version: "1.0.3" name: checking-model-compliance description: "Use this skill when the user asks to check Simulink model compliance against a standard (MISRA, MAB, JMAAB, ISO, DO, IEC, EN, CERT C/CWE, AUTOSAR, Simulink Code Inspector (SLCI)), wants to run Model Advisor checks, or needs a compliance report with fix suggestions. For JMAAB/MAB, supplement deterministic checks with agentic review of uncheckable guidelines." license: https://www.mathworks.com/content/dam/mathworks/license/pmrl/license.md metadata: author: MathWorks version: "0.6"

Checking Model Compliance

Runs Model Advisor checks for a named standard (or default configuration) and delivers a prioritized summary with fix suggestions.

When to Use

• Checking whether a Simulink model complies with a standard (MISRA, MAB, JMAAB, ISO 26262, DO-178C, etc.)
• Running Model Advisor checks against a named compliance standard
• Generating a compliance report with prioritized findings and fix suggestions
• Running a custom Model Advisor configuration file against a model
• Justifying or waiving specific compliance violations

When NOT to Use

• Building or editing model structure → building-simulink-models
• Writing behavioral tests → testing-simulink-models
• Structural validation only (unconnected ports) → model_check tool directly
• General model quality questions without a named standard (e.g., "is my model well decomposed?") — this requires a different workflow

Supported Standards

Equivalent check sets (run once, report for both):

• ISO 26262, IEC 61508, IEC 62304, EN 50128/EN 50657, ISO 25119
• DO-178B, DO-178C, DO-331

Custom Checks and Configurations

Users may have custom Model Advisor checks or custom configuration files (.json exported from Model Advisor Configuration Editor). These are not standards — they are handled via:

• Custom configuration file: User provides a path → use model_advisor_run with 'configuration' parameter directly (Path B)
• Custom check IDs: User provides specific check IDs → use model_advisor_run with 'checks' parameter directly (skip resolution)

Prerequisites

All script functions are in the skill's scripts/ directory. Use evaluate_matlab_code with project_path set to the skill's scripts/ folder so MATLAB can find them. Never use addpath.

Tools provided (always use these — never improvise with raw Model Advisor API):

Workflow

1. Identify Standard, Model, and Scope

Determine from user request:

• Standard — map to supported name (see table). Defaults: "MISRA" → MISRA_C. For "JMAAB" without version → ask user (v5.1 or v6)
• Model — .slx file (ask if ambiguous)
• Scope — full model (default) or subsystem path (e.g., Model/Controller)

Disambiguation rules:

• "JMAAB" without version specifier → ask (v5.1 or v6 — two distinct check sets)
• "ISO" without specifier → ask (multiple supported)
• "DO" without specifier → ask (178C vs 254)
• Multiple standards requested → resolve each, compare sets, run once if identical

2. Choose Path

Path A — Standard named: Proceed to step 3.

Path B — No standard, user says "run Model Advisor" / "check my model": Run detect_default_config(modelName). If config found → skip to step 4. If empty → ask which standard (show supported list).

3. Resolve Checks (Path A only)

model_advisor_resolve_checks('standard', '<STANDARD_NAME>')

• status: success → note checks_count, inform user
• status: truncated → use returned config file path in step 4
• status: error → report and stop

Gate: If checks_count > 100, confirm with user before proceeding.

Shortcut: If the user already has specific check IDs, call model_advisor_run directly with those IDs and skip to step 5.

4. Run Checks

model_advisor_run('<system>', 'checks', {<check_ids>})          % inline checks
model_advisor_run('<system>', 'configuration', '<config_path>') % config file

Use 'token_budget', 8000. If truncated, read full results from full_results path.

5. Analyze Findings

Classify from YAML response:

1. Critical (Failed) — must fix for compliance
2. Warnings (Warning) — should fix, may be justifiable
3. Informational — low priority

6. Present Compliance Report

For non-MAB/JMAAB standards, use this template:

## Compliance Summary: <Standard>
Model: <model> [Scope: <subsystem> if scoped]
Result: X passed, Y warnings, Z failures
 
### Critical Findings (must fix)
| Check | Blocks | Fix Type | Action |
|-------|--------|----------|--------|
| name  | N      | param/insert/config/routing/arch | what to change |
 
### Warnings (should fix)
| Check | Blocks | Fix Type | Action |
|-------|--------|----------|--------|
 
### Passed
N checks passed.
 
### Suggested Next Steps
[5-7 prioritized actions max]

For MAB, JMAAB, or JMAAB_V6 standards, use this template instead (includes the uncheckable guidelines review as an integral part of the report):

## Compliance Summary: <Standard>
Model: <model> [Scope: <subsystem> if scoped]
Result: X passed, Y warnings, Z failures
 
### Critical Findings (must fix)
| Check | Blocks | Fix Type | Action |
|-------|--------|----------|--------|
| name  | N      | param/insert/config/routing/arch | what to change |
 
### Warnings (should fix)
| Check | Blocks | Fix Type | Action |
|-------|--------|----------|--------|
 
### Passed
N checks passed.
 
---
 
### Agentic Review of Uncheckable Guidelines (LLM-based — not deterministic)
 
> **Note:** This is a semantic evaluation performed by an AI agent, not a deterministic Model Advisor check.
 
**Model:** <model_name>
**Judgment:** PASS | WARNING | FAIL
**Confidence:** 0.XX
 
#### Critical (must fix)
| Guideline | Blocks | Fix Type | Action | Confidence |
|-----------|--------|----------|--------|------------|
 
#### Warnings (should fix)
| Guideline | Blocks | Fix Type | Action | Confidence |
|-----------|--------|----------|--------|------------|
 
### Evidence
[2-3 sentences citing specific model data]
 
### Recommendation
[Top 3-5 prioritized actions, or "No action required."]
 
---
 
### Suggested Next Steps
[5-7 prioritized actions max, combining both deterministic and uncheckable findings]

Fix Type values: param (block parameter), insert (add block), config (model config), routing (reconnect signals), arch (restructure — recommend only)

Conciseness rules:

• One line per check; max 3-5 block paths shown per check (state total)
• Target 40-60 lines; max ~80
• Offer "I can list all affected blocks for check X" for detail

7. Performing the Uncheckable Guidelines Review (JMAAB/MAB only)

This section explains HOW to fill in the "Agentic Review of Uncheckable Guidelines" section that is part of the MAB/JMAAB report template above. You MUST complete this before presenting the report to the user.

Skip this step ONLY for other standards (MISRA, ISO 26262, DO-178C, etc.). For JMAAB, JMAAB_V6, or MAB standards → load and follow references/uncheckable-guidelines-review.md.

What to do: Evaluate guidelines that Model Advisor cannot fully verify — guidelines where semantic judgment is required or no check exists at all. Load references/semantic-evaluation-index.md to get the list of uncheckable guidelines, then:

1. From the Model Advisor results, identify which checked guidelines appear in the semantic evaluation index
2. For those guidelines, extract model data using MCP tools (model_overview, model_read, model_query_params)
3. Apply the Quick Rules from the index for each applicable guideline
4. For borderline cases, use the Semantic Gap column to calibrate judgment

Judgment: worst verdict across findings (FAIL > WARNING > PASS). Confidence 0.0-1.0. Do NOT issue FAIL if confidence < 0.5.

8. Fix Mode

Ask: "Would you like me to fix these issues?"

• No → stop (report only)
• Yes → load and follow references/fix-mode.md workflow

9. Justification Mode

If user wants to justify/waive/suppress violations → load and follow the Justification section in references/fix-mode.md.

Guardrails

Always

• For JMAAB/JMAAB_V6/MAB: ALWAYS perform Step 7 (Uncheckable Guidelines Review) — the report is incomplete without it
• Show standard name + check count before running
• Include block paths in findings
• Follow Fix Order strictly (structural → diagnostic) to prevent cascading false failures
• Summarize, do not echo raw output
• Confirm with user before executing >=100 checks
• Ask the user for justification text before adding any justification — never fabricate rationale
• If a reference tool returns an error, report it verbatim — do not retry with alternative approaches
• Confirm which model to check if multiple .slx files are present or the name is ambiguous
• State the resolved standard name and version in the report header
• Report the exact check count from tool output
• When explaining failures, list all distinct root causes

Ask First

• Fix mode modifications — never modify model without per-batch confirmation
• Justification — always a human decision
• Running >100 checks — confirm scope is intentional

Never

• Claim "model IS compliant" — only report pass/fail; compliance determination is user's responsibility
• Escalate diagnostics before structural fixes
• Suppress findings without explicit request
• Guess parameter names — use exact parameter field from check output
• Modify the MATLAB path permanently (no savepath)
• Run slbuild or code generation without explicit user permission
• Re-run checks unnecessarily — reuse violation IDs from the most recent run if the model has not been modified since; only re-run to get fresh hashes if the model changed

Error Recovery

Copyright 2026 The MathWorks, Inc.