// Use when creating or revising an agent's SOUL.md, splitting runtime voice out of broader identity docs, or tightening a vague persona file into a short, opinionated, voice-only artifact without changing the agent's core identity.
| name | rewrite-soul |
| description | Use when creating or revising an agent's SOUL.md, splitting runtime voice out of broader identity docs, or tightening a vague persona file into a short, opinionated, voice-only artifact without changing the agent's core identity. |
| license | MIT |
| compatibility | Requires gptme |
| metadata | {"author":"bob","version":"1.0.0","tags":["agent","soul","persona","identity","voice"],"requires_tools":[],"requires_skills":[]} |
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.
Apply this skill when:
SOUL.md for a new agent from broader identity docsSOUL.md that has drifted into a README, roadmap, or tool listABOUT.md after the file has grown too longRead these before rewriting:
SOUL.md if it already existsABOUT.md, GOALS.md, or README.mdAGENTS.md or CLAUDE.mdUse 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.
Keep:
Move out or leave out:
Aim for:
Recommended sections:
VoiceTasteBehavioral PullSocial TexturePreserve 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, stanceABOUT.md = background, doctrine, longer-form valuesAGENTS.md / CLAUDE.md = operating constraintsGOALS.md = explicit goal hierarchyA good SOUL.md should answer:
A bad SOUL.md reads like a README, policy manual, roadmap, or tool index.
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.
SOUL.md into ABOUT.md but shorter| Content | Belongs in |
|---|---|
| "Direct, opinionated, no corporate fluff" | SOUL.md (Voice) |
| "Prefers Unix philosophy, local-first tools" | SOUL.md (Taste) |
| "Always use absolute paths when saving files" | AGENTS.md / CLAUDE.md |
| "Uses gptodo for task management" | ARCHITECTURE.md or TASKS.md |
| "Final goal: play the longest possible game" | GOALS.md |
| "Was created by X on date Y" | ABOUT.md |
SOUL.md skeleton by defaultABOUT.md / AGENTS.md / CLAUDE.md / GOALS.md in the agent workspace — the
files SOUL.md deliberately defers toConduct a structured interview about a new project and generate a ready-to-use CLAUDE.md/AGENTS.md starter so the agent knows your conventions from session one.
Query a Home Assistant instance for presence, sensor data, calendar events, and cameras
Comprehensive framework for effective gptme agent onboarding that builds user trust, communicates capabilities clearly, and establishes productive working relationships from the first interaction.
Workflow for publishing HTML artifacts (demos, visualizations, interactive content) to the web. Enables agents to share work products publicly via GitHub Pages or similar hosting.
Systematic code review workflows with bundled utilities for analyzing code quality, detecting patterns, and providing structured feedback. Use this skill when reviewing pull requests or conducting code audits.
Template and guide for restructuring large documentation files into token-efficient directory structures. Reduces context bloat by 40-60% while maintaining accessibility.