What does this PR do?

The think parameter is Ollama-native and was being sent to ALL custom providers when reasoning was disabled, causing HTTP 422 rejections from Mistral, Fireworks, Together.ai, vLLM, and LM Studio — all of which validate request bodies and reject unrecognized parameters.

Root cause: _build_api_kwargs() used provider == "custom" as the guard condition, matching every custom endpoint regardless of whether it understands the think parameter.

Fix:

• Add _is_ollama() detection method using two complementary signals:
  1. _ollama_num_ctx — set during init via config or /api/show probe (proves endpoint is Ollama)
  2. URL heuristics — "ollama" or ":11434" in the base URL (fallback for when probe failed)
• Replace provider == "custom" guard with _is_ollama() in the think=False injection path

Related Issue

Closes #11237

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)
• ✨ New feature (non-breaking change that adds functionality)
• 🔒 Security fix
• 📝 Documentation update
• ✅ Tests (adding or improving test coverage)
• ♻️ Refactor (no behavior change)
• 🎯 New skill (bundled or hub)

Changes Made

• Add _is_ollama() detection method (num_ctx probe + URL heuristics)
• Replace provider == "custom" guard with _is_ollama() in the think=False injection path

How to Test

1. Configure a non-Ollama custom provider (e.g. Mistral, Fireworks) with reasoning_effort: none
2. Send a message — should no longer get HTTP 422
3. Configure an Ollama endpoint with reasoning_effort: none
4. Send a message to a thinking model (e.g. Qwen3) — should still suppress <think/> blocks

Automated: 23 new tests in tests/run_agent/test_think_param_ollama_scope.py:

• 7 _is_ollama() unit tests (num_ctx probe, URL patterns, cloud/LM Studio/vLLM rejections)
• 5 parametrized cloud custom provider tests (Mistral, Fireworks, Together, vLLM, generic)
• 1 LM Studio test (local but not Ollama)
• 7 Ollama positive tests (URL detection, probe detection, effort/enabled combos)
• 3 non-custom provider tests (OpenRouter, Anthropic unaffected)

Checklist

Code

• I've read the Contributing Guide
• My commit messages follow Conventional Commits (fix(scope):, feat(scope):, etc.)
• I searched for existing PRs to make sure this isn't a duplicate
• My PR contains only changes related to this fix/feature (no unrelated commits)
• I've run pytest tests/ -q and all tests pass
• I've added tests for my changes (required for bug fixes, strongly encouraged for features)
• I've tested on my platform: macOS 15 (Darwin 24.6.0)

Documentation & Housekeeping

• I've updated relevant documentation (README, docs/, docstrings) — or N/A
• I've updated cli-config.yaml.example if I added/changed config keys — or N/A
• I've updated CONTRIBUTING.md or AGENTS.md if I changed architecture or workflows — or N/A
• I've considered cross-platform impact (Windows, macOS) per the compatibility guide — or N/A
• I've updated tool descriptions/schemas if I changed tool behavior — or N/A

Screenshots / Logs

pytest tests/run_agent/test_think_param_ollama_scope.py -q
# 23 passed