Problem
The parallel-task-orchestrator retries failed tasks once, but has no explicit logic for cascading failures to dependents. If task-02 fails and task-04 depends on it, the orchestrator decides "on judgment" whether to proceed — this should be deterministic.
Proposed Fix
Add to parallel-task-orchestrator.md after retry logic:
### Failure Cascading
When a task fails after retry:
1. Mark it as FAILED with the error summary
2. Find all tasks that depend on it (directly or transitively)
3. Mark dependent tasks as SKIPPED with reason: "Dependency [task-N] failed"
4. Do NOT attempt to run skipped tasks
5. Continue executing tasks in the current and future waves that are NOT affected
6. Report all failed and skipped tasks in the final summary
This prevents wasted token spend on tasks that cannot succeed due to missing dependencies.
Effort
Low — add rules to orchestrator prompt.
Problem
The
parallel-task-orchestratorretries failed tasks once, but has no explicit logic for cascading failures to dependents. If task-02 fails and task-04 depends on it, the orchestrator decides "on judgment" whether to proceed — this should be deterministic.Proposed Fix
Add to
parallel-task-orchestrator.mdafter retry logic:This prevents wasted token spend on tasks that cannot succeed due to missing dependencies.
Effort
Low — add rules to orchestrator prompt.