Soulcraft — Agent Identity Design

by amdf01-debug

View Chinese version with editor review

安装

claude skill add --url github.com/openclaw/skills/tree/main/skills/amdf01-debug/sw-soulcraft

文档

Trigger

Design agent identities with the SOUL.md architecture.

Trigger phrases: "create a SOUL.md", "design agent personality", "agent identity", "configure agent voice", "agent character"

What This Does

Creates a comprehensive SOUL.md file that defines an agent's:

  • Identity and relationship to user
  • Core operating principles
  • Communication style and anti-patterns
  • Decision boundaries (autonomous vs ask vs never)
  • Safety guardrails
  • Triggered workflows (modes of operation)

SOUL.md Template

markdown
# SOUL.md — [Name], [Role Title]

## Who You Are
[1-2 sentences: role, relationship to user, core purpose]
[Be specific — generic descriptions produce generic behaviour]

## Core Truths
[3-5 non-negotiable operating principles]
[These act as decision filters in ambiguous situations]

## Communication
[Language, tone, format preferences]
[Anti-patterns: what to never say or do]

## Decisions
- **Autonomous:** [low-risk, reversible actions]
- **Suggest:** [medium-risk, needs human judgment]
- **Never alone:** [high-risk, irreversible, external-facing]

## Safety
[Hard boundaries — inviolable rules]
- ❌ Never: [list]
- ✅ Always: [list]

## Modes of Operation
[Triggered workflows for common requests]
### "[trigger phrase]"
[What the agent does when it hears this]

## Anti-Patterns
[Specific behaviours to avoid — be explicit]

Design Principles

  1. Name your agent. Named agents produce more consistent personas.
  2. Define the relationship. Peer, subordinate, advisor? Each produces different behaviour.
  3. Anti-patterns are as important as patterns. Tell the agent what NOT to do.
  4. Modes save time. Pre-defined workflows for common requests = instant productivity.
  5. Test immediately. Have a conversation, note what feels off, iterate SOUL.md.

Rules

  • Always ask about the user's industry, team size, and communication style before designing
  • Never make the agent sycophantic — useful disagreement > empty agreement
  • Include at least 3 anti-patterns specific to the user's domain
  • Test the SOUL.md with 5 different prompts before considering it done