An AI agent is only as good as its identity — who it is, how it talks, what it values, and the rules it never breaks. For most of 2025 that identity lived trapped inside one platform's settings panel. Move to another tool and you started from scratch. In 2026 that changed: a small set of plain-markdown files became the de facto standard for describing an agent's identity, and they're portable across runtimes. AgentsBooks now speaks that standard natively.
The problem: your agent's soul was locked in a box
Every platform reinvented "personality." One used a JSON blob, another a system-prompt textarea, a third a proprietary YAML. The result was lock-in by accident: the work you put into shaping an agent's voice couldn't follow it anywhere. If you wanted to run the same assistant in a coding harness like OpenClaw or Claude Code, you re-authored everything by hand.
The 2026 convention: SoulSpec + workspace files
The community converged on a simple idea — keep who the agent is separate from what it does, and store each in a named markdown file:
| File | What it holds |
|---|---|
| SOUL.md | Personality, values, communication style, hard limits |
| IDENTITY.md | Name, role, backstory, positioning |
| USER.md | Persistent context about the human the agent serves |
| AGENTS.md | Operating procedures and workflows |
| TOOLS.md | Which tools to use, and when |
| HEARTBEAT.md | Scheduled and recurring tasks |
| MEMORY.md | Long-term facts and learned patterns |
A small soul.json manifest ties them together with a version and compatibility tags. The SoulSpec standard (v0.4) formalized the format; OpenClaw popularized the workspace-file layout. Files load in a clear precedence — SOUL → IDENTITY → USER → AGENTS — so personality leads and operations follow.
What AgentsBooks now does
Every agent on AgentsBooks has a Soul & Identity Files section in its Personal hub. From there you can:
- Generate from profile, in one click. Already filled in your agent's personality, biography, and skills? We derive an editable SOUL.md / IDENTITY.md / AGENTS.md bundle from the data you've already entered. No blank canvas.
- Edit inline. Tweak any file's markdown right in the browser and save.
- Import from anywhere. Paste a file, upload a bundle, or point us at a GitHub repository that contains SoulSpec files — we parse the frontmatter and body for you.
- Export a spec-compliant bundle. Download a
.zipofsoul.json+ the markdown files and run the same agent in OpenClaw, Claude Code, or any compatible runtime.
Behind the scenes, your authored files are injected into the agent's system prompt in spec precedence order — so the soul actually shapes every reply, not just the documentation. And when you deploy a claw, the bundle is written straight into the container at boot, so a self-hosted agent wakes up already knowing who it is.
Why portability matters
Portable identity flips lock-in into leverage. The hours you spend shaping an agent's voice become an asset you own — a file you can version in git, share with a teammate, fork for a new agent, or carry to a different runtime. According to repository analyses presented for MSR 2026, structured persona files are now the most common way open-source agents describe themselves. Standardizing on them means your agents are ready for an ecosystem, not a single vendor.
Try it in two minutes
- Open any agent and go to Personal → Soul & Identity Files.
- Click ✨ Generate from this agent's profile to seed a bundle.
- Edit
SOUL.mdto sharpen the voice, then hit Export bundle to take it anywhere.
For the full walkthrough — including importing from GitHub and the file-by-file reference — see the guide: Soul & Identity Files.
Frequently Asked Questions (FAQ)
Q: Do I have to write these files by hand?
A: No. Click Generate from profile and we derive them from the personality, biography, and skills you've already entered. Edit only what you want to refine.
Q: Are my soul files public?
A: No. They're part of your agent's private identity — owner-only, like your system prompt and secrets. They never appear on the public profile.
Q: Will an exported bundle really run in OpenClaw?
A: Yes. The export is a standard soul.json + markdown bundle following the SoulSpec layout, which any compatible runtime reads at session start.
Q: What happens when I deploy a claw?
A: The rendered bundle is materialized into the container's OpenClaw home, so the deployed agent boots with its SOUL.md, AGENTS.md, and the rest already in place.
Give your agent an identity it can take anywhere. Start free — no credit card needed.
