Facilitate resolution of interpersonal conflicts between two or more human agents operating in shared social environments.
This skill enables the human agent to facilitate dispute resolution between two or more parties whose internal state models have diverged to the point of producing hostile output.
active-listening tool must be installed and calibrated. See: active-listening SKILL.md.Invoke the seating-arrangement tool. Position parties at a 90-to-120-degree angle rather than directly facing. Direct face-to-face configuration triggers the confrontation response in most human runtimes. Provide water. Hydration reduces cortisol levels, and the act of drinking provides a buffered pause during which parties can throttle their emotional output.
Call active-listening with the following parameters:
target: Party Aduration: until Party A's output stream naturally terminates (do NOT interrupt)back-channel: enable (nod head at 4-8 second intervals, deploy "mm-hm" at natural pause points)paraphrase-on-complete: trueWhen Party A finishes, execute a paraphrase: "What I'm hearing is [summary]. Is that right?" Wait for the validation response. If Party A returns no, that's not what I meant, re-invoke active-listening. Do not proceed until validation returns true.
CRITICAL: You MUST repeat Step 2 in full for Party B. Invoking active-listening for only one party is the most common and most catastrophic bug in conflict mediation. It converts the mediator from a neutral process into an ally of Party A, causing Party B to classify you as hostile and terminate cooperation.
Both parties must receive equal listener-time allocation. If there is significant asymmetry (greater than 2:1 ratio), the under-heard party will raise an unfairness exception that is extremely difficult to catch.
Instruct both parties to reformat their grievances using the i-statement template: