# Phase 8.2: LLM Endpoint / Thread Tools
**Estimated Time:** 1-2 days
**Dependencies:** Phase 8.1
**Status:** 📋 Planning
## Overview
Create data-driven tool configurations for LLM endpoints (Anthropic, OpenAI) and thread wrappers. These are pure YAML configs that chain to `http_client` - no code, just data.
## What This Phase Accomplishes
- Creates `anthropic_messages` tool config (base LLM endpoint)
- Creates `anthropic_thread` tool config (thread wrapper with transcript)
- Creates `openai_chat` and `openai_thread` equivalents
- Tests streaming to file + return sinks
## Files to Create
- `.ai/tools/llm/anthropic_messages.yaml` (new - HTTP tool, YAML config)
- `.ai/tools/llm/openai_chat.yaml` (new - HTTP tool, YAML config)
- `.ai/tools/threads/anthropic_thread.yaml` (new - HTTP tool, YAML config)
- `.ai/tools/threads/openai_thread.yaml` (new - HTTP tool, YAML config)
- `tests/integration/test_thread_tools.py` (new)
**Note:** LLM endpoint tools are HTTP tools (YAML configs), not runtime tools. Only runtime tools (like sinks) are Python-only.
## Architecture Notes
**Key Principle:** These are pure config tools. They chain to `http_client` primitive. No Python code needed - just YAML configuration.
**Tool Chain:**
```
anthropic_thread → anthropic_messages → http_client (primitive)
```
**Config Merging:** Child configs override parent configs. `anthropic_thread` inherits from `anthropic_messages` and adds thread-specific config (transcript path, etc.).
## Task Order
1. Create anthropic_messages tool config
2. Create anthropic_thread tool config
3. Create openai_chat tool config
4. Create openai_thread tool config
5. Write integration tests
## Success Criteria
- [ ] All four tool configs exist
- [ ] Tool chains resolve correctly
- [ ] Streaming works with file + return sinks
- [ ] Body templating works with parameters
- [ ] Transcript storage works for thread tools
- [ ] Tests verify end-to-end streaming
## Related Sections
- Doc lines 505-630: Layer 2 (LLM Endpoint Tools)
- Doc lines 632-696: Layer 3 (Thread Tools)
