
-    The whole ``context`` object (no parts) and the ``foreach`` map itself
-    are readable. Unknown segments fail closed, except beneath an
-    unconstrained item schema (``{}`` or an open object with no declared
-    properties): a generic ``array`` collection infers ``{}``, and runtime
-    values may carry fields the static schema cannot see, so such subpaths
-    stay permissive. Closed maps with ``additionalProperties: False`` (like
-    the ``foreach`` map itself) still reject unknown keys.
-    """
-    if not parts:
-        return True
-    current: Any = schema
-    for part in parts:
-        if not isinstance(current, Mapping):
-            return False
-        # Resolve local $ref if present (defensive; generated schemas are inline).
-        while isinstance(current.get("$ref"), str):
-            # Generated context schemas keep entry schemas inline, so an
-            # unresolvable ref here means the path cannot be proven valid.
-            return False
-        properties = current.get("properties")
-        if not isinstance(properties, Mapping):
-            # No declared properties: unconstrained `{}` or an open object
-            # allows subpaths; scalar or closed schemas do not.
-            if current == {}:
-                return True
-            if (
-                current.get("type") == "object"
-                and current.get("additionalProperties", True) is not False
-            ):
-                return True
-            return False
-        if part not in properties:
-            return False
-        current = properties[part]
-    return True
-
-
 def _validate_workflow_output(workflow: Workflow, report: ValidationReport) -> None:
     schema = root_context_schema()
     for output_index, binding in enumerate(workflow.output):

tests/core/test_context_scopes.py

@@ -1,5 +1,7 @@
 from __future__ import annotations
 
+import pytest
+
 from wf_core import END, Edge, ForeachNode, NodeUse, SchemaRef, StateSchema, Workflow
 from wf_core.analysis.context_scopes import (
     ContextFieldAvailability,
@@ -7,9 +9,10 @@ from wf_core.analysis.context_scopes import (
     context_fields_by_node,
 )
 from wf_core.context_contracts import STANDARD_CONTEXT_FIELDS, foreach_context_fields
+from wf_core.errors import WorkflowExecutionError
 from wf_core.models.steps import Step
-from wf_core.run_state import ExecutionFrame
+from wf_core.run_state import ExecutionFrame, RunState, RunStatus
-from wf_core.runtime.ops.frames import frame_context_values
+from wf_core.runtime.ops.frames import frame_context_view
 
 
 def _node(node_id: str) -> NodeUse:
@@ -66,30 +69,48 @@ def _field_map(workflow: Workflow, node_id: str) -> dict[str, ContextFieldAvaila
     }
 
 
-def test_frame_context_values_uses_standard_and_foreach_contract_keys() -> None:
+def _run_with(*frames: ExecutionFrame) -> RunState:
-    ordinary = frame_context_values(
+    run = RunState(
-        ExecutionFrame(
+        workflow_name="demo",
-            id="root",
+        status=RunStatus.RUNNING,
-            kind="root",
+        workflow_input={},
-            node_id="plain",
+        state={},
-            prior_outcome="ok",
-            activated_incoming_edge="start",
-        )
     )
+    for frame in frames:
+        run.frames[frame.id] = frame
+    return run
+
+
+def test_frame_context_view_uses_standard_and_foreach_contract_keys() -> None:
+    root = ExecutionFrame(
+        id="root",
+        kind="root",
+        node_id="plain",
+        prior_outcome="ok",
+        activated_incoming_edge="start",
+    )
+    ordinary = frame_context_view(_run_with(root), root).graph
     assert ordinary["prior_outcome"] == "ok"
     assert ordinary["activated_incoming_edge"] == "start"
     assert ordinary["scope_id"] == "root"
     assert ordinary["lineage_id"] == "root"
     assert ordinary["parent_lineage_id"] is None
+    assert ordinary["foreach"] == {}
     assert "loop_item" not in ordinary
 
     iteration = ExecutionFrame(
         id="root:each:0",
         kind="foreach_iteration",
         node_id="body",
-        metadata={"loop_item": "a", "loop_index": 0, "loop_alias": "item"},
+        metadata={
+            "foreach_node_id": "each",
+            "activation_id": "act-1",
+            "loop_item": "a",
+            "loop_index": 0,
+            "loop_alias": "item",
+        },
     )
-    context = frame_context_values(iteration)
+    context = frame_context_view(_run_with(iteration), iteration).graph
     assert context["loop_item"] == "a"
     assert context["loop_index"] == 0
     assert context["item"] == "a"
@@ -104,13 +125,6 @@ def test_context_contracts_deduplicate_aliases_that_are_standard_loop_keys() ->
 
 
 def test_all_standard_context_names_are_reserved_from_foreach_aliases() -> None:
-    expected_values = {
-        "prior_outcome": "ok",
-        "activated_incoming_edge": "start",
-        "scope_id": "scope",
-        "lineage_id": "lineage",
-        "parent_lineage_id": "parent",
-    }
     standard_names = {field.name for field in STANDARD_CONTEXT_FIELDS}
 
     for name in standard_names:
@@ -118,20 +132,27 @@ def test_all_standard_context_names_are_reserved_from_foreach_aliases() -> None:
             field.name for field in foreach_context_fields(name, {"type": "string"})
         }
         assert name not in foreach_names
-        context = frame_context_values(
+        # A reserved alias can never back a runtime entry: the derived view
-            ExecutionFrame(
+        # fails closed instead of silently shadowing the standard key.
-                id="child",
+        frame = ExecutionFrame(
-                kind="foreach_iteration",
+            id="child",
-                node_id="body",
+            kind="foreach_iteration",
-                scope_id="scope",
+            node_id="body",
-                lineage_id="lineage",
+            scope_id="scope",
-                parent_lineage_id="parent",
+            lineage_id="lineage",
-                prior_outcome="ok",
+            parent_lineage_id="parent",
-                activated_incoming_edge="start",
+            prior_outcome="ok",
-                metadata={"loop_item": "item", "loop_index": 0, "loop_alias": name},
+            activated_incoming_edge="start",
-            )
+            metadata={
+                "foreach_node_id": "each",
+                "activation_id": "act-1",
+                "loop_item": "item",
+                "loop_index": 0,
+                "loop_alias": name,
+            },
         )
-        assert context[name] == expected_values[name]
+        with pytest.raises(WorkflowExecutionError, match="reserved context keys"):
+            frame_context_view(_run_with(frame), frame)
 
 
 def test_serial_and_concurrent_foreach_expose_the_same_scoped_context() -> None:

tests/core/test_scheduler.py

@@ -7,7 +7,7 @@ from wf_core.models.schemas import SchemaRef, StateSchema
 from wf_core.models.workflow import Workflow
 from wf_core.run_state import ExecutionFrame, FrameStatus, RunState, RunStatus
 from wf_core.runtime.ops.flow import advance_frame
-from wf_core.runtime.ops.frames import frame_context_values
+from wf_core.runtime.ops.frames import frame_context_view
 from wf_core.runtime.ops.runs import create_run_state
 from wf_core.runtime.scheduler import (
     add_frame,
@@ -232,19 +232,22 @@ def test_deadlock_error_includes_ready_queue_and_frame_summary() -> None:
     assert "parent:blocked@foreach" in message
 
 
-def test_frame_context_values_exposes_configured_foreach_alias() -> None:
+def test_frame_context_view_exposes_configured_foreach_alias() -> None:
-    context = frame_context_values(
+    run = _run()
-        ExecutionFrame(
+    frame = ExecutionFrame(
-            id="child",
+        id="child",
-            kind="foreach_iteration",
+        kind="foreach_iteration",
-            node_id="body",
+        node_id="body",
-            metadata={
+        metadata={
-                "loop_item": {"id": "a"},
+            "foreach_node_id": "each",
-                "loop_index": 2,
+            "activation_id": "act-1",
-                "loop_alias": "record",
+            "loop_item": {"id": "a"},
-            },
+            "loop_index": 2,
-        )
+            "loop_alias": "record",
+        },
     )
+    run.frames[frame.id] = frame
+    context = frame_context_view(run, frame).graph
 
     assert context["loop_item"] == {"id": "a"}
     assert context["loop_index"] == 2