Sheet 07 · ART-14
Prompt and agent record
A record for any prompt, agent, or automation that colleagues will rely on — what it does, what it assumes, how it was tested, and who maintains it. Reusable resources are only reusable if someone can find and trust them.
Artifact details
- Format
- Register
- Owner
- The author, with the AI Program Manager maintaining the register
- When it is used
- When a prompt or agent is shared beyond its author.
- Retention
- Until retired, plus one year.
Contributors: AI champions, Data & Reporting, IT & Security for anything with system access
Downloads
The blank template is a standalone register with every field and its guidance, ready to fill in or print. Printing uses this page’s print stylesheet, which drops the navigation and the download controls.
Required fields
The guidance matters more than the field name. A field labelled “risks” with no guidance gets filled in with “standard AI risks” and the artifact stops being worth reading.
| Field | Guidance | Required |
|---|---|---|
| Name and purpose | What task this does, in one line. | Required |
| Owner and maintainer | A named person. Unowned prompts rot. | Required |
| Tool and version | Where it runs and against what. | Required |
| Inputs and assumptions | What it expects, and what it silently assumes. | Required |
| Known failure modes | What it gets wrong. The most valuable field for anyone reusing it. | Required |
| How it was tested | Test set, sample size, error rate observed. | Required |
| Review status and last reviewed | Prompts drift as models change. | Required |
| Data classification permitted | The ceiling for input to this prompt. | Required |
Worked example
Filled in against the sample record set, so the template can be judged on what a real entry looks like rather than on its headings.
PR-014 — Specification difference summary
- Purpose: summarize differences between a project specification section and the corresponding firm standard, for engineer adjudication.
- Owner: Alan Petrov, QA Lead. Tool: Cadence Search, standards corpus. Data ceiling: confidential.
- Inputs: one specification section and the standard reference. Assumes the standard in the library is the current revision — which is not always true, and is the source of most errors.
Testing and known failure modes
- Tested against 26 reviews. False-positive rate 31%, above the 20% threshold set for the pilot.
- Known failure modes: flags differences against superseded revisions that were never retired from the library; misses conflicts expressed as cross-references rather than direct text.
- Review status: in pilot. Last reviewed 2026-08-18. Do not reuse outside PIL-002 scope without talking to the owner.
Blank template
The same fields with nothing in them. Print this page, or use the download above to get a standalone file.
Name and purpose
What task this does, in one line.
Owner and maintainer
A named person. Unowned prompts rot.
Tool and version
Where it runs and against what.
Inputs and assumptions
What it expects, and what it silently assumes.
Known failure modes
What it gets wrong. The most valuable field for anyone reusing it.
How it was tested
Test set, sample size, error rate observed.
Review status and last reviewed
Prompts drift as models change.
Data classification permitted
The ceiling for input to this prompt.