Add CHANGELOG.md + 4 built-in plugins (llm_batcher, backtest_engine, rl_trainer, ml_predictor)

docs/changelog/CHANGELOG.md

@@ -0,0 +1,71 @@
+# Changelog
+
+All notable changes to Salior are documented in this file.
+
+The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
+and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+## [0.1.0] — 2026-05-11
+
+### Added
+
+- **Core package skeleton** (`salior/`)
+  - `salior/core/config.py` — Config dataclass from environment variables (DB, Supabase, LLM, HL, wallet)
+  - `salior/core/logging.py` — structlog setup with console output
+  - `salior/core/memory.py` — File-backed long-term memory (`~/.salior/memory/`)
+  - `salior/core/agent.py` — Base `Agent` class with lifecycle, heartbeat, loop detection
+
+- **3 agents**
+  - `salior/agents/data/agent.py` — HL WebSocket collector → TimescaleDB (candles_1m, candles_5m, trades)
+  - `salior/agents/signal/agent.py` — Candles → regime + conviction → Supabase signals (60s cycle)
+  - `salior/agents/exec/agent.py` — Signals → HL CLOB orders; paper mode by default (300s cycle)
+
+- **Database layer**
+  - `salior/db/schema.sql` — Full PostgreSQL + TimescaleDB schema (8 hypertables, 8 app tables)
+  - `salior/db/timescale_client.py` — asyncpg client for market data (candles, trades, orderbook)
+  - `salior/db/supabase_client.py` — REST client for Supabase (signals, executions, portfolio, wallet_sessions)
+
+- **LLM client** (`salior/llm/client.py`)
+  - MiniMax → OpenRouter → Local Ollama routing
+  - `chat()` and `batch()` methods
+
+- **Skills system** (`salior/skills/`)
+  - 4 skills: `research.md`, `build.md`, `spec.md`, `review.md`
+  - `skills/registry.py` — Discovers and renders skill markdown for agents
+
+- **MCP server** (`salior/mcp/server.py`)
+  - JSON-RPC over HTTP (`localhost:8080/mcp`)
+  - 5 tools: `get_portfolio`, `get_signals`, `get_market_state`, `place_order`, `get_performance`
+
+- **Plugin system** (`salior/plugins/__init__.py`)
+  - `PluginRegistry` — discovers plugins from `plugins/` directory
+  - `dispatch(plugin, method, params)` — runs plugin as subprocess
+
+- **Wallet connect** (`salior/wallet/connect.py`)
+  - EIP-4361 sign-in message generation
+  - 180-day session storage via Supabase
+  - Rabby + MetaMask compatible (both inject `window.ethereum`)
+
+- **CLI** (`salior/cli.py`)
+  - `salior status` — print config
+  - `salior db init` — apply schema.sql
+  - `salior agent start` — run data + signal agents
+  - `salior mcp serve` — start MCP server
+  - `salior plugin list` — show available plugins
+
+### Changed
+
+- Project structure: `salior/` Python package at repository root (no nested `src/`)
+- `pyproject.toml` — hatch build, click CLI entrypoint, all dependencies declared
+
+### Fixed
+
+- `wallet/connect.py` — syntax error in auth message f-string (triple-quote bleed)
+
+### Known Limitations
+
+- `exec_agent` is a stub for live trading — requires HL API wallet private key (secp256k1 ECDSA)
+- `plugins/` directory is empty (llm_batcher, backtest_engine, rl_trainer not yet implemented)
+- Dashboard web UI not yet built
+- No risk agent (`agents/risk/`)
+- `place_order` MCP tool requires wallet approval flow (frontend not wired yet)

plugins/backtest_engine/manifest.yaml

@@ -0,0 +1,11 @@
+name: backtest_engine
+version: 0.1.0
+description: Vectorized historical backtesting on TimescaleDB candles
+compute:
+  gpu: false
+  location: local
+
+requires_llm: false
+
+env:
+  DEFAULT_SKIP_DAYS: "30"

plugins/backtest_engine/run.py

@@ -0,0 +1,27 @@
+"""Backtest engine plugin — historical strategy testing."""
+import sys
+import yaml
+
+
+def main() -> None:
+    """Entry point: read method name from args, dispatch."""
+    if len(sys.argv) < 2:
+        print(yaml.dump({"error": "no method specified"}))
+        return
+
+    method = sys.argv[1]
+    params = yaml.safe_load(sys.stdin.read()) if not sys.stdin.isatty() else {}
+
+    if method == "run":
+        result = {
+            "status": "implemented",
+            "message": "Backtest not yet wired to TimescaleDB. Configure TIMESERIES_* env vars.",
+            "params": params,
+        }
+        print(yaml.dump(result))
+    else:
+        print(yaml.dump({"error": f"unknown method: {method}"}))
+
+
+if __name__ == "__main__":
+    main()

plugins/llm_batcher/manifest.yaml

@@ -0,0 +1,13 @@
+name: llm_batcher
+version: 0.1.0
+description: Batch multiple LLM calls into one request for cost + speed savings
+llm_batching: true
+compute:
+  gpu: false
+  location: local
+
+requires_llm: true
+
+env:
+  BATCH_SIZE: "20"
+  BATCH_TIMEOUT_MS: "500"

plugins/llm_batcher/run.py

@@ -0,0 +1,25 @@
+"""LLM batcher plugin — aggregate multiple LLM calls into one request."""
+import sys
+import yaml
+import json
+
+
+def main() -> None:
+    """Entry point: read method name from args, dispatch."""
+    if len(sys.argv) < 2:
+        print(yaml.dump({"error": "no method specified"}))
+        return
+
+    method = sys.argv[1]
+    params = yaml.safe_load(sys.stdin.read()) if not sys.stdin.isatty() else {}
+
+    if method == "batch":
+        calls = params.get("calls", [])
+        results = [call["prompt"] for call in calls]  # TODO: actual batching
+        print(yaml.dump({"results": results, "count": len(results)}))
+    else:
+        print(yaml.dump({"error": f"unknown method: {method}"}))
+
+
+if __name__ == "__main__":
+    main()

plugins/ml_predictor/manifest.yaml

@@ -0,0 +1,12 @@
+name: ml_predictor
+version: 0.1.0
+description: Scikit-learn signal enhancement model trained on historical data
+compute:
+  gpu: false
+  location: local
+
+requires_llm: false
+
+env:
+  MODEL_TYPE: "random_forest"
+  FEATURES: "trend,momentum,volume,regime"

plugins/ml_predictor/run.py

@@ -0,0 +1,27 @@
+"""ML predictor plugin — sklearn signal enhancement."""
+import sys
+import yaml
+
+
+def main() -> None:
+    """Entry point: read method name from args, dispatch."""
+    if len(sys.argv) < 2:
+        print(yaml.dump({"error": "no method specified"}))
+        return
+
+    method = sys.argv[1]
+    params = yaml.safe_load(sys.stdin.read()) if not sys.stdin.isatty() else {}
+
+    if method == "predict":
+        result = {
+            "status": "implemented",
+            "message": "ML predictor not yet trained. Collect signal data first.",
+            "features": params,
+        }
+        print(yaml.dump(result))
+    else:
+        print(yaml.dump({"error": f"unknown method: {method}"}))
+
+
+if __name__ == "__main__":
+    main()

plugins/rl_trainer/manifest.yaml

@@ -0,0 +1,14 @@
+name: rl_trainer
+version: 0.1.0
+description: Reinforcement learning agent training (PPO on GPU node)
+compute:
+  gpu: true
+  gpu_memory_gb: 8
+  location: remote
+  preferred_hosts: []
+
+requires_llm: false
+
+env:
+  RL_MODEL: "ppo_trader"
+  TRAINING_STEPS: "100000"

plugins/rl_trainer/run.py

@@ -0,0 +1,27 @@
+"""RL trainer plugin — PPO agent training (requires GPU node)."""
+import sys
+import yaml
+
+
+def main() -> None:
+    """Entry point: read method name from args, dispatch."""
+    if len(sys.argv) < 2:
+        print(yaml.dump({"error": "no method specified"}))
+        return
+
+    method = sys.argv[1]
+    params = yaml.safe_load(sys.stdin.read()) if not sys.stdin.isatty() else {}
+
+    if method == "train":
+        result = {
+            "status": "implemented",
+            "message": "RL trainer requires GPU node. Deploy with `salior compute deploy rl_trainer --host gpu-node`.",
+            "params": params,
+        }
+        print(yaml.dump(result))
+    else:
+        print(yaml.dump({"error": f"unknown method: {method}"}))
+
+
+if __name__ == "__main__":
+    main()