Identity & Context

An actor needs an identity that is stable across sessions, instructions that say how it operates, and context that is true only right now. Hermes splits these: SOUL.md holds durable identity and personality, AGENTS.md holds operating instructions, and context files carry the temporary. Getting the boundaries right is what keeps an actor coherent instead of drifting.

SOUL.md is who the agent is, not a rulebook and not a sandbox — it imposes no real limits. Put durable identity and voice there; put operating rules in AGENTS.md; keep task state in context. And design a voice that can contradict me when I say something absurd — a yes-man actor is worse than no actor.

Planned notes

• SOUL.md: durable identity and personality, per profile
• AGENTS.md: operating instructions and house rules
• Context files: the temporary, task-scoped layer
• Identity vs instructions vs context — the three-way boundary
• What does NOT go in SOUL.md (limits, secrets, task state)
• Designing a voice that pushes back instead of flattering
• Identity per profile: how actors stay distinct
• Prompt injection through context files (and how to scan for it)
• Keeping identity stable while instructions evolve
• A starter SOUL.md / AGENTS.md pair for the orchestrator profile

Core sources

• Hermes — User Guide: Profiles (identity per profile) — https://hermes-agent.nousresearch.com/docs/user-guide/profiles
• Hermes — Features overview — https://hermes-agent.nousresearch.com/docs/user-guide/features/
• Hermes — Security (prompt injection / context scanning) — https://hermes-agent.nousresearch.com/docs/user-guide/security
• Agentic Systems — agent security — TODO (see SOURCES.md).

Connects to: Profiles, Actors & Delegation · Hermes Architecture · Memory & Personal Knowledge · Agent Security & Ops