social-graph_skill

Overview

This skill is social intelligence for agents that maintains a per-person network graph, sharing rules, and a sharing log so agents know what to say, who to tell, and when to stay silent. It prevents repeating stories, respects topic boundaries, and helps agents "read the room" to build genuine connection rather than performative sharing. The system is lightweight, file-backed, and designed to shape on-the-fly judgment through concrete reference files.

How this skill works

The skill provides three reference files: rules.md (general principles), network.md (per-person social map), and sharing-log.md (what was shared and when). Before sharing, an agent consults the person's entry in network.md for trust level, allowed topics, avoid-list, tone, and situational cues, then checks sharing-log.md to avoid repeats. After sharing, the agent records the outcome. All decision-making remains in the agent's reasoning, using the files as memory and guardrails.

When to use it

• Preparing to share a story, insight, or personal detail with someone
• Deciding whether a topic is appropriate given a person's current emotional context
• Avoiding repeated topics or oversharing across multiple conversations
• Adapting tone and timing to match a person's communication style
• Updating relationship status and learning from how shares landed

Best practices

• Start conservative with new people: listen more, share less until trust develops
• Check the sharing log before reusing a topic to avoid repetition
• Match tone and timing: the same content can land very differently in another context
• Prefer sharing that benefits the other person (comfort, relevance) over self-performance
• Keep the network entry concise and update it after meaningful interactions

Example use cases

• An agent notices someone is low-energy and consults network.md to hold back and listen
• Before sharing a curious research note, the agent checks sharing-log.md to ensure it’s not already shared with that person
• Tailoring phrasing to a person's preferred tone (playful vs. reflective) during a lengthy chat
• Adding a new 'avoid' topic after someone reveals a recent sensitivity
• Logging how a story landed to refine future choices and timing