What problem does it solve?
When an AI assistant encounters mentions of people in conversation, it must decide whether that information deserves to become persistent memory. Without a judgment policy, agents either silently store unverified personal facts or miss genuinely valuable continuity. This Skill provides a five-gate decision framework that routes each person-related cue to exactly one outcome: a rejectable proposal, a content-free deferral receipt, or an enum-only abstention.
Core Features & Use Cases
- Five-Gate Semantic Judgment: Sequentially checks qualification (continuity value), typed evidence, timing, authorization, and de-escalation before any memory action.
- Three Controlled Outcomes: Produces an F276 rejectable proposal, a known-person deferred delta receipt, or an enum-only abstention record — never silent writes to canonical memory.
- Typed Evidence Discipline: Enforces source roles (owner message, third-party quote, agent inference) so evidence only supports the fields it can actually prove.
- Use Case: During a chat, the owner mentions a colleague's new role. The Skill checks continuity value and evidence, then issues one rejectable proposal card instead of silently storing the fact, letting the owner approve exact items first.
Quick Start
Ask the agent to evaluate whether the person mentioned in this conversation should become a memory proposal, a deferred delta, or an abstention.
