Generate Requirement Drafts

Generate Requirement Drafts

Generate draft requirements from Simulink models using the richest artifact the environment supports. All outputs are drafts requiring human review and approval before baselining.

When to Use

• Drafting or updating requirements from a Simulink model
• Creating .slreqx or .yaml requirements artifacts
• Establishing model-to-requirement traceability

When NOT to Use

• Importing requirements from an existing source of truth (ReqIF, DOORS, Word, Excel) — use Requirements Toolbox import APIs directly
• Writing or running tests — use the testing-simulink-models skill
• Free-form prose notes with no saved artifact
• User explicitly wants a different text format (markdown, Word)

Output Conventions

• Default file names: <Model>_Requirements.slreqx or <model>_requirements.yaml
• Stable IDs: REQ_<SYSTEM>_001, REQ_<SYSTEM>_002, … — use one consistent project-wide prefix with sequential integers across all requirements, even when they derive from different subsystems. Do not switch to per-subsystem prefixes or hierarchical decimal numbering (e.g. CB-1.1, TP-1.1).
• Follow repo conventions if .slreqx or .yaml files already exist
• When updating an existing artifact, preserve existing IDs — append new IDs, never renumber
• Every generated requirement must be marked as draft — use "draft" in Keywords (slreq) or status: Draft + keywords: [draft] (YAML)

Writing Good Requirements

Requirements must express behavioral intent (WHAT the system shall do), not restate model topology (HOW it is implemented). Put the WHY in Rationale. Use EARS (Easy Approach to Requirements Syntax) patterns:

EARS Patterns

Choose the pattern that best fits the model behavior:

Examples — Model Concepts to EARS Requirements

Rules

• Write requirements using EARS patterns — pick the pattern that matches the model behavior
• Use block/subsystem names (e.g. Saturation, BrakeLogic) only in Rationale, Description, or provenance notes — never in the requirement Summary. (A workspace parameter rendered as VarName (value) is required in the Summary and is not a block name.)
• When referencing a numeric value, include the workspace variable name and resolved value: VarName (value). If the model uses a literal with no variable, record the numeric literal directly
• Put the WHY in Rationale, not in the requirement statement
• One subsystem may support zero, one, or several behavioral requirements — don't force a 1:1 mapping
• For a Stateflow state or mode, separate entering it from being in it: the entry transition is event-driven (When <trigger>, … shall transition to <state>), but the behavior that holds throughout the mode is state-driven (While <state>, … shall <response>). When asked for a mode's behavior, prefer the While <state> form — do not fold it into a When entry statement

Backend Decision Gate

Choose once at the start, then stay on that path.

Probe (run via evaluate_matlab_code):

hasSlreq = ~isempty(which('slreq.new'));
if hasSlreq, try, slreq.find(Type="Link"); catch, hasSlreq = false; end, end
disp(hasSlreq)

Workflow

• Call model_overview, model_read, model_query_params, model_resolve_params as MCP tools — never pass their names to evaluate_matlab_code.
• Use evaluate_matlab_code only for MATLAB code, such as the backend probe and the slreq.* APIs in Path A.

1. Understand the model — use the model_overview and model_read tools to understand subsystems, interfaces, control logic.

2. Extract parameters — use the model_query_params / model_resolve_params tools for thresholds, gains, sample times that should appear in requirement text. Record both variable names and resolved values.

3. Build a capture table (backend-neutral):

   Id	ParentId	Summary	Description	SourceBlock	Rationale	ASIL	Priority	Keywords

4. Run the backend path (A or B below).

5. Review using the gate below.

Path A — Requirements Toolbox (.slreqx)

Use when probe succeeds. For full API patterns see references/slreq-patterns.md.

model = "CruiseControl";
load_system(model);
rs = slreq.new(model + "_Requirements");

% Add requirements with hierarchy — note behavioral "shall" statements
req = add(rs, Id="REQ_CC_001", ...
    Summary="If throttle command exceeds ThrottleMax (1.0) or falls below ThrottleMin (0.0), then the controller shall clamp the output to the valid range.", ...
    Description="Derived from CruiseControl/ThrottleCmd.");
req.Rationale = "Prevents invalid actuator commands that could damage the throttle body.";
req.Keywords = ["draft","auto-generated","control","safety"];

child = add(req, Id="REQ_CC_002", ...
    Summary="When brake pedal input exceeds BrakeThreshold (0.0), the controller shall disengage cruise control.", ...
    Description="Derived from CruiseControl/BrakeLogic.");
child.Keywords = ["draft","auto-generated","safety","brake"];

% Create traceability links: use block handle from SID (blk_X → X) for robustness
lnk = slreq.createLink(Simulink.ID.getHandle(model + ":5"), req);
slreq.createLink(Simulink.ID.getHandle(model + ":8"), child);

save(rs);
% Save all link sets
linkSets = slreq.find(Type="LinkSet");
for i = 1:numel(linkSets), save(linkSets(i)); end

Link rules:

• Source = model element, destination = requirement → link type is automatically Implement
• Prefer subsystem-level links; use block-level only when the requirement is block-specific
• Every model-derived requirement must have at least one traceability link
• The model must be loaded (load_system) before creating links with block paths

Path B — Structured YAML fallback

Use when Requirements Toolbox is unavailable. For schema and field rules see references/yaml-requirements.md.

1. Generate a <model>_requirements.yaml file from the capture table
2. Set status: Draft and include draft in keywords for every requirement
3. Self-review: all required fields set, asil/status/priority use valid values, IDs sequential, derived_from references valid IDs or is null
4. Validate:

python -c "import yaml, sys; yaml.safe_load(open(sys.argv[1])); print('OK')" path/to/requirements.yaml

Guardrails

Review Gate

Before finishing, verify (both backends):

• Every requirement has Id, Summary, and Description
• Every requirement is marked as DRAFT (via Keywords or status field)
• Summaries use EARS patterns, not block-topology restatements
• Numeric values include parameter name + value where a workspace variable exists
• IDs are stable and sequential; existing IDs preserved on regeneration

slreq path only:

• Model-derived requirements have a traceability link to the source model element
• Link direction is model element → requirement (not reversed)
• Subsystem-level linking preferred over block-level
• Requirement set and all link sets are saved

YAML path only:

• File parses without errors
• asil and priority use valid enum values (or Unset when unknown)

References

• references/slreq-patterns.md — Requirements Toolbox API cookbook
• references/yaml-requirements.md — Structured YAML schema and field rules
• assets/slreq_from_model.m — End-to-end slreq example
• assets/requirements.yaml — Structured YAML example

Copyright 2026 The MathWorks, Inc.