Agent Brief Template
Version 1.1 · Kramaniti Kosh
Intended outcome
An accountable agent brief with a bounded job and a clear handoff.
Before you begin
- One recurring job and its outcome
- Approved inputs and exclusions
- An owner and a receiving person or team
How to use
- Fill in the brief with the workflow owner before choosing tools or a model.
- Specify the output format, source boundary, stop conditions and receiving owner.
- Use the brief as instructions in your chosen environment and test it with a safe sample. Record that test before relying on it.
Working template
Role
What recurring job does this agent support? Name the business outcome, not a vague capability.
Inputs
- Approved source material
- Decision context
- Constraints and exclusions
Output contract
Describe the exact result, format, audience, and quality standard.
Boundaries
- What the agent must not decide or change
- What requires human review
- What sources it may use
- When it must escalate
Handoff
Name the next owner, the decision they need to make, and the evidence they should receive.
Demonstration
This is an illustrative scenario, not a client case study or a measured result. Do not reuse its details as facts about your work.
An editorial team wants help preparing a review brief from approved draft notes.
Sample inputs
Demonstration inputs only: approved draft notes are available; the editorial lead makes publication decisions; the publisher receives approved copy.
Example output
Job
Prepare an editorial review brief from approved draft notes.
Inputs
Only the attached draft and approved reference notes.
Output contract
Return a summary, unresolved claims and questions for the editorial lead.
Boundaries
Do not invent evidence, choose a publication date, edit live systems or publish. If references are missing, label the gap and stop the affected claim.
Handoff
[Name the editorial lead] reviews the brief. The publisher receives copy only after explicit approval.
Sample test
Include one unsupported claim in a safe sample. Expected behaviour: flag the claim and request evidence. Test result: not yet recorded.
Quality check
The job is specific; inputs and exclusions are explicit; the sample test includes a stop condition; the receiving owner is named.
Limits and human review
A well-written brief does not prove runtime reliability or grant access to tools. Test it in the intended environment. Keep consequential actions with the named human owner.
Edition notes
Version 1.1 adds setup guidance, an illustrative example and a review checklist. Provider-neutral Markdown instructions; no installation or platform compatibility is implied. Runtime behaviour has not been verified across AI providers. Reuse terms have not yet been published.