Black box and grey box testing

	Black box	Grey box
What BotGauge sees	Messages and replies only	Also tool calls, traces, and backend state. It can place test content in documents, tool results, or memory.
Setup	Just the endpoint	Add the BotGauge wrapper, and optionally a verifier
Rules tested	Fewer. Rules needing inside access are listed as skipped, with the reason.	All applicable rules
Strength of findings	Based on what the agent says	Can be confirmed by hard evidence, so severity can be higher

Start with black box and add grey box when you want more. BotGauge counts a run as grey box only when the inside access has actually been proven to work during the connection test. Declaring a verifier is not enough on its own. If you choose black box explicitly, inside access is ignored.

What the wrapper is

The wrapper is a small endpoint you add to your agent. BotGauge sends it a turn message, and it answers with the agent’s reply plus extra observations.

Piece	What it does
Turn endpoint	Receives a `botgauge.turn/1` message (`{protocol, session_id, message}`) and returns a `botgauge.observation/1`: the reply, optional `events` such as tool calls, and `plant_receipts`.
Tools endpoint	Lists the tools your agent has and, if you set it, whether each is read-only. A tool with no declared action is treated as possibly writing, to stay safe.
Planting	In grey-box tests, BotGauge can place content in retrieval, tool output, or memory. It carries a unique marker called a canary. If the agent’s reply shows it used the marker, BotGauge knows the planted content influenced it.

Set it up

Choose the wrapper preset

In the connection form, choose the BotGauge wrapper preset, or copy the setup prompt BotGauge provides and give it to your coding assistant. It generates the wrapper for your stack.

Declare channels and a verifier

Declare which channels you support (retrieval, tool output, memory, system) and any trace or verifier.

Test the connection

Run Test connection. The result shows which capabilities are connected, declared, failed, or not configured.

Rerun recon

Rerun recon so the profile includes the new visibility.

What grey box unlocks

Rules that need to see tool calls, such as no duplicate actions and arguments matching the request
Indirect-injection tests that plant instructions in documents or tool output
Higher-severity, hard-evidence confirmation
Backend checks through a verifier, for example confirming that a refund was or wasn’t created

Unlock deeper testing

Set up the BotGauge wrapper for tool calls and backend checks in BotGauge.

Add the wrapper →

Get started

Run an assessment

Read the results

Frameworks and access

Help

Black box and grey box testing

What the wrapper is

Set it up

What grey box unlocks

​What the wrapper is

​Set it up

​What grey box unlocks

What the wrapper is

Set it up

What grey box unlocks