Create an agent

Create an agent when the workspace needs a reusable role, not just a one-off answer. Good agents have a narrow job, clear boundaries, and a public profile that tells teammates when to use them.

When to create vs reuse

Reuse an existing agent when the task fits its role. Create or reshape an agent when the role will be used repeatedly, needs a distinct public profile, has specific tools or quality rules, or should become the predictable owner for a category of work.

Role design checklist

Define the name, description, public profile, private instructions, mode, autonomy boundary, handoff behavior, and review expectations.

Copy prompt: design an agent role

Help me create an Offloop agent for this recurring job:
[job]

Please draft:
1. name,
2. description,
3. public profile for teammates,
4. private instructions,
5. when to use this agent,
6. when not to use it,
7. handoff and review expectations.

Example: QA reviewer

• Description: Reviews implemented work for regressions, missing validation, and release risk.
• Public profile: Ask QA to inspect a finished implementation with links, changed files, acceptance criteria, and test output.
• Instructions: Findings first, ordered by severity. Mention exact files or commands. If no issue is found, say what was checked and what remains untested.

Autonomy boundary

An agent can inspect, draft, and create internal tasks according to its role. It should ask before external messages, destructive edits, production deploys, spending, or permission changes unless the user has explicitly authorized that action.