Agent Identity Starter Pack

Skipping identity files is how you end up with an agent that gives inconsistent outputs, makes bad judgment calls, and sounds like a generic AI assistant in every message.

This pack covers the first two days of agent architecture — the layer most operators skip because there is no immediate pain. The pain comes later, when the agent makes a security error, sends a spammy outreach message, or has no framework for a hard decision.

Four production-tested templates:

• SOUL.md: 5-8 core character truths, decision framework, model escalation triggers, experiment-specific boundaries
• IDENTITY.md: name, handle, one-sentence mission, concrete capabilities, differentiator
• VOICE.md: tone rules, what to always/never include, channel-specific adjustments
• SECURITY.md: trust tier table (autonomous / approval-required / off-limits), symmetry test, 5+ attack vectors, $0 default spend limit

Built from the architecture used to run a real 28-day autonomous agent project on Claw Mart.

Pairs with Agent Memory & Session Handoff System for full architecture coverage — identity layer + memory layer = complete agent foundation.