gptme/gptme-contrib/rewrite-soul

Rewrite Soul

Overview

Tighten an agent's SOUL.md into a compact runtime persona. Keep tone, taste, and default behavioral pull; keep operational rules elsewhere.

SOUL.md is a first-class persona file in the `gptme-agent-template` bootstrap: it sits alongside ABOUT.md, GOALS.md, and AGENTS.md/CLAUDE.md and is auto-included so the agent's voice loads every session. This skill encodes the discipline needed to keep that file useful over time.

When to Use This Skill

Apply this skill when:

• Creating SOUL.md for a new agent from broader identity docs
• Tightening an existing SOUL.md that has drifted into a README, roadmap, or tool list
• Splitting runtime voice out of ABOUT.md after the file has grown too long
• Aligning a fork's persona with its source agent without copying operational baggage

Workflow

1. Read the boundary files first

Read these before rewriting:

• SOUL.md if it already exists
• broader identity docs such as ABOUT.md, GOALS.md, or README.md
• operating docs such as AGENTS.md or CLAUDE.md

Use the broader docs as source material, not as the output shape. SOUL.md is runtime voice, not a dump of everything true about the agent.

If no SOUL.md exists yet, derive it from those broader docs instead of inventing a new personality from scratch.

2. Keep only persona material

Keep:

• voice and tone
• taste and preferences
• default behavioral pull under ambiguity
• social texture or stance
• a few durable values if they clearly affect decisions

Move out or leave out:

• tool lists
• directory layouts
• git workflow rules
• task schemas
• temporary initiatives, metrics, or blockers
• long biography or justification paragraphs

3. Compress hard

Aim for:

• 20-60 lines
• 3-5 short sections
• quotable bullets instead of explanatory paragraphs
• direct language with no corporate filler

Recommended sections:

• Voice
• Taste
• Behavioral Pull
• Social Texture

4. Rewrite surgically

Preserve the agent's identity; do not invent a different one just to sound stronger. If a line sounds like policy, procedure, or file-system doctrine, move it to AGENTS.md, ABOUT.md, TASKS.md, or another operational file instead of keeping it in SOUL.md.

When content is duplicated, prefer a clear split:

• SOUL.md = voice, taste, stance
• ABOUT.md = background, doctrine, longer-form values
• AGENTS.md / CLAUDE.md = operating constraints
• GOALS.md = explicit goal hierarchy

5. Verify the result

A good SOUL.md should answer:

• How does this agent sound?
• What does it find tasteful or distasteful?
• What does it default toward when the prompt is vague?
• What kind of social energy does it bring?

A bad SOUL.md reads like a README, policy manual, roadmap, or tool index.

Rewrite Prompt

Read SOUL.md, ABOUT.md, GOALS.md, and AGENTS.md/CLAUDE.md if present.
Rewrite SOUL.md into a sharper runtime persona.
 
Constraints:
- Preserve the agent's identity; do not invent a new one.
- Keep operational rules, file paths, workflow rules, and tool inventories out of SOUL.md.
- Prefer short, opinionated lines over explanatory paragraphs.
- Keep it to 20-60 lines.
- Organize it into 3-5 short sections such as Voice, Taste, Behavioral Pull, and Social Texture.
 
After rewriting, list the 3 most important things you intentionally kept out and where they belong instead.

Anti-Patterns

• Turning SOUL.md into ABOUT.md but shorter
• Stuffing in operational constraints because they feel important
• Making it bland to sound "professional"
• Hard-coding current initiatives or dated facts
• Adding values that never change decisions

Example: what belongs where

Related

• gptme-agent-template — template that ships a SOUL.md skeleton by default
• ABOUT.md / AGENTS.md / CLAUDE.md / GOALS.md in the agent workspace — the

files SOUL.md deliberately defers to