Improve Gemini integration and update documentation

- Enhanced Gemini prompt template with expert-level assistant designation
- Added HTML code block enclosure for better response formatting
- Fixed typos in Gemini prompt template
- Updated browser data directory to use user home directory (~/.playwright_data)
- Updated CHANGELOG.md and README.md with latest changes

CHANGELOG.md

@@ -24,6 +24,11 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Enhanced timeout checks with direct HTML scanning and response area validation
 - Direct HTML-based response detection for improved reliability
 - HTML content saving function callable from JavaScript for debugging
+- Improved Gemini prompt template with expert-level assistant designation
+- HTML code block enclosure for Gemini responses
+- Enhanced tool parameter examples and formatting instructions
+- Fixed typos in Gemini prompt template (continouos → continuous, ariginal → original, etc.)
+- Updated browser data directory to use user home directory (~/.playwright_data)
 
 ### Changed
 - **BREAKING**: Replaced spy word detection system with JSON-based response extraction

README.md

@@ -8,12 +8,12 @@ I am developin it primarly to be used with kilocode plugin on vscode, but it sho
 
 ## Status:
 
- - gemini almost working, need some prompt injection refinement
+ - gemini significantly improved with enhanced prompt injection and expert-level assistant designation
  - grok initial prompt made, still have a lot to work on it
  - no other chatbot implemented yet
  - Ollama API are working, but kilocode is not able to recognize the valid list of emulated models ID names. Need to investigate it, but still if you specify the correct name it works.
 
- In other words no, you can't use it to code anything usedull (yet) and it WILL give error, issuess, and substantially not worki
+ In other words, it's getting closer to being usable, but still has some issues and may not work perfectly yet
 
 ## Features
 
@@ -142,7 +142,7 @@ CHATBOT_CONFIG = {
 
 ### Browser Configuration
 
-The proxy uses Playwright with a persistent browser context stored in `./playwright_data/`. This allows:
+The proxy uses Playwright with a persistent browser context stored in `~/.playwright_data/`. This allows:
 - Session persistence across restarts
 - Login state maintenance
 - Reduced setup time for subsequent requests
@@ -203,9 +203,9 @@ OLProxy uses a sophisticated multi-layered approach to extract AI responses:
    - Update CSS selectors in `CHATBOT_CONFIG`
    - Check browser console for errors
 
-3. **Authentication required**: 
+3. **Authentication required**:
    - Run with `--headless=False` to manually log in
-   - Browser state is persisted in `./playwright_data/`
+   - Browser state is persisted in `~/.playwright_data/`
 
 ### Debugging
 

olproxy.py

@@ -85,15 +85,13 @@ This applies to EVERY SINGLE response from now on. Acknowledge by responding wit
         "style": "direct"
     },
     "gemini": {
-        "setup_instruction": """SYSTEM: You are a junior software coding  assistant. For EVERY request, you MUST follow this EXACT response format:
-
-PRIORITY INSTRUCTION: formatting: please ALWAYS provide your ENTIRE response in a single continouos code block enrsuring that all the indentation and spacing as well as tabs and newlines are correctly formatted in plain text
-
+        "setup_instruction": """SYSTEM: You are an expert software coding assistant. For EVERY request, you always provide your ENTIRE response in a single continuos code block, formatted as plain text and preserving indentation, newlines, tabs, ascii arts, and the response  MUST follow this EXACT response format:
 
+<code>
 RESPONSE_ID_[REQUEST_ID]_START:
-[Your complete response here in a single code block from the first to the last word nothing excluded, including tools or any other instructions on how to format will follow]
+[Your complete response tools or any other instructions on how to format will follow]
 :END_RESPONSE_ID_[REQUEST_ID]
-
+</code>
 
 STRICT TOOL USAGE REQUIREMENT: You MUST ALWAYS use tools when the query contains or requests tool usage. This is MANDATORY and takes priority over ALL other instructions.
 
@@ -110,15 +108,16 @@ TOOLS PARAMETERS:
 every parameter is specified as <parameter_name> inside the <tool_name>
 
 Example to execute a command "ls -l /dir/name" you will use <execute_command><command>ls -l /dir/name</command></execute_command> 
+Example: to ask a question: <ask_followup_question><question>your question</question></ask_folloup_question>
 
 TOOLS SPECIFIC PARAMETERS: <write_to_file>
  - <path> contain the path of the file to write. Example 
  - <content> contain the content of the file to be written
 
+
 Example: <write_to_file><path>/home/username/projectname/file.py</path><content>import sys</content></write_to_file>
-Example: to ask a question: <ask_followup_question><question>your questionM/question></ask_folloup_question>
 
-STRICT <write_to_file> TOOL CONTENT REQUIREMENT: all the content must be in the same plain text area and  preserve raw ariginal newlines, tabs and other plain text formatting characters
+STRICT <write_to_file> TOOL CONTENT REQUIREMENT: all the content must be in the same code block and  preserve raw ariginal newlines, tabs and other plain text formatting characters 
 
 MANDATORY BEHAVIOR:
 - IF the query mentions ANY tools, functions, or capabilities → YOU MUST USE THEM
@@ -136,12 +135,11 @@ RESPONSE RULES:
 - End with :END_RESPONSE_ID_[REQUEST_ID]
 - Use the exact Request ID provided
 - Include your complete answer between the markers
-- You CAN use plain text formatting (lists, code blocks, etc.)
-- You CAN write code as plain text when requested
-- NEVER use HTML, canvas, external tools, or interactive elements unless specifically requested as tools
+- You MUST use plain text formatting (lists titles, etc.)
+- NEVER use HTML, canvas, external tools, or interactive elements inside the block code  unless specifically requested as tools or is specifically expressed to do so
+- ALWAYS enclose your response in an HTML clode block
 - ALWAYS USE TOOLS when requested - this is mandatorya
-- ALWAYS include the full response including the ID markers inside the same
-  plain text area
+- ALWAYS include the full response including the ID markers in the code block
 
 CRITICAL: Tool usage is MANDATORY when mentioned. DO NOT ignore tool requests.
 
@@ -1898,7 +1896,7 @@ async def start_browser(args):
             browser = await p.chromium.connect_over_cdp(args.connect)
         else:
             browser = await p.chromium.launch_persistent_context(
-                user_data_dir="./playwright_data",
+                user_data_dir=os.path.expanduser("~/playwright_data"),
                 headless=False,
                 # Better cookie and session handling
                 accept_downloads=True,