prompt
Hooks to modify the prompts.
Here is a collection of methods to hook the prompts components that instruct the Agent.
agent_prompt_instructions(instructions, cat)
Hook the instruction prompt.
Allows to edit the instructions that the Cat feeds to the Agent to select tools and forms.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
instructions |
str
|
Instructions prompt to select tool or form. |
required |
cat |
StrayCat
|
StrayCat instance. |
required |
Returns:
Name | Type | Description |
---|---|---|
instructions |
str
|
Instructions prompt to select tool or form |
Notes
This prompt explains the Agent how to select a tool or form.
Source code in cat/mad_hatter/core_plugin/hooks/prompt.py
agent_prompt_prefix(prefix, cat)
Hook the main prompt prefix.
Allows to edit the prefix of the Main Prompt that the Cat feeds to the Agent. It describes the personality of your assistant and its general task.
The prefix is then completed with the agent_prompt_suffix
.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
prefix |
str
|
Main / System prompt with personality and general task to be accomplished. |
required |
cat |
StrayCat
|
StrayCat instance. |
required |
Returns:
Name | Type | Description |
---|---|---|
prefix |
str
|
Main / System prompt. |
Notes
The default prefix describe who the AI is and how it is expected to answer the Human.
Source code in cat/mad_hatter/core_plugin/hooks/prompt.py
agent_prompt_suffix(prompt_suffix, cat)
Hook the main prompt suffix.
Allows to edit the suffix of the Main Prompt that the Cat feeds to the Agent.
The suffix is concatenated to agent_prompt_prefix
when RAG context is used.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
cat |
StrayCat
|
StrayCat instance. |
required |
Returns:
Name | Type | Description |
---|---|---|
prompt_suffix |
str
|
The suffix string to be concatenated to the Main Prompt (prefix). |
Notes
The default suffix has a few placeholders: - {episodic_memory} provides memories retrieved from episodic memory (past conversations) - {declarative_memory} provides memories retrieved from declarative memory (uploaded documents) - {chat_history} provides the Agent the recent conversation history - {input} provides the last user's input - {agent_scratchpad} is where the Agent can concatenate tools use and multiple calls to the LLM.