> ## Documentation Index
> Fetch the complete documentation index at: https://omniloy.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.
# Guardrails and red flags
> The agent's clinical safety limits and detection of critical situations
OlivIA applies two clinical safety mechanisms: **guardrails** (what the agent must never do) and **red flags** (critical situations that trigger immediate escalation).
## Guardrails
Guardrails are prohibited agent behaviors. Categories:
| Category | Prevents the agent from… |
| ---------------------------- | ------------------------------------ |
| `no_diagnosis` | giving a diagnosis |
| `no_medication` | recommending or adjusting medication |
| `no_prognosis` | issuing a prognosis |
| `no_contradicting_clinician` | contradicting the clinical team |
Each guardrail has a **severity** (`warning` or `critical`) and optional guidance. If a protocol configures none, a **safe default set** applies (no diagnosis and no medication as critical; no prognosis and no contradicting the clinician as warning).
**How they are enforced — two layers:**
1. **Preventive:** the rules are injected into the agent's instructions as inviolable rules that override the protocol context and any patient request.
2. **Detective (judge):** before speaking, a judge evaluates the phrase the agent is about to say against each category. If it detects a violation, **the phrase is blocked**. The system is *fail-safe*: if the judge is unavailable, the phrase is treated as unsafe and not spoken.
## Red flags
A **red flag** is a critical clinical criterion that, when detected on the call, triggers an immediate response. It is defined per protocol with a **criterion** and a **transfer phone** (up to 10 per version). Declaring a criterion is what enables its detection: if a protocol declares none, the agent has no ability to escalate by red flag.
Common suggested criteria: suicidal ideation, severe chest pain, severe breathing difficulty, severe bleeding — plus any custom criteria you define.
**What happens when a red flag is detected:**
1. The alert is logged and **escalated immediately** to the clinical team.
2. The run is closed as **unresolved**.
3. The agent says the `red_flag.reassurance` phrase (if configured).
4. The call is **transferred** to the criterion's phone (or the organization's emergency phone). If there is no number or the transfer fails, the agent closes the call safely, without medical guarantees.
```mermaid theme={null}
sequenceDiagram
autonumber
participant A as Agent
participant C as Clinical team
participant P as Patient
A->>C: Log and escalate the alert
A->>A: Close the run as unresolved
A->>P: Reassurance phrase
A->>P: Transfer the call
```