feat: Claude Sonnet 4 1M context; Cursor rules budget 10% (cap 100k)

[jacsamell]

PR Type

Enhancement

---

Description

• Enable 1M context model support with updated cursor rules budgeting
• Increase cursor rules token limit from 20k to 100k tokens
• Reduce context ratio from 25% to 10% with hard cap
• Add OpenAI/LiteLLM compatibility shim for type mismatches

---

Changes walkthrough 📝

	Relevant files
Dependencies	litellm_ai_handler.py Add OpenAI/LiteLLM compatibility shim for type handling    pr_agent/algo/ai_handlers/litellm_ai_handler.py • Add compatibility shim for OpenAI/LiteLLM type mismatches • Handle ResponseTextConfig to ResponseFormatTextConfig renaming • Wrap aliasing in try/except for best effort compatibility +32/-1    requirements.txt Pin dependency versions for compatibility                                requirements.txt • Pin LiteLLM to version 1.75.8 and OpenAI to 1.100.1 • Update comment about version compatibility requirements +3/-3
Configuration changes	utils.py Update cursor rules token budgeting parameters                      pr_agent/git_providers/utils.py • Increase max_cursor_rules_tokens from 20k to 100k • Reduce cursor_rules_context_ratio from 0.25 to 0.10 +2/-2      configuration.toml Update default cursor rules configuration values                  pr_agent/settings/configuration.toml • Update default max_cursor_rules_tokens to 100000 • Update default cursor_rules_context_ratio to 0.10 +2/-2
Documentation	README.md Document updated cursor rules budgeting behavior                  README.md • Document new cursor rules budgeting (10% of context, 100k cap) • Explain 1M-context model support for up to 100k rule tokens +2/-0

---

Need help?

Type /help how to ... in the comments thread for any questions about PR-Agent usage.

Check out the documentation for more information.

[github-actions]

PR Reviewer Guide 🔍

(Review updated until commit d4454ca)

Here are some key observations to aid the review process:

⏱️ Estimated effort to review: 3 🔵🔵🔵⚪⚪

AI Review Score: 75%

🧪 No relevant tests

Confidence score [1-100]: 78

Complexity score [1-10]: 6

Security score [1-10]: 8

Auto approve recommendation: false

Auto approve reasoning: This PR includes dependency version pinning and runtime monkey-patching of modules through the compatibility shim. These changes could potentially break in different environments or cause import failures. The lack of tests for the new compatibility shim functionality also raises concerns about reliability.

Requires human approval: Complex Logic

💡 Code suggestions Error Logging Add logging to the compatibility shim exception handler to help debug issues when the shim fails to apply correctly. Existing code: except Exception: # Best effort compatibility; proceed even if aliasing fails pass Improved code: except Exception as e: # Best effort compatibility; proceed even if aliasing fails import logging logging.debug(f"OpenAI compatibility shim failed: {e}") pass

🔒 No security concerns identified

⚡ Recommended focus areas for review Runtime Patching The compatibility shim uses setattr to monkey-patch OpenAI module attributes at import time. This approach is fragile and could fail silently or cause unexpected behavior if module structures change. Consider alternative approaches or add comprehensive testing. try: # Shim: align Litellm's expected OpenAI type name with current OpenAI SDK # Older Litellm versions import `ResponseTextConfig`, which was renamed to # `ResponseFormatTextConfig` in newer OpenAI SDKs. from openai.types.responses import ( response as _openai_response_mod, response_create_params as _openai_response_params_mod, ) if ( not hasattr(_openai_response_mod, "ResponseTextConfig") and hasattr(_openai_response_mod, "ResponseFormatTextConfig") ): setattr( _openai_response_mod, "ResponseTextConfig", _openai_response_mod.ResponseFormatTextConfig, ) # Alias param type as well if needed if ( not hasattr(_openai_response_params_mod, "ResponseTextConfigParam") and hasattr(_openai_response_params_mod, "ResponseFormatTextConfigParam") ): setattr( _openai_response_params_mod, "ResponseTextConfigParam", _openai_response_params_mod.ResponseFormatTextConfigParam, ) except Exception: # Best effort compatibility; proceed even if aliasing fails pass Version Pinning Pinning exact versions of litellm and openai may cause dependency conflicts in environments with other packages that require different versions. Consider using compatible version ranges instead of exact pins. litellm==1.75.8 openai==1.100.1

---

🔍 Manual review required

Reason: Manual review required: Complex Logic