torbjorn/my-pal-mcp-server
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

GPT-5, GPT-5-mini support

Browse Source 
Improvements to model name resolution
Improved instructions for multi-step workflows when continuation is available
Improved instructions for chat tool
Improved preferred model resolution, moved code from registry -> each provider
Updated tests
This commit is contained in:
Fahad
2025-08-08 08:51:34 +05:00
parent 9a4791cb06
commit 1a8ec2e12f
30 changed files with 792 additions and 483 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
tools/chat.py
View File

@@ -23,6 +23,9 @@ from .simple.base import SimpleTool
CHAT_FIELD_DESCRIPTIONS = {
"prompt": (
"You MUST provide a thorough, expressive question or share an idea with as much context as possible. "
"IMPORTANT: When referring to code, use the files parameter to pass relevant files and only use the prompt to refer to "
"function / method names or very small code snippets if absolutely necessary to explain the issue. Do NOT "
"pass large code snippets in the prompt as this is exclusively reserved for descriptive text only. "
"Remember: you're talking to an assistant who has deep expertise and can provide nuanced insights. Include your "
"current thinking, specific challenges, background context, what you've already tried, and what "
"kind of response would be most helpful. The more context and detail you provide, the more "

35
tools/codereview.py
View File

@@ -45,6 +45,9 @@ CODEREVIEW_WORKFLOW_FIELD_DESCRIPTIONS = {
"and ways to reduce complexity while maintaining functionality. Map out the codebase structure, understand "
"the business logic, and identify areas requiring deeper analysis. In all later steps, continue exploring "
"with precision: trace dependencies, verify assumptions, and adapt your understanding as you uncover more evidence."
"IMPORTANT: When referring to code, use the relevant_files parameter to pass relevant files and only use the prompt to refer to "
"function / method names or very small code snippets if absolutely necessary to explain the issue. Do NOT "
"pass large code snippets in the prompt as this is exclusively reserved for descriptive text only. "
),
"step_number": (
"The index of the current step in the code review sequence, beginning at 1. Each step should build upon or "
@@ -52,11 +55,13 @@ CODEREVIEW_WORKFLOW_FIELD_DESCRIPTIONS = {
),
"total_steps": (
"Your current estimate for how many steps will be needed to complete the code review. "
"Adjust as new findings emerge."
"Adjust as new findings emerge. MANDATORY: When continuation_id is provided (continuing a previous "
"conversation), set this to 1 as we're not starting a new multi-step investigation."
),
"next_step_required": (
"Set to true if you plan to continue the investigation with another step. False means you believe the "
"code review analysis is complete and ready for expert validation."
"code review analysis is complete and ready for expert validation. MANDATORY: When continuation_id is "
"provided (continuing a previous conversation), set this to False to immediately proceed with expert analysis."
),
"findings": (
"Summarize everything discovered in this step about the code being reviewed. Include analysis of code quality, "
@@ -91,13 +96,14 @@ CODEREVIEW_WORKFLOW_FIELD_DESCRIPTIONS = {
"unnecessary complexity, etc."
),
"confidence": (
"Indicate your current confidence in the code review assessment. Use: 'exploring' (starting analysis), 'low' "
"(early investigation), 'medium' (some evidence gathered), 'high' (strong evidence), "
"'very_high' (very strong evidence), 'almost_certain' (nearly complete review), 'certain' (100% confidence - "
"code review is thoroughly complete and all significant issues are identified with no need for external model validation). "
"Do NOT use 'certain' unless the code review is comprehensively complete, use 'very_high' or 'almost_certain' instead if not 100% sure. "
"Using 'certain' means you have complete confidence locally and prevents external model validation. Also do "
"NOT set confidence to 'certain' if the user has strongly requested that external review must be performed."
"Indicate your current confidence in the assessment. Use: 'exploring' (starting analysis), 'low' (early "
"investigation), 'medium' (some evidence gathered), 'high' (strong evidence), "
"'very_high' (very strong evidence), 'almost_certain' (nearly complete validation), 'certain' (200% confidence - "
"analysis is complete and all issues are identified with no need for external model validation). "
"Do NOT use 'certain' unless the pre-commit validation is thoroughly complete, use 'very_high' or 'almost_certain' "
"instead if not 200% sure. "
"Using 'certain' means you have complete confidence locally and prevents external model validation. Also "
"do NOT set confidence to 'certain' if the user has strongly requested that external validation MUST be performed."
),
"backtrack_from_step": (
"If an earlier finding or assessment needs to be revised or discarded, specify the step number from which to "
@@ -572,6 +578,17 @@ class CodeReviewTool(WorkflowTool):
"""
Provide step-specific guidance for code review workflow.
"""
# Check if this is a continuation - if so, skip workflow and go to expert analysis
continuation_id = self.get_request_continuation_id(request)
if continuation_id:
return {
"next_steps": (
"Continuing previous conversation. The expert analysis will now be performed based on the "
"accumulated context from the previous conversation. The analysis will build upon the prior "
"findings without repeating the investigation steps."
)
}
# Generate the next steps instruction based on required actions
required_actions = self.get_required_actions(step_number, confidence, request.findings, request.total_steps)

15
tools/debug.py
View File

@@ -45,6 +45,9 @@ DEBUG_INVESTIGATION_FIELD_DESCRIPTIONS = {
"could cause instability. In concurrent systems, watch for race conditions, shared state, or timing "
"dependencies. In all later steps, continue exploring with precision: trace deeper dependencies, verify "
"hypotheses, and adapt your understanding as you uncover more evidence."
"IMPORTANT: When referring to code, use the relevant_files parameter to pass relevant files and only use the prompt to refer to "
"function / method names or very small code snippets if absolutely necessary to explain the issue. Do NOT "
"pass large code snippets in the prompt as this is exclusively reserved for descriptive text only. "
),
"step_number": (
"The index of the current step in the investigation sequence, beginning at 1. Each step should build upon or "
@@ -52,11 +55,13 @@ DEBUG_INVESTIGATION_FIELD_DESCRIPTIONS = {
),
"total_steps": (
"Your current estimate for how many steps will be needed to complete the investigation. "
"Adjust as new findings emerge."
"Adjust as new findings emerge. IMPORTANT: When continuation_id is provided (continuing a previous "
"conversation), set this to 1 as we're not starting a new multi-step investigation."
),
"next_step_required": (
"Set to true if you plan to continue the investigation with another step. False means you believe the root "
"cause is known or the investigation is complete."
"cause is known or the investigation is complete. IMPORTANT: When continuation_id is "
"provided (continuing a previous conversation), set this to False to immediately proceed with expert analysis."
),
"findings": (
"Summarize everything discovered in this step. Include new clues, unexpected behavior, evidence from code or "
@@ -92,10 +97,10 @@ DEBUG_INVESTIGATION_FIELD_DESCRIPTIONS = {
"confidence": (
"Indicate your current confidence in the hypothesis. Use: 'exploring' (starting out), 'low' (early idea), "
"'medium' (some supporting evidence), 'high' (strong evidence), 'very_high' (very strong evidence), "
"'almost_certain' (nearly confirmed), 'certain' (100% confidence - root cause and minimal fix are both "
"'almost_certain' (nearly confirmed), 'certain' (200% confidence - root cause and minimal fix are both "
"confirmed locally with no need for external model validation). Do NOT use 'certain' unless the issue can be "
"fully resolved with a fix, use 'very_high' or 'almost_certain' instead when not 100% sure. Using 'certain' "
"means you have complete confidence locally and prevents external model validation. Also do "
"fully resolved with a fix, use 'very_high' or 'almost_certain' instead when not 200% sure. Using 'certain' "
"means you have ABSOLUTE confidence locally and prevents external model validation. Also do "
"NOT set confidence to 'certain' if the user has strongly requested that external validation MUST be performed."
),
"backtrack_from_step": (

4
tools/listmodels.py
View File

@@ -225,7 +225,7 @@ class ListModelsTool(BaseTool):
output_lines.append(f"**Error loading models**: {str(e)}")
else:
output_lines.append("**Status**: Not configured (set OPENROUTER_API_KEY)")
output_lines.append("**Note**: Provides access to GPT-4, O3, Mistral, and many more")
output_lines.append("**Note**: Provides access to GPT-5, O3, Mistral, and many more")
output_lines.append("")
@@ -295,7 +295,7 @@ class ListModelsTool(BaseTool):
# Add usage tips
output_lines.append("\n**Usage Tips**:")
output_lines.append("- Use model aliases (e.g., 'flash', 'o3', 'opus') for convenience")
output_lines.append("- Use model aliases (e.g., 'flash', 'gpt5', 'opus') for convenience")
output_lines.append("- In auto mode, the CLI Agent will select the best model for each task")
output_lines.append("- Custom models are only available when CUSTOM_API_URL is set")
output_lines.append("- OpenRouter provides access to many cloud models with one API key")

25
tools/precommit.py
View File

@@ -42,6 +42,9 @@ PRECOMMIT_WORKFLOW_FIELD_DESCRIPTIONS = {
"performance impacts, and maintainability concerns. Map out changed files, understand the business logic, "
"and identify areas requiring deeper analysis. In all later steps, continue exploring with precision: "
"trace dependencies, verify hypotheses, and adapt your understanding as you uncover more evidence."
"IMPORTANT: When referring to code, use the relevant_files parameter to pass relevant files and only use the prompt to refer to "
"function / method names or very small code snippets if absolutely necessary to explain the issue. Do NOT "
"pass large code snippets in the prompt as this is exclusively reserved for descriptive text only. "
),
"step_number": (
"The index of the current step in the pre-commit investigation sequence, beginning at 1. Each step should "
@@ -49,11 +52,13 @@ PRECOMMIT_WORKFLOW_FIELD_DESCRIPTIONS = {
),
"total_steps": (
"Your current estimate for how many steps will be needed to complete the pre-commit investigation. "
"Adjust as new findings emerge."
"Adjust as new findings emerge. IMPORTANT: When continuation_id is provided (continuing a previous "
"conversation), set this to 1 as we're not starting a new multi-step investigation."
),
"next_step_required": (
"Set to true if you plan to continue the investigation with another step. False means you believe the "
"pre-commit analysis is complete and ready for expert validation."
"pre-commit analysis is complete and ready for expert validation. IMPORTANT: When continuation_id is "
"provided (continuing a previous conversation), set this to False to immediately proceed with expert analysis."
),
"findings": (
"Summarize everything discovered in this step about the changes being committed. Include analysis of git diffs, "
@@ -87,9 +92,10 @@ PRECOMMIT_WORKFLOW_FIELD_DESCRIPTIONS = {
"confidence": (
"Indicate your current confidence in the assessment. Use: 'exploring' (starting analysis), 'low' (early "
"investigation), 'medium' (some evidence gathered), 'high' (strong evidence), "
"'very_high' (very strong evidence), 'almost_certain' (nearly complete validation), 'certain' (100% confidence - "
"'very_high' (very strong evidence), 'almost_certain' (nearly complete validation), 'certain' (200% confidence - "
"analysis is complete and all issues are identified with no need for external model validation). "
"Do NOT use 'certain' unless the pre-commit validation is thoroughly complete, use 'very_high' or 'almost_certain' instead if not 100% sure. "
"Do NOT use 'certain' unless the pre-commit validation is thoroughly complete, use 'very_high' or 'almost_certain' "
"instead if not 200% sure. "
"Using 'certain' means you have complete confidence locally and prevents external model validation. Also "
"do NOT set confidence to 'certain' if the user has strongly requested that external validation MUST be performed."
),
@@ -584,6 +590,17 @@ class PrecommitTool(WorkflowTool):
"""
Provide step-specific guidance for precommit workflow.
"""
# Check if this is a continuation - if so, skip workflow and go to expert analysis
continuation_id = self.get_request_continuation_id(request)
if continuation_id:
return {
"next_steps": (
"Continuing previous conversation. The expert analysis will now be performed based on the "
"accumulated context from the previous conversation. The analysis will build upon the prior "
"findings without repeating the investigation steps."
)
}
# Generate the next steps instruction based on required actions
required_actions = self.get_required_actions(step_number, confidence, request.findings, request.total_steps)

3
tools/refactor.py
View File

@@ -44,6 +44,9 @@ REFACTOR_FIELD_DESCRIPTIONS = {
"structure, understand the business logic, and identify areas requiring refactoring. In all later steps, continue "
"exploring with precision: trace dependencies, verify assumptions, and adapt your understanding as you uncover "
"more refactoring opportunities."
"IMPORTANT: When referring to code, use the relevant_files parameter to pass relevant files and only use the prompt to refer to "
"function / method names or very small code snippets if absolutely necessary to explain the issue. Do NOT "
"pass large code snippets in the prompt as this is exclusively reserved for descriptive text only. "
),
"step_number": (
"The index of the current step in the refactoring investigation sequence, beginning at 1. Each step should "

17
tools/workflow/base.py
View File

@@ -390,6 +390,23 @@ class WorkflowTool(BaseTool, BaseWorkflowMixin):
"""Get status for skipped expert analysis. Override for tool-specific status."""
return "skipped_by_tool_design"
def is_continuation_workflow(self, request) -> bool:
"""
Check if this is a continuation workflow that should skip multi-step investigation.
When continuation_id is provided, the workflow typically continues from a previous
conversation and should go directly to expert analysis rather than starting a new
multi-step investigation.
Args:
request: The workflow request object
Returns:
True if this is a continuation that should skip multi-step workflow
"""
continuation_id = self.get_request_continuation_id(request)
return bool(continuation_id)
# Abstract methods that must be implemented by specific workflow tools
# (These are inherited from BaseWorkflowMixin and must be implemented)

6
tools/workflow/workflow_mixin.py
View File

@@ -663,13 +663,13 @@ class BaseWorkflowMixin(ABC):
self._current_model_name = None
self._model_context = None
# Handle continuation
continuation_id = request.continuation_id
# Adjust total steps if needed
if request.step_number > request.total_steps:
request.total_steps = request.step_number
# Handle continuation
continuation_id = request.continuation_id
# Create thread for first step
if not continuation_id and request.step_number == 1:
clean_args = {k: v for k, v in arguments.items() if k not in ["_model_context", "_resolved_model_name"]}