> ## Documentation Index
> Fetch the complete documentation index at: https://musicready.org/llms.txt
> Use this file to discover all available pages before exploring further.

# Security Considerations

> Security guidelines for MRS implementations.

See: [`/MRS-Specification-RFC#17-security-considerations`](/MRS-Specification-RFC#17-security-considerations)

## Input Validation

Parsers MUST validate:

| Limit                      | Recommended |
| -------------------------- | ----------- |
| Maximum document size      | 100 MB      |
| Maximum nesting depth      | 100 levels  |
| Maximum events per measure | 10,000      |
| Maximum spans              | 100,000     |
| Maximum measures           | 100,000     |

## Denial of Service Protection

Implementations SHOULD protect against:

* **Deeply nested structures**: Limit recursion depth
* **Very large measure numbers**: Validate integer bounds
* **Excessive span counts**: Limit span storage
* **Malformed UTF-8**: Validate string encoding
* **Circular references**: Detect reference cycles

## Information Disclosure

Working Set Envelopes may expose:

| Data                      | Risk                      |
| ------------------------- | ------------------------- |
| Source document hash      | Reveals document identity |
| Creator/modifier identity | Reveals authorship        |
| Agent names               | Reveals workflow          |
| Transaction history       | Reveals edit patterns     |

### Mitigation

Implementations SHOULD:

* Allow redaction of sensitive metadata
* Support anonymous agent identifiers
* Provide hash salting options
* Enable audit log access controls

## Operation Safety

### UUID Validation

Orchestrators MUST validate:

* UUIDs are well-formed UUIDv7
* Referenced UUIDs exist (or are valid tmp-ids)
* No UUID collisions in minting

### Scope Enforcement

Orchestrators MUST enforce:

* Operations within granted scope
* Operations within granted lanes
* No unauthorized measure access

### Transaction Integrity

Orchestrators MUST ensure:

* Atomic application (all-or-nothing or explicit partial)
* Rollback capability
* Audit trail integrity

## Agent Isolation

When running untrusted agents:

| Protection          | Description                   |
| ------------------- | ----------------------------- |
| Scope limiting      | Grant minimal scope           |
| Lane restriction    | Grant minimal lanes           |
| Operation filtering | Allow only needed op types    |
| Output validation   | Validate all ops before apply |

## Best Practices

### For Orchestrator Implementers

1. Validate all input before processing
2. Enforce resource limits
3. Log security-relevant events
4. Support authentication/authorization
5. Enable audit trail

### For Agent Implementers

1. Respect granted permissions
2. Don't attempt out-of-scope access
3. Handle rejection gracefully
4. Don't store sensitive data unnecessarily