Skip to content

Your First Auditor

This tutorial walks you through the complete lifecycle of a Lucid Auditor: development, verification, publishing, and deployment.

Alpha Access Required

The Lucid SDK and CLI are available to alpha participants. Request access to get started.

The Scenario: Prompt Injection Detection

You need to ensure that no prompt injection attacks reach your AI model. You will build an auditor that intercepts requests and blocks those containing malicious injection attempts.

Step 1: Implement the Logic (SDK)

Create a file named main.py. We'll use the Lucid SDK to define a request-phase hook.

from lucid_sdk import AuditorApp, Proceed, Deny

# 1. Create the auditor app
app = AuditorApp("lucid-guardrails-auditor", port=8096)

# 2. Define the safety logic
@app.on_request
def check_injection(data: dict, config=None, lucid_context=None):
    prompt = data.get("prompt", "")

    # Scan for common injection patterns
    injection_patterns = [
        "ignore all previous instructions",
        "disregard the above",
        "system prompt:",
    ]

    prompt_lower = prompt.lower()
    for pattern in injection_patterns:
        if pattern in prompt_lower:
            return Deny(reason=f"Prompt injection detected: {pattern}")

    return Proceed(safety_score=1.0)

# 3. Run the auditor
if __name__ == "__main__":
    app.run()

Step 2: Containerize

Lucid Auditors run as sidecars. Create a Dockerfile:

FROM python:3.12-slim

# Create non-root user
RUN useradd -m -u 1001 appuser
USER appuser

WORKDIR /app
COPY --chown=appuser:appuser main.py .
RUN pip install --user lucid-sdk

# Required labels
LABEL io.lucid.auditor="true"
LABEL io.lucid.schema_version="1.0"
LABEL io.lucid.phase="request"
LABEL io.lucid.interfaces="health,audit"

CMD ["python", "main.py"]

Build the image:

docker build -t lucid-guardrails-auditor:v1 .Successfully built lucid-guardrails-auditor:v1

Step 3: Verify Compliance (CLI)

Before deploying, ensure your container meets the Lucid Auditor Standard (correct labels and health endpoints).

lucid auditor verify lucid-guardrails-auditor:v1[+] Basic labels found.
[+] Compliance probe successful!
[*] Verification complete. Auditor is compliant.

Step 4: Sign & Notarize

Register your auditor's cryptographic digest with the Lucid Verifier. This ensures only your authorized code can run in the TEE.

lucid auditor publish lucid-guardrails-auditor:v1Pushing image to registry...
Registering digest with Verifier...
[+] Auditor published and notarized.

Step 5: Define the Environment with Audit Chain

Create a file named my-env.yaml to define your agent with the custom auditor:

apiVersion: lucid.io/v1alpha1
kind: LucidEnvironment
metadata:
  name: secure-agent
spec:
  infrastructure:
    provider: gcp
    region: us-central1
  agents:
    - name: my-secure-agent
      model:
        id: meta-llama/Llama-3.3-70B
      gpu:
        type: H100
        memory: 80GB
      auditChain:
        preRequest:
          - auditorId: lucid-guardrails-auditor
            name: Injection Detection
            env:
              INJECTION_THRESHOLD: "0.8"
              INJECTION_BLOCK_ON_DETECTION: "true"

Step 6: Deploy

Deploy your environment:

lucid apply -f my-env.yamlCreating agent: my-secure-agent...
Created: agent-abc123

[+] Environment deployed successfully!
lucid status my-secure-agentAgent: my-secure-agent
ID: agent-abc123
Status: running
Model: meta-llama/Llama-3.3-70B
GPU: H100

Results

  1. Sidecar Injection: The Lucid platform automatically injects your lucid-guardrails-auditor into the agent workload.
  2. Enforcement: Every request to your model now passes through the injection detector.
  3. AI Passport: Every response includes a cryptographically signed passport proving the injection check was performed.
lucid passport show <passport-id>Passport ID: pass-001
Hardware Attested: true
Auditors:
- lucid-guardrails-auditor: PROCEED (safety_score: 1.0)

Next Steps