Add obol-stack-dev skill and LLM routing integration tests

[bussyjd]

Summary

• obol-stack-dev skill — comprehensive developer skill for LLM smart-routing through llmspy, covering architecture, CLI wrappers, overlay generation, integration testing, and troubleshooting
• Integration tests — full-cluster tests (//go:build integration) that deploy 3 OpenClaw instances and validate inference through all 3 provider paths (Ollama, Anthropic, OpenAI) via llmspy
• README — expanded Model Providers section documenting obol model setup/status, added OpenClaw commands section

Skill structure

.agents/skills/obol-stack-dev/
├── SKILL.md                              # Entry point
└── references/
    ├── dev-environment.md                # Setup, building, env vars
    ├── llmspy-routing.md                 # 2-tier architecture, 3 provider paths
    ├── obol-cli.md                       # Command map, flag quirks, programmatic usage
    ├── overlay-generation.md             # values-obol.yaml generation
    ├── integration-testing.md            # Test matrix, helpers, writing new tests
    └── troubleshooting.md               # Every gotcha (auth, EOF, labels, flags)

Integration test results (all passing)

Test	Provider	Model	Route	Status
TestIntegration_OllamaInference	Ollama	ollama/gpt-oss:120b-cloud	OpenClaw → llmspy → Ollama	PASS
TestIntegration_AnthropicInference	Anthropic	ollama/claude-sonnet-4-5-20250929	OpenClaw → llmspy → Anthropic	PASS
TestIntegration_OpenAIInference	OpenAI	ollama/gpt-4o-mini	OpenClaw → llmspy → OpenAI	PASS
TestIntegration_MultiInstance	Ollama (x3)	First available	3 instances side-by-side	PASS

All tests go through obol CLI verbs: obol openclaw sync, obol model setup, obol openclaw token, obol openclaw delete --force.

Test plan

• Unit tests pass: go test ./internal/openclaw/
• Integration tests pass: go test -tags integration -v -timeout 15m ./internal/openclaw/
• Cleanup works: test namespaces deleted after each test
• Verify skill is discoverable by agents via .claude/skills/ symlink