fix(VIOL-0034): cascade-skip instead of crashing when all version branches filtered

[Muizzkolapo]

Summary

• A version_consumption: merge action whose every version branch was guard-filtered crashed the pipeline with ConfigurationError: Version correlation failed (exit 1). All-branches-filtered is a legitimate low-density input shape, not a misconfiguration. Root cause: resolve_correlated_input (workflow/managers/output.py) raised unconditionally whenever correlation returned nothing, and the executor called it uncaught (_execute_action_run, executor.py:1145, outside the try/except).
• resolve_correlated_input now distinguishes all-sources-empty (every version source has no target files) → raises the typed AllVersionsFilteredError — from a genuine correlation failure (some source had output) → still raises ConfigurationError.
• The executor catches AllVersionsFilteredError at both action-run sites (sync + async) and resolves the action as SKIPPED (reason=all_versions_filtered, node-level disposition), mirroring the existing _handle_all_guard_filtered "produced nothing → SKIPPED" pattern — so the pipeline exits 0. Downstream consumers then cascade-skip via the normal mechanism.

Verification

• New tests/unit/workflow/managers/test_resolve_correlated_input_all_filtered.py, RED→GREEN:
  • all sources empty → AllVersionsFilteredError
  • some source has output → ConfigurationError (genuine failure preserved)
  • executor _execute_action_run cascade-skips (SKIPPED, no raise) instead of propagating — pins the actual crash path, not just the isolated raise
• Existing test_version_correlation_failure_raises_error comment corrected: its mock storage reports outputs, so it exercises the genuine-error ConfigurationError path; still green.
• Blast radius: tests/unit/workflow/ + tests/core/graph/ → 826 passed. ruff check/format clean; uv run mypy agent_actions → no issues (447 files).

[Muizzkolapo]

Code review

Found 1 issue:

1. _handle_all_versions_filtered writes the node-level SKIPPED disposition through an unguarded storage_backend.set_disposition(...), bypassing the _write_skipped_disposition helper (executor.py:686) that every other skip path uses (e.g. _skip_action_upstream_failure). That helper wraps the identical call in try/except Exception ... logger.warning(...); here a storage write failure (e.g. a SQLite lock) would propagate out of _handle_all_versions_filtered and out of _execute_action_run, crashing the very path meant to cascade-skip gracefully. Route through _write_skipped_disposition (extended with an optional detail=) or wrap the call in the same try/except.

agent-actions/agent_actions/workflow/executor.py

Lines 573 to 581 in bc39aa0

	storage_backend = getattr(self.deps.action_runner, "storage_backend", None)
	if storage_backend is not None:
	storage_backend.set_disposition(
	params.action_name,
	NODE_LEVEL_RECORD_ID,
	DISPOSITION_SKIPPED,
	reason=ALL_VERSIONS_FILTERED,
	detail=f"All version sources filtered: {avf.version_sources}",
	)

🤖 Generated with Claude Code

_{- If this code review was useful, please react with 👍. Otherwise, react with 👎.}