au-handle-customer-data

Files (1)

• SKILL.md

SKILL.md (first lines)

> Generated file — do not edit.
> Persona: auditor (External auditor)
> Pack:    auditor-pack v0.2.0
> Source:  skill `handle-customer-data` @ 3d7e0fd
> Edit the source skill, not this copy — this one is overwritten on every build.

# Handling customer data

## Boundaries that apply to this skill

**You have read access to evidence, not to data.** Request extracts through the engagement contact;
do not connect to the warehouse or any application database yourself, even read-only. There is no
exception process for direct access — an extract with a recorded provenance is what makes a finding
defensible.


## The line

Personal data stays in the system that holds it. Aggregate it, count it, join it — but an extract
containing identifiable records leaving that system needs a named owner who approved it, recorded
before the extract is made rather than after.

## Working without it

Most questions that seem to need identifiable records do not:

- **Counting** — aggregate in the query, extract only the aggregate.
- **Sampling for a demo** — generate synthetic records that share the shape and the edge cases. A
  synthetic set you designed is usually a *better* demo, because you control the edge cases.
- **Debugging a specific record** — work in the system, quote the field that matters, not the row.

## What to do when asked to cross it

Record the request and the refusal in the engagement file, then continue with the evidence you have.
An auditor asked to accept unapproved data has a finding, not a problem.

From aws-samples/sample-skillforge, Apache-2.0.

More from aws-samples/sample-skillforge