Agent Templates System

OpenClaw Agent Templates system reference and workflow guide. Use when creating, editing, reviewing, debugging, or extending the built-in Agent Templates fea...

Security Scan

The skill is an internal documentation / workflow guide for OpenClaw's Agent Templates feature and its requested/declared footprint aligns with that purpose.

Purpose & Capability

The name/description match the included files: design notes, implementation map, schema, UI guidance, and an example template. There are no unrelated env vars, binaries, or install steps requested.

Instruction Scope

SKILL.md and the references explicitly instruct readers to inspect repository source files and describe the create-from-template flow that writes config, workspace files, identity files, and seeds memory. That behavior is coherent for a template materialization feature, but it does mean using the feature will cause local file and config writes (into the OpenClaw state dir and target workspace).

Install Mechanism

This is instruction-only with no install steps, downloads, or code to write to disk as part of skill installation.

Credentials

The skill requests no credentials or env vars. However template definitions support optional clawhubUrl entries (example uses https://clawhub.ai/...), which implies that creating an agent could pull or reference remote skills — a reasonable capability for this feature but one you should review before materializing templates that reference external endpoints.

Persistence & Privilege

always is false and there is no install that persists or auto-enables the skill. The documentation explains components that will modify OpenClaw config and write workspace files when materializing templates — expected for this feature and scoped to the agent being created.

Assessment

This skill is a documentation-only guide for the Agent Templates feature and appears coherent with that purpose. Before using the create-from-template flow in production: 1) review any template JSON (especially workspace.files and memorySeeds) so you know exactly what files will be written and where; 2) verify the product's path-traversal protections and that workspace paths resolve where you expect (the docs state this check exists, but confirm in code/runtime); 3) inspect any included skills that have clawhubUrl or other external references — understand whether materializing the template will fetch or execute remote code; 4) test in a sandbox/non-production environment to confirm config updates and file writes behave as expected; and 5) be cautious with raw JSON editing (parser errors and malformed definitions can lead to unexpected materialization).

Use this skill when working on OpenClaw's built-in Agent Templates feature.

What this system does

The Agent Templates system lets OpenClaw store reusable agent blueprints in SQLite, edit them in the Control UI, and materialize real agents from those blueprints.

A template can define:

• identity defaults
• included skills
• partial agent config
• workspace files
• memory seeds

From the UI, a user can:

• list templates
• create templates
• edit templates
• delete templates
• create a real agent from a selected template

Core workflow

1. Read references/implementation.md first.
2. If UI work is involved, also read references/ui-design.md.
3. If you need payload examples, read references/template-schema.md and inspect assets/examples/.
4. When changing behavior, keep storage, gateway methods, and UI state in sync.
5. When validating a bug or feature, test both:
   • template CRUD
   • agent creation from template

Source of truth files

Read these files in the OpenClaw repo when making code changes:

• src/agents/templates.ts
• src/gateway/server-methods/agent-templates.ts
• ui/src/ui/controllers/templates.ts
• ui/src/ui/views/templates.ts

Important behaviors

• Templates are stored in SQLite under the OpenClaw state dir.
• Template definitions are normalized before storage/use.
• Creating an agent from a template updates OpenClaw config, creates a workspace, writes identity/workspace files, and seeds memory.
• Included skills are merged with template config and create-time overrides.
• Workspace file paths are constrained to stay inside the target workspace.

Practical guidance

• Prefer small, composable template definitions.
• Put durable persona/role setup in workspace.files.
• Put starter memory in workspace.memorySeeds.
• Use definition.skills for reusable attached skills.
• Keep UI wording aligned with the underlying JSON schema.
• If the UI allows raw JSON editing, make parser errors user-legible.

Debugging checklist

• Can agentTemplates.list return the template?
• Does agentTemplates.get return the expected normalized definition?
• Does the UI render the selected template and preserve edits?
• Does agents.createFromTemplate create a non-conflicting agent id?
• Were SOUL.md, IDENTITY.md, AGENTS.md, and any seeded files written as expected?
• Did config get updated with workspace, identity, and merged skills?

References in this skill

• references/implementation.md — architecture and code map
• references/template-schema.md — practical schema notes and field behavior
• references/ui-design.md — Control UI layout and UX notes
• assets/examples/agent-template-definition.json — example template payload