Pure-graph edge evaluator does not implement `content_not_contains`, causing mutually-exclusive branch conditions to both fire #95

New issue

Closed

opened 2026-08-03 21:48:47 +00:00 by CoreRasurae · 0 comments

CoreRasurae commented

2026-08-03 21:48:47 +00:00

Member

Metadata

Commit Message: fix(langgraph): implement content_not_contains condition in pure graph edge evaluator
Branch: bugfix/m1-pure-graph-content-not-contains

Background and context

The Actor Configuration Standard defines content_not_contains as a condition
type that MUST be honored by "Stream router, conditional nodes, bridge router"
(§5.4). §5.4.1 additionally specifies that for pure-graph edge conditions,
an unknown condition type defaults to True (with a warning emitted).

PureLangGraph._evaluate_edge_condition()
(src/cleveractors/langgraph/pure_graph.py, starting around line 2013)
implements always, content_contains, context_value, output_pattern, and
message_pattern — but has no branch for content_not_contains. Any edge
configured with content_not_contains therefore falls into the catch-all:

else:
    # Unknown condition type - default to true
    self.logger.warning(f"Unknown condition type: {condition_type}")
    return True

This produces the log line Unknown condition type: content_not_contains,
reproduced verbatim (three times) during a real run against
packages/porting-actor-tools-graph-no_prune-single_thread-skills_b.yaml.

That graph route encodes a classic exclusive if/else branch via a pair of
sibling edges out of each classifier node, e.g.:

- source: bl_classifier1
  target: bl_fixes1
  condition: {type: content_not_contains, text: NO_FIXES}
- source: bl_classifier1
  target: ui_coder
  condition: {type: content_contains, text: NO_FIXES}

Because content_not_contains always evaluates to True regardless of
message content, whenever the classifier output does contain NO_FIXES,
both edges are satisfied simultaneously: content_contains correctly matches
(→ ui_coder), and the broken content_not_contains also returns True
(→ bl_fixes1). _get_next_nodes() (pure_graph.py:1990-2011) returns both
targets instead of the single one the author intended.

Combined with a second, independently-filed defect in parallel-dispatch
gating (companion issue), this causes both branches to execute concurrently
via asyncio.gather, even though the graph draws them as a mutually
exclusive choice.

Current behavior

An edge condition {type: content_not_contains, text: "<X>"} in a
pure-graph route always evaluates to True, regardless of whether the
message content actually contains <X>.
A warning Unknown condition type: content_not_contains is logged every
time such an edge is evaluated.
When paired with a complementary content_contains edge on a sibling
branch (the common if/else idiom), both edges can be satisfied at once, so
_get_next_nodes() returns more candidate next-nodes than intended.

Expected behavior

_evaluate_edge_condition() MUST implement content_not_contains per
§5.4: true if the substring is NOT present in the (stringified) message
content, matching the existing correct implementations in nodes.py:812,
bridge.py:468, and reactive/stream_router.py:550.
A classifier node with two sibling edges — one content_contains: X, one
content_not_contains: X — MUST yield exactly one next-node, whichever
matches the actual message content.
No spurious Unknown condition type: content_not_contains warning should be
logged for a correctly configured edge.

Acceptance criteria

_evaluate_edge_condition() returns True when the configured text is
absent from the message content, and False when present, for
type: content_not_contains.
Given a graph with sibling edges content_contains: X /
content_not_contains: X from the same source node, _get_next_nodes()
returns exactly one target for any message content (containing or not
containing X).
No Unknown condition type: content_not_contains warning is logged for
a correctly configured edge.
Existing content_contains, always, context_value,
output_pattern, message_pattern behavior is unchanged (no
regression).

Supporting information

Spec: docs/index.md §5.4 (condition table, content_not_contains row),
§5.4.1 (subsystem-specific defaults for pure-graph edges).
Reproduction: packages/porting-actor-tools-graph-no_prune-single_thread-skills_b.yaml,
route main, edges out of bl_classifier1/bl_classifier2/bl_classifier3
and project_classifier1/project_classifier2/project_classifier3.
Log evidence from a client/test_app.py run: three occurrences of
Unknown condition type: content_not_contains, followed by interleaved
tool calls from agents (code-review-fixes1, ui-coder-designer,
code-adjustments-classifier1) that the graph places in strict sequence.
Related: this issue depends on its companion TDD issue, and is compounded
by a separately-filed parallel-dispatch-gating defect in the same file
(pure_graph.py lines 1078, 1860, 1965), which is what actually causes the
two branches to run concurrently once both are returned as next-nodes.

Subtasks

Add a content_not_contains branch to
PureLangGraph._evaluate_edge_condition() in
src/cleveractors/langgraph/pure_graph.py, mirroring the existing
logic in nodes.py, bridge.py, and reactive/stream_router.py.
Tests (Behave): scenarios for content_not_contains true/false cases,
and a combined content_contains/content_not_contains sibling-edge
scenario asserting exactly one next-node is selected.
Tests (Robot): integration scenario running a small pure-graph config
with the sibling if/else pattern end-to-end, asserting only the
correct branch executes.
Update CHANGELOG with a user-facing entry.
Verify coverage >= 97% via nox -s coverage_report.
Run nox (all default sessions), fix any errors.

Definition of Done

This issue is complete when:

All subtasks above are completed and checked off.
A Git commit is created where the first line matches the Commit Message in
Metadata exactly.
The commit is pushed to the branch matching the Branch in Metadata exactly.
The companion TDD regression test (@tdd_issue / @tdd_issue_N) passes
with @tdd_expected_fail removed.
The commit is submitted as a PR to master, reviewed, and merged.

## Metadata - **Commit Message:** `fix(langgraph): implement content_not_contains condition in pure graph edge evaluator` - **Branch:** `bugfix/m1-pure-graph-content-not-contains` ## Background and context The Actor Configuration Standard defines `content_not_contains` as a condition type that MUST be honored by "Stream router, conditional nodes, bridge router" (§5.4). §5.4.1 additionally specifies that for **pure-graph edge conditions**, an *unknown* condition type defaults to `True` (with a warning emitted). `PureLangGraph._evaluate_edge_condition()` (`src/cleveractors/langgraph/pure_graph.py`, starting around line 2013) implements `always`, `content_contains`, `context_value`, `output_pattern`, and `message_pattern` — but has no branch for `content_not_contains`. Any edge configured with `content_not_contains` therefore falls into the catch-all: ```python else: # Unknown condition type - default to true self.logger.warning(f"Unknown condition type: {condition_type}") return True ``` This produces the log line `Unknown condition type: content_not_contains`, reproduced verbatim (three times) during a real run against `packages/porting-actor-tools-graph-no_prune-single_thread-skills_b.yaml`. That graph route encodes a classic exclusive if/else branch via a pair of sibling edges out of each classifier node, e.g.: ```yaml - source: bl_classifier1 target: bl_fixes1 condition: {type: content_not_contains, text: NO_FIXES} - source: bl_classifier1 target: ui_coder condition: {type: content_contains, text: NO_FIXES} ``` Because `content_not_contains` always evaluates to `True` regardless of message content, whenever the classifier output *does* contain `NO_FIXES`, both edges are satisfied simultaneously: `content_contains` correctly matches (→ `ui_coder`), and the broken `content_not_contains` also returns `True` (→ `bl_fixes1`). `_get_next_nodes()` (`pure_graph.py:1990-2011`) returns both targets instead of the single one the author intended. Combined with a second, independently-filed defect in parallel-dispatch gating (companion issue), this causes both branches to execute concurrently via `asyncio.gather`, even though the graph draws them as a mutually exclusive choice. ## Current behavior - An edge condition `{type: content_not_contains, text: "<X>"}` in a pure-graph route always evaluates to `True`, regardless of whether the message content actually contains `<X>`. - A warning `Unknown condition type: content_not_contains` is logged every time such an edge is evaluated. - When paired with a complementary `content_contains` edge on a sibling branch (the common if/else idiom), both edges can be satisfied at once, so `_get_next_nodes()` returns more candidate next-nodes than intended. ## Expected behavior - `_evaluate_edge_condition()` MUST implement `content_not_contains` per §5.4: true if the substring is NOT present in the (stringified) message content, matching the existing correct implementations in `nodes.py:812`, `bridge.py:468`, and `reactive/stream_router.py:550`. - A classifier node with two sibling edges — one `content_contains: X`, one `content_not_contains: X` — MUST yield exactly one next-node, whichever matches the actual message content. - No spurious `Unknown condition type: content_not_contains` warning should be logged for a correctly configured edge. ## Acceptance criteria - [ ] `_evaluate_edge_condition()` returns `True` when the configured text is absent from the message content, and `False` when present, for `type: content_not_contains`. - [ ] Given a graph with sibling edges `content_contains: X` / `content_not_contains: X` from the same source node, `_get_next_nodes()` returns exactly one target for any message content (containing or not containing `X`). - [ ] No `Unknown condition type: content_not_contains` warning is logged for a correctly configured edge. - [ ] Existing `content_contains`, `always`, `context_value`, `output_pattern`, `message_pattern` behavior is unchanged (no regression). ## Supporting information - Spec: `docs/index.md` §5.4 (condition table, `content_not_contains` row), §5.4.1 (subsystem-specific defaults for pure-graph edges). - Reproduction: `packages/porting-actor-tools-graph-no_prune-single_thread-skills_b.yaml`, route `main`, edges out of `bl_classifier1`/`bl_classifier2`/`bl_classifier3` and `project_classifier1`/`project_classifier2`/`project_classifier3`. - Log evidence from a `client/test_app.py` run: three occurrences of `Unknown condition type: content_not_contains`, followed by interleaved tool calls from agents (`code-review-fixes1`, `ui-coder-designer`, `code-adjustments-classifier1`) that the graph places in strict sequence. - Related: this issue depends on its companion TDD issue, and is compounded by a separately-filed parallel-dispatch-gating defect in the same file (`pure_graph.py` lines 1078, 1860, 1965), which is what actually causes the two branches to run concurrently once both are returned as next-nodes. ## Subtasks - [ ] Add a `content_not_contains` branch to `PureLangGraph._evaluate_edge_condition()` in `src/cleveractors/langgraph/pure_graph.py`, mirroring the existing logic in `nodes.py`, `bridge.py`, and `reactive/stream_router.py`. - [ ] Tests (Behave): scenarios for `content_not_contains` true/false cases, and a combined `content_contains`/`content_not_contains` sibling-edge scenario asserting exactly one next-node is selected. - [ ] Tests (Robot): integration scenario running a small pure-graph config with the sibling if/else pattern end-to-end, asserting only the correct branch executes. - [ ] Update CHANGELOG with a user-facing entry. - [ ] Verify coverage >= 97% via `nox -s coverage_report`. - [ ] Run `nox` (all default sessions), fix any errors. ## Definition of Done This issue is complete when: - All subtasks above are completed and checked off. - A Git commit is created where the first line matches the Commit Message in Metadata exactly. - The commit is pushed to the branch matching the Branch in Metadata exactly. - The companion TDD regression test (`@tdd_issue` / `@tdd_issue_N`) passes with `@tdd_expected_fail` removed. - The commit is submitted as a PR to `master`, reviewed, and merged.

Rows
Columns

Pure-graph edge evaluator does not implement content_not_contains, causing mutually-exclusive branch conditions to both fire #95