requirements-decomposerlisted
Install: claude install-skill YevheniiaQA/qa-agent-toolkit
# Requirements Decomposer
Test scenarios written against prose are untraceable. This skill converts prose
into numbered, atomic requirements that scenarios can point at — which is what
makes coverage measurable at all.
## The decomposition rule
A requirement is atomic when it can be verified true or false by a single check.
"The user can manage integrations" is not atomic. "Deleting an integration
returns 204 and removes it from GET /integrations" is.
Split by layer as well as by story: a User Story describes an end-to-end user
outcome; a Task describes one component's behaviour (an endpoint, a form, a job).
Scenarios generated later follow the test pyramid off this split — stories become
end-to-end tests, tasks become component/API tests.
## ID scheme
Stable IDs are the backbone of traceability. Never renumber an existing ID; append.
```
F-<TICKET>-<LAYER>-<NN> functional requirement F-DEVAI-283-BE-01
NF-<TICKET>-<LAYER>-<NN> non-functional requirement NF-DEVAI-283-BE-01
AC-<TICKET>-<NN> acceptance criterion AC-DEVAI-283-01
```
`LAYER` ∈ `BE` | `FE` | `E2E` | `INT`.
## Output structure
```
02-analysis/<TICKET-ID>/requirements/
├── US-001.md User Story + AC
├── TASK-001.md Backend task + FR/NFR
├── TASK-002.md Frontend task + FR/NFR
├── OPEN_QUESTIONS.md Everything the source did not answer
├── NOTES-classification.md Why things were split the way they were
└── _index.json M