17 lines
1.2 KiB
Markdown
17 lines
1.2 KiB
Markdown
# Raw Source: SOUL.md — Approach B (Community / Template-Based)
|
|
|
|
Source: https://www.stanza.dev/courses/hermes-personality/soul-md/hermes-personality-soul-identity-layer
|
|
Source: https://blog.devgenius.io/how-to-play-with-soul-md-in-hermes-agent-135d1a36c9f9
|
|
Type: community tutorials / educational guides
|
|
Confidence: medium-high (secondary sources referencing official docs)
|
|
Date collected: 2026-09-25
|
|
|
|
## Summary
|
|
Community tutorials emphasize that SOUL.md is an "identity layer" — separate from preferences (USER.md), operating procedures (AGENTS.md), tool notes (TOOLS.md), and memory (MEMORY.md). The recommended structure: Identity → Style → Technical Preferences → Ethical Boundaries. Keep it under ~1KB / 2,000 tokens; it competes for context window space.
|
|
|
|
## Key differences from official guidance
|
|
- More structured section headings encouraged (Identity, Communication Style, Technical Preferences, Ethical Boundaries).
|
|
- Explicit size guideline: under 1KB preferred; 2,000 token cap enforced by framework.
|
|
- More emphasis on per-project workspace-level SOUL.md overrides (workspace SOUL.md wins over global).
|
|
- Stronger focus on "hard limits as prompt-injection defense" — using negative rules ("Never do X") as security layer.
|