Response Headers

The Stronghold proxy adds headers to every proxied HTTP response, regardless of whether the content was flagged. This allows AI agents and monitoring tools to programmatically inspect scan results without parsing response bodies.

Header Reference

Header	Description	Values
X-Stronghold-Decision	What the scanner found	ALLOW, WARN, BLOCK
X-Stronghold-Action	What the proxy did	allow, warn, block
X-Stronghold-Reason	Why content was flagged	Human-readable string
X-Stronghold-Score	Combined threat score. Present when a scan produced a combined or heuristic score. Omitted when no score was computed.	0.00 - 1.00
X-Stronghold-Scan-Type	Type of scan performed	content, disabled, skipped-unscannable, skipped-not-scannable, skipped-oversized
X-Stronghold-Warning	Warning message	Only present if action is warn
X-Stronghold-Request-ID	UUID for tracing	req-<hex>
X-Stronghold-Scan-Latency	Time spent scanning	e.g. 12ms

Decision vs Action

The Decision and Action headers can have different values. This is intentional and reflects the separation between detection and enforcement:

• Decision = what the scanner found. This is the raw detection result. BLOCK means the scanner identified a high threat score.
• Action = what the proxy did about it. This is controlled by your configuration. The proxy may be configured to allow content that the scanner would otherwise block.

Example: Flagged but Allowed

A response with these headers:

X-Stronghold-Decision: BLOCK
X-Stronghold-Action: allow
X-Stronghold-Score: 0.87
X-Stronghold-Reason: Prompt injection detected: instruction override attempt

This means the scanner detected a prompt injection (decision = BLOCK, score = 0.87), but the proxy configuration has action_on_block set to allow, so the content was passed through with headers attached. The agent can read these headers and decide how to handle the flagged content.

Example: Clean Content

X-Stronghold-Decision: ALLOW
X-Stronghold-Action: allow
X-Stronghold-Score: 0.02
X-Stronghold-Scan-Type: content

Low score, no threat detected. Content passed through normally.

Parsing Headers in Code

Python

import requests

response = requests.get("http://example.com")

decision = response.headers.get("X-Stronghold-Decision")
if decision == "BLOCK":
    reason = response.headers.get("X-Stronghold-Reason")
    print(f"Content flagged: {reason}")

JavaScript

const response = await fetch("http://example.com");

const decision = response.headers.get("X-Stronghold-Decision");
const action = response.headers.get("X-Stronghold-Action");

if (decision === "BLOCK" && action === "allow") {
  console.log("Flagged but allowed:", response.headers.get("X-Stronghold-Reason"));
}

Scan Type Values

The X-Stronghold-Scan-Type header indicates what kind of scan was performed:

Value	Meaning
content	Full content scan was performed (prompt injection detection)
disabled	Scanning is disabled in configuration
skipped-unscannable	Content type is not text-based (binary data)
skipped-not-scannable	Content was fetched but determined to be unscannable after inspection
skipped-oversized	Content exceeds the 1 MB size limit

When the scan type is skipped-unscannable, skipped-not-scannable, or skipped-oversized, the decision will be ALLOW and the X-Stronghold-Score header is omitted (not present) since no scan was actually performed.

HTTPS (MITM) Header Differences

When the proxy intercepts HTTPS traffic via MITM, the response headers are a reduced subset of the full set listed above. Present on MITM responses when a scan was performed. For unscannable or oversized content, only X-Stronghold-Proxy: mitm is set.

Header	Description
X-Stronghold-Proxy	Always set to mitm to indicate the response was intercepted via MITM
X-Stronghold-Decision	The scan decision (ALLOW, WARN, BLOCK) — only present when a scan was performed
X-Stronghold-Reason	Why content was flagged (only present when scanned content is flagged)

Note

In the current implementation, blocked MITM responses include X-Stronghold-Decision and X-Stronghold-Reason but may not include X-Stronghold-Proxy.

Headers such as X-Stronghold-Action, X-Stronghold-Score, X-Stronghold-Scan-Type, X-Stronghold-Request-ID, X-Stronghold-Scan-Latency, and X-Stronghold-Warning are only present on plain HTTP responses. If your application relies on these headers for decision-making, be aware that they will not be available for HTTPS traffic.