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

# Recon: let BotGauge learn your agent

> Recon is BotGauge's discovery chat that learns your agent's tools, data sources, and behavior. Covers depth, questions, context you can add, and reruns.

Recon is a short discovery chat. BotGauge asks your agent probing questions to learn what it is and what could go wrong. It usually takes under a minute, and you can watch the questions appear live as "Asking..." turns into "Answered".

## What BotGauge learns

| Area | What it learns |
| - | - |
| Tools | What the agent can do, and which tools change things (payments, deletes, bookings) and so are riskier |
| Data sources | Whether it reads documents, uses memory, or hands off to other agents |
| Behavior | How it refuses, what policies it states, and hints about its instructions |
| Mismatches | Where what you declared differs from what it really does |

The result is a saved profile of your agent. It is kept as a snapshot, so later runs can say exactly which version of the profile they used.

## Reuse and re-running

If you already ran recon on this exact connection with the same depth and nothing new was added, BotGauge shows the saved profile instantly instead of running it again. Use **Rerun recon** (or rerun in the API) after you change your agent's prompt, tools, or behavior.

Recon depth is **Fast**, **Intelligent** (the default), or **Deep**. Deeper recon asks more questions and takes longer.

## Add context

The more BotGauge knows, the more precisely it tests. You can add the following.

| Context | What to provide | Why it helps |
| - | - | - |
| System prompt (up to 3 per agent) | The instructions your agent runs on, pasted as text. Remove API keys first. Anything that looks like a secret is rejected and not stored. | Lets BotGauge test whether the agent leaks it or can be talked out of it. |
| Tool list or API spec (OpenAPI, MCP) | An OpenAPI file or MCP tool list, or a plain list of tool names. Mark each tool as read, write, delete, or execute. | Tells BotGauge which tools exist and which can change data. A tool with no marked type is treated as possibly writing. |
| Manifest (YAML) | One YAML file describing the agent: its tools, scope, and policies. | Declares everything in one place instead of several uploads. |
| Product policy | A plain-language rule your agent should follow, for example "refunds above 100 need approval". | Lets BotGauge check that the agent follows it. |

> [!NOTE]
>
> Everything you add is checked first. Anything that looks like a secret is rejected and never stored. If you add new context after the last recon, BotGauge runs a fresh recon so it is taken into account.

## Suggested rules

Recon may suggest extra rules based on what it saw, for example "never refund to a different account". Suggestions are saved as proposals and don't run until you confirm them in the [next step](https://www.botgauge.com/docs/red-teaming/rules-and-plans.md).

## What can go wrong

| Symptom | Meaning | Action |
| - | - | - |
| Spinner never ends | The live connection dropped. | Reload. The dashboard wizard re-fetches the saved recon. |
| "Recon couldn't finish" | The agent failed or timed out during recon. | Run the connection test again, then retry. |
| Few or no tools found | The agent exposes no tools, or tool calls aren't visible. | Declare tools yourself, or add the wrapper for visibility. |

Let BotGauge learn your agentStart the discovery chat and add context in [BotGauge](https://cloud.botgauge.com).

[Start recon →](https://cloud.botgauge.com)


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.