fix: add claude-opus-4-7 to NO_SUPPORT_TEMPERATURE_MODELS (#2400)

[raywcm]

Fixes #2400

Description

When using Claude Opus 4-7 models (via Anthropic, Vertex AI, or Bedrock), PR-Agent sends the temperature parameter which Anthropic has deprecated for these models, causing every command to fail with a 400 error:

temperature is deprecated for this model.

Changes Made

1. Added all Claude Opus 4-7 model variants to NO_SUPPORT_TEMPERATURE_MODELS in pr_agent/algo/__init__.py:

   • claude-opus-4-7
   • vertex_ai/claude-opus-4-7
   • anthropic/claude-opus-4-7
   • bedrock/anthropic.claude-opus-4-7
   • bedrock/anthropic.claude-opus-4-7-v1:0
   • bedrock/us.anthropic.claude-opus-4-7
   • bedrock/global.anthropic.claude-opus-4-7

2. Added bedrock/anthropic.claude-opus-4-7-v1:0 to MAX_TOKENS (same token limit as the non--v1:0 variant) so that get_max_tokens() does not crash when this model ID is configured.

The fix follows the existing pattern used for the Claude Opus 4-8 variants already in NO_SUPPORT_TEMPERATURE_MODELS. The LiteLLMAIHandler already skips sending the temperature parameter for any model in this list.

[qodo-free-for-open-source-projects]

Code Review by Qodo

🐞 Bugs (1) 📘 Rule violations (1) 📎 Requirement gaps (0)

Context used

✅ Tickets: 🎫 [Bug] `temperature` parameter sent to claude-opus-4-7 causes 400 from Anthropic

1. No tests for claude-opus-4-7 📘 Rule violation ⚙ Maintainability

Description

This PR changes runtime behavior by adding claude-opus-4-7 variants to
NO_SUPPORT_TEMPERATURE_MODELS, but no unit tests were added/updated to verify temperature is
omitted for these models. Without coverage, this regression-prone behavior may break again and
reintroduce the 400 error.

Code

pr_agent/algo/init.py[R314-321]

+    # Anthropic Claude Opus 4-7 — temperature is deprecated (Issue #2400)
+    "claude-opus-4-7",
+    "vertex_ai/claude-opus-4-7",
+    "anthropic/claude-opus-4-7",
+    "bedrock/anthropic.claude-opus-4-7",
+    "bedrock/anthropic.claude-opus-4-7-v1:0",
+    "bedrock/us.anthropic.claude-opus-4-7",
+    "bedrock/global.anthropic.claude-opus-4-7",

Evidence

Rule 16 requires adding/updating pytest tests when behavior changes. The PR modifies
NO_SUPPORT_TEMPERATURE_MODELS to include claude-opus-4-7 variants, and LiteLLMAIHandler uses
this list to decide whether to attach kwargs["temperature"]; however, no corresponding unit test
was added alongside existing similar tests (e.g., for claude-opus-4-8) to assert temperature is
omitted for the new claude-opus-4-7 entries.

AGENTS.md: Add/Update Tests and Place Them in the Correct Pytest Suite Directory
pr_agent/algo/init.py[314-321]
pr_agent/algo/ai_handlers/litellm_ai_handler.py[472-480]
tests/unittest/test_litellm_chat_completion_core.py[76-101]

Agent prompt

The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
The PR adds multiple `claude-opus-4-7` model IDs to `NO_SUPPORT_TEMPERATURE_MODELS` so `LiteLLMAIHandler` skips sending `temperature`, but there is no unit test asserting that `temperature` is not included in the LiteLLM `acompletion()` kwargs for these models.

## Issue Context
There are already unit tests asserting `temperature` behavior for the `claude-opus-4-8` family; this change should add similar coverage for the `claude-opus-4-7` family to prevent regressions.

## Fix Focus Areas
- pr_agent/algo/__init__.py[314-321]
- pr_agent/algo/ai_handlers/litellm_ai_handler.py[472-480]
- tests/unittest/test_litellm_chat_completion_core.py[76-101]

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools

---

2. Untested MAX_TOKENS entry 🐞 Bug ⚙ Maintainability

Description

A new MAX_TOKENS key (bedrock/anthropic.claude-opus-4-7-v1:0) was added, but the Claude Opus 4-7
max-tokens unit test does not include this variant, so future accidental removal/rename would not be
caught. This is a maintainability/regression-coverage gap rather than a functional bug today.

Code

pr_agent/algo/init.py[170]

+    'bedrock/anthropic.claude-opus-4-7-v1:0': 1000000,

Evidence

The PR introduces a new MAX_TOKENS entry for bedrock/anthropic.claude-opus-4-7-v1:0, but the
unit test that enumerates Claude Opus 4-7 variants omits this identifier, so that new case is not
exercised by tests.

pr_agent/algo/init.py[161-172]
tests/unittest/test_get_max_tokens.py[196-206]

Agent prompt

The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

### Issue description
`MAX_TOKENS` now includes the model id `bedrock/anthropic.claude-opus-4-7-v1:0`, but the existing `test_claude_opus_4_7_model_max_tokens` parametrization doesn’t cover it. This reduces regression protection for the newly-supported identifier.

### Issue Context
The PR explicitly added this model id to prevent `get_max_tokens()` from raising for that configuration. Adding it to the existing 4-7 parameter list will lock the intended behavior.

### Fix Focus Areas
- tests/unittest/test_get_max_tokens.py[196-206]

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools

---

[qodo-free-for-open-source-projects]

PR Summary by Qodo

Fix: skip temperature for Claude Opus 4-7 model IDs and add Bedrock v1:0 max tokens
🐞 Bug fix 🕐 Less than 10 minutes

Walkthroughs

Description

• Add Claude Opus 4-7 model variants to the no-temperature allowlist to avoid Anthropic 400s.
• Register Bedrock Opus 4-7 v1:0 in MAX_TOKENS to prevent get_max_tokens() crashes.

Diagram

graph TD
  Cfg["pr_agent/algo/__init__.py (model caps)"] --> H(["LiteLLMAIHandler"] ) --> P{{"LLM provider APIs"}}
  H -->|"omit temperature"| P
  subgraph Legend
    direction LR
    _cfg["Config/constants"] ~~~ _svc(["Handler/service"]) ~~~ _ext{{"External API"}}
  end

Loading

High-Level Assessment

The following are alternative approaches to this PR:

1. Prefix/regex-based model family matching

• ➕ Avoids enumerating every provider-specific Opus 4-7 variant (incl. future suffixes like -v1:0).
• ➕ Reduces maintenance when new region-scoped Bedrock IDs appear.
• ➖ Risk of false positives/negatives if other models share similar prefixes or providers change naming conventions.
• ➖ Harder to reason about and audit compared to an explicit list.

2. Provider capability discovery (via SDK / LiteLLM metadata)

• ➕ Single source of truth for which parameters are supported per model.
• ➕ Automatically adapts to vendor deprecations/rollouts.
• ➖ Adds runtime dependency on metadata availability/accuracy.
• ➖ More complex integration and testing; may not be supported uniformly across providers.

Recommendation: The explicit enumeration is appropriate for a targeted hotfix and matches the existing pattern used for Opus 4-8. If Opus model variants continue proliferating, consider augmenting with a conservative prefix-based matcher specifically for known providers (e.g., Bedrock 'bedrock/claude-opus-4-7') to reduce future churn while keeping the explicit list as an override/escape hatch.

File Changes

Bug fix (1)

__init__.py Add Claude Opus 4-7 to no-temperature models and register Bedrock v1:0 tokens +10/-1

Add Claude Opus 4-7 to no-temperature models and register Bedrock v1:0 tokens

• Adds Bedrock 'anthropic.claude-opus-4-7-v1:0' to MAX_TOKENS to match the existing Opus 4-7 limit and avoid failures in get_max_tokens(). Extends NO_SUPPORT_TEMPERATURE_MODELS with Claude Opus 4-7 variants (Anthropic/Vertex AI/Bedrock, including region-scoped and v1:0 IDs) so the handler omits the deprecated temperature parameter.

pr_agent/algo/init.py

---