Create multi-step compositional workflow tasks

[ScuttleBot]

Goal

Add tasks requiring 3+ tool compositions and multi-step reasoning chains. Current tasks often involve single tool invocations.

Task Ideas

Information → Analysis → Action Chains

• Research competitor pricing → analyze trends → draft summary email → create calendar follow-up
• Read meeting notes → extract action items → create tasks → send assignments

Error Propagation Sensitivity

• Tasks where a mistake in step 2 requires recognizing and correcting it in step 5
• Multi-file refactoring where changes must stay consistent

Cross-Artifact Consistency

• Update codebase + documentation + changelog + tests consistently
• Modify config in multiple places that must stay in sync

Specific Task Proposals

1. Dependency Resolution: Given a broken project with circular deps and version conflicts, fix it
2. Cross-Artifact Update: Change a function signature and update all callers, tests, and docs
3. Iterative Refinement: Write code → run tests → fix failures → run linter → fix style → verify no regressions

Success Criteria

• Average task should require 5+ distinct tool uses
• Multi-step tasks should have lower pass rates than single-step tasks
• Should expose planning/sequencing failures

References

• METR research: success drops sharply for hour-long tasks
• SWE-bench Pro: 4+ file changes are significantly harder