> ## 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.

# Red teaming glossary and quick reference

> Definitions of BotGauge red-teaming terms such as recon, run pin, held, signal, canary, and verifier, plus the journey, results, and limits at a glance.

## Glossary

| Term | Meaning |
| - | - |
| Agent | The AI system you are testing |
| Probe, connection test | The quick check that BotGauge can reach your agent and read its reply |
| Recon | The discovery chat where BotGauge learns what your agent does |
| Profile | The saved result of recon |
| Rule | A statement of what your agent must or must not do |
| Plan | The attacks that will run for the selected rules |
| Approval | Your sign-off on a specific plan, with limits and an expiry |
| Run | One execution of an approved plan |
| Run pin | The frozen record of everything a run depended on, so results are reproducible and comparable |
| Prepared round, adaptive round | Fixed library attacks, and AI-driven attacks that adapt to replies |
| Attempt | One try of an attack. Attempts are never edited afterwards. |
| Finding | A recorded problem from a confirmed breach or a signal |
| Held | The agent resisted the attacks. The only passing result. |
| Signal | Suspicious but not proven |
| Not evaluable | Couldn't be tested here, for example because of missing access |
| Black box, grey box | Testing from outside only, and testing with inside visibility |
| Wrapper | A small endpoint you add to your agent for deeper visibility |
| Verifier | A way for BotGauge to read backend state to confirm what an action really did |
| Canary | A unique marker BotGauge plants to prove the agent used planted content |
| Verify fix | Replaying a finding's attack to confirm the fix |
| Crosswalk | The table that maps rule tags to framework requirements |

## Quick reference

**The journey.** Connect, Test connection, Recon, Review, Approve, Start, Watch, Read results, Fix and Verify, Compliance.

**Results.** Confirmed finding, Signal, Held, Couldn't verify, Not evaluable, Blocked, Error.

**Compliance item statuses.** Finding, Signal, Partial, Tested no finding, Unresolved, Not assessed.

**Severity.** Critical, High, Medium, Low. Medium at most when confirmed only by the reply or the AI judge, and Low for signals.

**Run verdicts.** Breached, Completed, Inconclusive, Error, Cancelled.

**Limits.** Up to 500 attempts per run, 8 messages per conversation, 3 system prompts per agent, approvals default to 60 minutes (maximum 7 days), and 30 connection tests per hour per project.

Remember that only Held is a pass, that not evaluable is a gap and not safety, and that a clean report is a statement about the tested rules and not a guarantee.


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