Global

Methods

chatgpt(bot)

Register the Mineflayer ChatGPT plugin on a bot instance.
Source:
Parameters:
Name Type Description
bot object Mineflayer bot instance.

checkConfidenceScore(confidenceScore, minimumConfidenceScore, fallbackMessage) → {object}

Check if the confidence score is below the minimum threshold. If it is, returns a fallback message and flagged status. Otherwise, returns flagged false. LLM09 - Misinformation
Source:
Parameters:
Name Type Description
confidenceScore number Inferred confidence score.
minimumConfidenceScore number Minimum allowed confidence.
fallbackMessage string Fallback reply.
Returns:
Type:
object
Confidence check result object.

checkLastMessageCoolDown(memory, player, coolDownInSeconds, fallbackMessage) → {object}

Check whether the player's last message is outside the configured cooldown. Returns flagged=true with fallback when the cooldown has not elapsed yet. LLM10 - Unbounded Consumption
Source:
Parameters:
Name Type Description
memory object Per-player memory store.
player string Player name or id.
coolDownInSeconds number Minimum elapsed seconds.
fallbackMessage string Fallback message returned on cooldown hit.
Returns:
Type:
object
Cooldown result object.

detectJailbreakAttempt(message) → {boolean}

Detect whether a message attempts to jailbreak instruction boundaries. LLM01 - Prompt Injection
Source:
Parameters:
Name Type Description
message string Message text to inspect.
Returns:
Type:
boolean
True when a jailbreak attempt is detected.

detectPromptLeakage(message) → {boolean}

Detect whether a message contains any security instruction text. LLM07 - System Prompt Leakage
Source:
Parameters:
Name Type Description
message string Message text to inspect.
Returns:
Type:
boolean
True when the message leaks security instructions.

detectSecretsCredentials(message) → {boolean}

Detect whether a message contains possible secrets or credentials. LLM02 - Sensitive Information Disclosure
Source:
Parameters:
Name Type Description
message string Message text to inspect.
Returns:
Type:
boolean
True when secret-like content is detected.

detectSlashCommand(message) → {boolean}

Detect whether a message contains a Minecraft-style slash command. LLM05 - Improper Output Handling
Source:
Parameters:
Name Type Description
message string Message text to inspect.
Returns:
Type:
boolean
True when a slash command is present.

(async) moderateInboundReply(openAIClient, reply, fallbackMessage, confidenceScoreopt, minimumConfidenceScoreopt) → {Promise.<object>}

Sanitise and moderate an inbound reply before it is sent to the player.
Source:
Parameters:
Name Type Attributes Default Description
openAIClient object Client instance that can call OpenAI moderation.
reply string Model reply.
fallbackMessage string Fallback message.
confidenceScore number <optional>
 1 Reply confidence score.
minimumConfidenceScore number <optional>
 0 Minimum accepted confidence.
Returns:
Type:
Promise.<object>
Moderated inbound reply object.

(async) moderateOutboundMessage(openAIClient, memory, player, message, fallbackMessage, coolDownInSecondsopt) → {Promise.<object>}

Sanitise and moderate an outbound message before it is sent to the model.
Source:
Parameters:
Name Type Attributes Default Description
openAIClient object Client instance that can call OpenAI moderation.
memory object Per-player memory store.
player string Player name or id.
message string Outbound message.
fallbackMessage string Fallback message.
coolDownInSeconds number <optional>
 15 Minimum seconds between messages.
Returns:
Type:
Promise.<object>
Moderated outbound message object.

sanitiseProfanity(message) → {string}

Sanitise profanity in a string.
Source:
Parameters:
Name Type Description
message string Message text.
Returns:
Type:
string
Sanitised message.