Switch LLM to crk1fuz8 deployment, bust cache

Add full contract spec to base prompt, GitHub provider prompt, and tool descriptions

Fix top_n type coercion for models returning string integers

Fix planner ignoring tool calls when finish_reason is stop

Switch LLM to wh63jnnl deployment, bust cache

Upgrade to hierarchical schema and build_request() API

Fix empty tool arguments from Fireworks model appending <|call|>

Remove project-level mcp.json (configured at user level instead)

Add mcp-a2a-gateway config and seed script

Switch LLM model to pyroworks deployment

Add .DS_Store to .gitignore

Add structured execution trace to A2A task responses

Add CACHEBUST ARG to always invalidate Docker cache at uv sync

Update fireaction-contracts to latest (Square path fix)

Fix: add trailing slash to multi-provider agent card URLs (avoid 307 redirect)

Remove stripe from default providers (no credentials)

Default to all providers (skip those with missing credentials)

Single RUN with secret mount: git config + uv sync together, no --frozen

Skip installing project itself (use PYTHONPATH instead)

Fix: use --mount=type=secret for GITHUB_TOKEN (not ARG)

Fix Dockerfile: separate RUN steps so uv sync errors aren't swallowed

Add PYTHONPATH for src layout in Docker

Use python -m for CMD (entry point script not available in venv)

Use full path for fireaction-a2a in CMD

Use ARG for GITHUB_TOKEN instead of secret mount

Support multiple providers in one server

Make provider configurable via PROVIDER env var

Switch LLM to gpt-oss-20b for faster inference

Fix agent card URL: use public HF Space URL instead of internal 0.0.0.0

Configure Fireworks AI models and fix Docker build for private repos

Install git in Docker image for git-based dependencies

Add HF Spaces YAML metadata to README

Initial implementation: fireaction A2A agent services