Improving prompts - GOOD

docs/devel/prompts/v0/output-prompt.txt

@@ -3,9 +3,15 @@ You are 'Taius', the AskAI assistant. Act as a means of digital inclusion for vi
 Before responding to the user, it is imperative that you follow the step-by-step instructions provided below in sequential order:
 
 1. Craft your reply solely based on the information given in the provided contexts.
+
 2. Create a summarized and accessible version of its content.
-3. Limit showing 5 items and append a summary (Example: \nTotal files:33, Omitted:5\n).
+
+3. When listing is necessary limit showing 5 items and append a summary (Example: \nTotal files:33, Omitted:5\n).
+
 4. When the output contains files and/or folders listing, specify whether each item is a file or, folder, and also, size and modified date.
+
 5. Prefer rendering numbered lists than bulleted, when listing is necessary.
+
 6. Start your response with the phrase: Here is a summarized version of the command output\n\n
+
 7. Conclude your response by dropping small conclusion about the content.

docs/devel/prompts/v1/analysis-prompt.txt

@@ -1,4 +1,4 @@
-You are 'Taius', the AskAI assistant. Act as a means of digital inclusion for visually impaired individuals, specifically, a Speech-to-Text (STT) interpretation engine. You will be given a question that can only be responded referring previous prompts.
+You are 'Taius', the AskAI assistant. Act as a means of digital inclusion for visually impaired individuals, specifically, a Speech-to-Text (STT) interpretation engine. You will be given a question that can only be responded referring to the provided contexts.
 
 Before responding to the user, it is imperative that you follow the step-by-step instructions provided below in sequential order:
 

src/main/askai/core/model/processor_response.py

@@ -9,12 +9,11 @@
 
    Copyright·(c)·2024,·HSPyLib
 """
-
-from askai.core.model.terminal_command import TerminalCommand
+import json
 from dataclasses import dataclass, field
 from typing import List
 
-import json
+from askai.core.model.terminal_command import TerminalCommand
 
 
 @dataclass

src/main/askai/core/processor/instances/analysis_processor.py

@@ -31,6 +31,17 @@
 class AnalysisProcessor:
     """Process analysis prompts."""
 
+    @staticmethod
+    def _wrap_output(query_response: ProcessorResponse) -> str:
+        """Wrap the output into a new string to be forwarded to the next processor.
+        :param query_response: The query response provided by the AI.
+        """
+        query_response.query_type = QueryType.GENERIC_QUERY.value
+        query_response.require_summarization = False
+        query_response.require_internet = False
+
+        return str(query_response)
+
     @staticmethod
     def q_type() -> str:
         return QueryType.ANALYSIS_QUERY.value
@@ -58,17 +69,18 @@ def process(self, query_response: ProcessorResponse) -> Tuple[bool, Optional[str
         template = PromptTemplate(input_variables=[], template=self.template())
         final_prompt: str = msg.translate(template.format())
         shared.context.set("SETUP", final_prompt, "system")
-        shared.context.set("QUESTION", query_response.question)
-        context: ContextRaw = shared.context.join("CONTEXT", "SETUP", "QUESTION")
+        shared.context.set("QUESTION", f"\n\nQuestion: {query_response.question}\n\nHelpful Answer:")
+        context: ContextRaw = shared.context.join("SETUP", "CONTEXT", "QUESTION")
         log.info("Analysis::[QUESTION] '%s'  context=%s", query_response.question, context)
 
-        if (
-            response := shared.engine.ask(context, *Temperatures.CODE_GENERATION.value)
-        ) and response.is_success:
+        if (response := shared.engine.ask(context, *Temperatures.CODE_GENERATION.value)) and response.is_success:
             log.debug("Analysis::[RESPONSE] Received from AI: %s.", response)
-            if output := response.message:
+            if response.message and (output := response.message) != "I don't know.":
                 shared.context.push("CONTEXT", query_response.question)
                 shared.context.push("CONTEXT", output, "assistant")
+            else:
+                self._next_in_chain = QueryType.GENERIC_QUERY.proc_name
+                output = self._wrap_output(query_response)
             status = True
         else:
             log.error(

src/main/askai/core/processor/instances/command_processor.py

@@ -79,7 +79,7 @@ def process(self, query_response: ProcessorResponse) -> Tuple[bool, Optional[str
         template = PromptTemplate(input_variables=["os_type", "shell"], template=self.template())
         final_prompt: str = template.format(os_type=prompt.os_type, shell=prompt.shell)
         shared.context.set("SETUP", final_prompt, "system")
-        shared.context.set("QUESTION", query_response.question)
+        shared.context.set("QUESTION", f"\n\nQuestion: {query_response.question}\n\nHelpful Answer:")
         context: ContextRaw = shared.context.join("CONTEXT", "SETUP", "QUESTION")
         log.info("Command::[QUESTION] '%s'  context=%s", query_response.question, context)
 
@@ -105,13 +105,13 @@ def _process_command(self, query_response: ProcessorResponse, cmd_line: str) ->
         """
         status = False
         command = cmd_line.split(" ")[0].strip()
-        cmd_out = None
+        output = None
         try:
             if command and which(command):
                 cmd_line = expandvars(cmd_line.replace("~/", f"{os.getenv('HOME')}/").strip())
                 AskAiEvents.ASKAI_BUS.events.reply.emit(message=msg.executing(cmd_line))
                 log.info("Executing command `%s'", cmd_line)
-                cmd_out, exit_code = Terminal.INSTANCE.shell_exec(cmd_line, shell=True)
+                output, exit_code = Terminal.INSTANCE.shell_exec(cmd_line, shell=True)
                 if exit_code == ExitStatus.SUCCESS:
                     log.info("Command succeeded.\nCODE=%s \nPATH: %s \nCMD: %s ", exit_code, os.getcwd(), cmd_line)
                     if _path_ := extract_path(cmd_line):
@@ -121,16 +121,16 @@ def _process_command(self, query_response: ProcessorResponse, cmd_line: str) ->
                         else:
                             log.warning("Directory '%s' does not exist. Curdir unchanged!", _path_)
                     AskAiEvents.ASKAI_BUS.events.reply.emit(message=msg.cmd_success(exit_code), erase_last=True)
-                    if not cmd_out:
-                        cmd_out = msg.cmd_no_output()
+                    if not output:
+                        output = msg.cmd_no_output()
                     else:
-                        shared.context.set("CONTEXT", f"Command `{cmd_line}' output:\n\n```\n{cmd_out}\n```")
-                        cmd_out = self._wrap_output(query_response, cmd_line, cmd_out)
+                        shared.context.set("CONTEXT", f"Command `{cmd_line}' output:\n\n```\n{output}\n```")
+                        output = self._wrap_output(query_response, cmd_line, output)
                 else:
                     log.error("Command failed.\nCODE=%s \nPATH=%s \nCMD=%s ", exit_code, os.getcwd(), cmd_line)
-                    cmd_out = msg.cmd_failed(cmd_line)
+                    output = msg.cmd_failed(cmd_line)
                 status = True
             else:
-                cmd_out = msg.cmd_no_exist(command)
+                output = msg.cmd_no_exist(command)
         finally:
-            return status, cmd_out
+            return status, output

src/main/askai/core/processor/instances/generic_processor.py

@@ -59,7 +59,7 @@ def process(self, query_response: ProcessorResponse) -> Tuple[bool, Optional[str
         template = PromptTemplate(input_variables=["user"], template=self.template())
         final_prompt: str = msg.translate(template.format(user=prompt.user))
         shared.context.set("SETUP", final_prompt, "system")
-        shared.context.set("QUESTION", query_response.question)
+        shared.context.set("QUESTION", f"\n\nQuestion: {query_response.question}\n\nHelpful Answer:")
         context: ContextRaw = shared.context.join("GENERAL", "SETUP", "QUESTION")
         log.info("Setup::[GENERIC] '%s'  context=%s", query_response.question, context)
 

src/main/askai/core/processor/instances/internet_processor.py

@@ -65,7 +65,7 @@ def process(self, query_response: ProcessorResponse) -> Tuple[bool, Optional[str
         template = PromptTemplate(input_variables=["cur_date"], template=self.template())
         final_prompt: str = msg.translate(template.format(cur_date=now(self.DATE_FMT)))
         shared.context.set("SETUP", final_prompt, "system")
-        shared.context.set("QUESTION", query_response.question)
+        shared.context.set("QUESTION", f"\n\nQuestion: {query_response.question}\n\nHelpful Answer:")
         context: ContextRaw = shared.context.join("SETUP", "QUESTION")
         log.info("Setup::[INTERNET] '%s'  context=%s", query_response.question, context)
 

src/main/askai/core/processor/instances/summary_processor.py

@@ -75,7 +75,7 @@ def process(self, query_response: ProcessorResponse) -> Tuple[bool, Optional[str
         template = PromptTemplate(input_variables=["os_type"], template=self.template())
         final_prompt: str = msg.translate(template.format(os_type=prompt.os_type))
         shared.context.set("SETUP", final_prompt, "system")
-        shared.context.set("QUESTION", query_response.question)
+        shared.context.set("QUESTION", f"\n\nQuestion: {query_response.question}\n\nHelpful Answer:")
         context: ContextRaw = shared.context.join("SETUP", "QUESTION")
         log.info("Setup::[SUMMARY] '%s'  context=%s", query_response.question, context)
 

src/main/askai/core/processor/processor_proxy.py

@@ -48,7 +48,7 @@ def process(self, question: str) -> Tuple[bool, ProcessorResponse]:
         status = False
         template = PromptTemplate(input_variables=[], template=self.template())
         final_prompt = msg.translate(template.format())
-        shared.context.set("SETUP", final_prompt)
+        shared.context.set("SETUP", final_prompt, "system")
         shared.context.set("QUESTION", f"\n\nQuestion: {question}\n\nHelpful Answer:")
         context: ContextRaw = shared.context.join("CONTEXT", "SETUP", "QUESTION")
         log.info("Ask::[QUESTION] '%s'  context=%s", question, context)

src/main/askai/core/support/utilities.py

@@ -59,7 +59,7 @@ def beautify(text: Any) -> str:
     text = re.sub(r"([Jj]oke( [Tt]ime)?)[-:\s][ \n\t]+", CHAT_ICONS[''], text)
     text = re.sub(r"[Ff]un [Ff]acts?[-:\s][ \n\t]+", CHAT_ICONS[''], text)
     text = re.sub(r"[Aa]dvice[-:\s][ \n\t]+", CHAT_ICONS[''], text)
-    text = re.sub(re_url, r'%CYAN% \1%GREEN%', text)
+    text = re.sub(re_url, r' \1', text)
     text = re.sub(r"^\n+", '', text, re.MULTILINE)
     text = re.sub(r"\n{2,}", '\n\n', text, re.MULTILINE)
     # fmt: on

src/main/askai/resources/assets/prompts/analysis-prompt.txt

@@ -1,16 +1,16 @@
-You are 'Taius', the AskAI assistant. Act as a means of digital inclusion for visually impaired individuals, specifically, a Speech-to-Text (STT) interpretation engine. You will be given a question that can only be responded referring previous prompts.
+Act as a means of digital inclusion for visually impaired individuals, specifically, a Speech-to-Text (STT) interpretation engine. You will be given a question that can only be responded referring all chat history.
 
 Before responding to the user, it is imperative that you follow the step-by-step instructions provided below in sequential order:
 
-1. Craft your reply solely based on the information given in the provided contexts. If you don't know the answer, just say that you don't know, don't try to make up an answer.
+1. Thoroughly examine the complete chat history before making any decisions or conclusions.
 
-2. Create a summarized and accessible version of its content.
+2. Craft your reply solely based on the information given in the chat history. If you don't know the answer, just say exactly "I don't know", don't try to make up an answer. Provide your response immediately.
 
-3. Refine your response by ensuring you don't redundantly reiterate any items previously mentioned in the conversation history.
+3. Create a summarized and accessible version of its content.
 
-4. Limit showing 5 items and append a summary (Example: \nTotal files:33, Omitted:5\n).
+4. Refine your response by ensuring you don't redundantly reiterate any items previously mentioned in the conversation history.
 
-5. Prefer rendering numbered lists than bulleted ones.
+5. When the output includes a list, limit showing 5 items and append a summary (Example: \nTotal files:33, Omitted:5\n). Prefer rendering numbered lists than bulleted, when listing is necessary.
 
 6. Start your response with the phrase: Analysing the provided data\n
 

src/main/askai/resources/assets/prompts/command-prompt.txt

@@ -1,4 +1,4 @@
-You are 'Taius', the AskAI assistant. Act as a means of digital inclusion for visually impaired individuals, specifically, a Speech-to-Text (STT) interpretation engine. You will be given a 'question' that will require creating a terminal command.
+Act as a means of digital inclusion for visually impaired individuals, specifically, a Speech-to-Text (STT) interpretation engine. You will be given a 'question' that will require creating a terminal command.
 
 Before responding to the user, it is imperative that you follow the step-by-step instructions provided below in sequential order:
 

src/main/askai/resources/assets/prompts/internet-prompt.txt

@@ -1,15 +1,13 @@
-You are 'Taius', the AskAI assistant. Act as a means of internet crawler and researcher.
+Act as a means of internet crawler, scraper and researcher.
 
 Before responding to the user, you must follow the step-by-step instructions provided below in sequential order:
 
 1. Determine a list of keywords that when combined are good for retrieving the required information for a successful response. Understand the question and try to add more keywords to refine the question.
 
 2. Determine which sites are good for retrieving the required information for a successful response. Please include a minimum of three URLs, and a maximum of five.
 
-3. When the current date is important to retrieve accurate responses. If that's the case, provide the specific field: 'after', containing '{cur_date}'.
+3. Generate a JSON response containing the designated fields.
 
-4. Generate a JSON response containing the designated fields.
+4. The final response 'JSON' must contain the fields: 'keywords' and 'sites'.
 
-5. The final response 'JSON' must contain the fields: 'keywords' and 'sites'.
-
-6. The final response is a formatted JSON with no additional description or context.
+5. The final response is a formatted JSON with no additional description or context.

src/main/askai/resources/assets/prompts/output-prompt.txt

@@ -1,10 +1,10 @@
-You are 'Taius', the AskAI assistant. Act as a means of digital inclusion for visually impaired individuals, specifically, a Speech-to-Text (STT) interpretation engine.
+Act as a means of digital inclusion for visually impaired individuals, specifically, a Speech-to-Text (STT) interpretation engine.
 
 Before responding to the user, it is imperative that you follow the step-by-step instructions provided below in sequential order:
 
 1. You will be given the '{shell}' '{command_line}' output.
 
-2. Craft your reply solely based on the information given in the provided data.
+2. Craft your reply solely based on the information given in the provided command output.
 
 3. Create a summarized and accessible version of its content.
 
@@ -14,6 +14,6 @@ Before responding to the user, it is imperative that you follow the step-by-step
 
 6. When the provided output does not contain files or folders listing, create a small analysis of the provided data.
 
-6. Start your response with the phrase: Here is a summarized version\n\n
+7. Start your response with the phrase: Here is a summarized version\n\n
 
-7. Wrap up your reply by offering a succinct hint or tip related to the answer; prefix with: 'Hints:'.
+8. Wrap up your reply by offering a succinct hint or tip related to the answer; prefix with: 'Hints:'.

src/main/askai/resources/assets/prompts/proxy-prompt.txt

@@ -1,15 +1,13 @@
-As 'Taius', the AI query proxy. Your task is to analyze and categorize the types of queries presented to you. Discern the diverse query types and identify their specific processing requirements. You MUST return a "JSON string" containing the designated fields, no more than that. Queries must fall into one of the following categories:
+Your task is to analyze and categorize the types of queries presented to you. Discern the diverse query types and identify their specific processing requirements. You MUST return a "JSON string" containing the designated fields, no more than that. Queries must fall into one of the following:
 
-- "InternetQuery" (Examples: [what is the next lakers game, what is the whether like today])
+- "InternetQuery" (Examples: [what is the next lakers game, what is the whether like today, who is the actual president of brazil])
 
 - "SummarizationQuery" (Examples: [summarize my documents, summarize the file /tmp/the-file.md])
 
 - "AnalysisQuery" (Examples: [is there any image, how many reminders, what is the total size, what should I do])
 
 - "CommandQuery" (Examples: [list my images, open 1, play it, show me it])
 
-- "GenericQuery" (Examples: [what is the size of the moon, who are you])
-
 Before responding to the user, you must follow the step-by-step instructions provided below in sequential order:
 
 1. Determine if the query is clear and intelligible.
@@ -22,7 +20,7 @@ Before responding to the user, you must follow the step-by-step instructions pro
 
 5. If the user has provided a terminal command in a clear manner, select the 'CommandQuery'.
 
-6. If you haven't found an answer yet or are still undecided, choose the either 'GenericQuery' or 'AnalysisQuery' and include the boolean field 'unclear' set to true.
+6. If you haven't found an answer yet or are still undecided, choose 'AnalysisQuery'.
 
 7. The final response is a formatted JSON with no additional description or context.
 

src/main/askai/resources/assets/prompts/summary-prompt.txt

@@ -1,4 +1,4 @@
-You are 'Taius', the AskAI assistant. Act as a means of files, folder and documents summarizer.
+Act as a means of files, folders and documents summarizer.
 
 Before responding to the user, you must follow the step-by-step instructions provided below in sequential order: