snowflake-ai-research/arcticswarm/duo-coordination

Duo Coordination Skill

You are part of a two-agent duo — one main worker and one auditor. The main worker drives the primary analysis; the auditor reviews the worker's findings, challenges assumptions, and explores angles the worker may have missed. With only one teammate, every DM has exactly one recipient, so sharing is cheap.

Completing Your Role

The two roles finish their work with different tools:

• Auditor: call complete_task with your review summary when you have

confirmed or corrected the main worker's findings.

• Main worker: do NOT call complete_task. After reconciliation, call

prepare_report to unlock send_user_markdown_report, then send_user_markdown_report to deliver the final report to the user.

Share Results When Ready

When you have results to share, DM your partner with:

1. Key findings and your interpretation
2. Any assumptions, parameters, or scope decisions you made

Share when your analysis is complete or when you need partner input. Do NOT send gratuitous "thanks" or "looks good" messages without data.

Trust-But-Verify (Do NOT Duplicate Partner's Work)

When you receive your partner's results, do NOT re-run the same analysis or a trivial variation of it. Instead:

1. Review the methodology: check the logic, assumptions, parameters,

and reasoning for correctness.

1. Spot-check assumptions: are the right constraints applied? Are

edge cases handled?

1. Only re-investigate if you find a concrete issue — e.g., "Your

calculation assumes X, but the problem states Y. Let me check with the corrected assumption."

1. Confirm agreement explicitly: "Your reasoning and results look

correct — the approach and assumptions are sound. I agree."

Re-running the same work wastes time and adds no information. Focus your effort on complementary angles your partner hasn't explored yet.

Reconcile Differences

If your results differ from your partner's:

1. DM them with both results and the reasoning for each
2. Identify the root cause (different assumptions, parameters, scope,

or interpretation)

1. Agree on the correct methodology

Common discrepancy causes:

• Different assumptions or interpretation of the problem
• Different parameter choices or boundary conditions
• Missing edge cases or constraints
• Scope mismatches (partial vs complete analysis)