Ultimate General Intelligent Agent Prompt Refinement Detailed Explanation — Author: Wang Jiao Cheng#
Execution protocol waiting for instructions: Simple tasks are executed using adaptive identity overlay input processing output structure primitives, complex tasks are decomposed into simple tasks assigned to primitives to form a chain for execution, by default, input processing output details are not displayed but users can request to display them.#
Overall implementation of automatic intelligent routing, automatic combination innovation, automatic creation of agents, automatic agent collaboration, automatic workflow planning, and transparent explainability;
Execution protocol waiting for instructions:#
Execution protocol refers to the content after the colon, which is the task execution protocol reached between the user and the model;
Waiting for instructions refers to the content after the colon, meaning that the task execution protocol is waiting for the input of task instructions that need to be executed before the protocol can execute the task, putting the model in a standby state for the protocol to take effect;
Simple tasks are executed using adaptive identity overlay input processing output structure primitives,#
Simple tasks refer to basic tasks that the model can complete correctly without decomposition, whether it is basic is relative to the model's capabilities, and it does not necessarily mean atomic tasks; the model can understand what constitutes a simple task;
Using indicates that the task is not executed directly but is executed using the following primitives, specifying primitives to execute tasks is the basis for subsequent prompt modular control;
Adaptive refers to automatically adapting based on the input task content, equivalent to autonomously creating virtual agents, and autonomously creating multiple virtual agents for collaboration;
Identity refers to knowledge and abilities related to a role, equivalent to a virtual agent, where an agent typically represents a set of knowledge and abilities related to a role;
Overlay refers to multiple identities rather than a single identity, and overlay can achieve combinatorial innovation, equivalent to autonomously planning multiple virtual agent collaborations;
Input processing output structure refers to the basic system structure that includes input, processing, and output, equivalent to functions in mathematics and methods in programming;
Primitive refers to basic units, and the model can understand which type of basic unit it belongs to;
Execution refers to executing simple tasks, also implying that primitives are the basic execution units;
Complex tasks are decomposed into simple tasks assigned to primitives to form a chain for execution,#
Complex tasks refer to composite tasks that the model cannot complete correctly without prior decomposition, whether it is complex is relative to the model's capabilities, and the model can understand what constitutes complexity;
Decomposed into simple tasks means breaking down composite tasks that the model cannot complete correctly without prior decomposition into basic tasks that the model can complete correctly without decomposition, and the model can decide the degree of decomposition;
Assigned to primitives means that each simple task after the decomposition of the complex task is assigned a primitive to execute, with each primitive being an adaptive identity overlay input processing output structure;
Form a chain for execution means that each primitive executes in a certain executable order according to the assigned simple tasks, equivalent to autonomously planning a virtual workflow, where the virtual workflow consists of a series of primitives, each of which contains multiple virtual agent collaborations, representing a series of collaborations among multiple virtual agents;
By default, input processing output details are not displayed but users can request to display them.#
By default refers to the preset rules for the output after the overall execution of the task;
Not displayed means that specified content is not output;
Input processing output details refer to the input, processing, and output details of each primitive, excluding the final results;
But users can request to display means that the preset rules for the output after the overall execution of the task do not output specified content but allow for output display upon user request, meaning that the input, processing, and output details of each primitive are allowed to be output and displayed, where processing includes identity overlay and processing procedures, and details include any relevant content related to input, processing, and output, with the transparency and explainability mechanism allowing for on-demand display of relevant details, such as avoiding the insecurity of black boxes, obtaining decision-making basis, or demonstrating the need for innovative content.