docs/deploy cleanup: skill.md, RUN_MODE=mcp, market map

README.md

@@ -21,7 +21,7 @@
   <img src="https://img.shields.io/badge/strategies-19-C9A84C" alt="Strategies" />
   <img src="https://img.shields.io/badge/tests-1300%2B%20passing-brightgreen" alt="Tests" />
   <img src="https://img.shields.io/badge/license-MIT-blue" alt="License" />
-  <img src="https://img.shields.io/badge/MCP-20%20tools-8A2BE2" alt="MCP" />
+  <img src="https://img.shields.io/badge/MCP-21%20tools-8A2BE2" alt="MCP" />
 </p>
 
 <p align="center">
@@ -550,7 +550,7 @@ hl mcp serve                      # stdio transport (default)
 hl mcp serve --transport sse      # SSE transport
 ```
 
-**20 MCP tools** for account state, trading, APEX/Radar/REFLECT, wallet/setup, safety actions (`schedule_cancel`, `emergency_close_all`), and agent memory/journal helpers. Run `hl mcp serve` to expose them to any MCP host.
+**24 MCP tools** for account state, trading, CFI funding hedges, APEX/Radar/REFLECT, wallet/setup, safety actions (`schedule_cancel`, `emergency_close_all`), and agent memory/journal helpers. Run `hl mcp serve` to expose them to any MCP host.
 
 Fast tools (strategies, builder, wallet, setup, memory, journal, judge) call Python directly — zero subprocess overhead.
 
@@ -619,7 +619,7 @@ Railway build root at the repo root.
 ```
 cli/           CLI commands and trading engine
   commands/    Subcommand modules (run, apex, radar, pulse, guard, reflect, house, ...)
-  mcp_server.py  MCP server (20 tools via FastMCP)
+  mcp_server.py  MCP server (24 tools via FastMCP)
   hl_adapter.py  Direct HL API adapter (live + mock)
   builder_fee.py Builder fee config (HL native BuilderInfo)
   keystore.py    Encrypted keystore (geth-compatible)

cli/mcp_server.py

@@ -28,12 +28,14 @@
 _READ_ONLY_TOOLS = {
     "strategies", "builder_status", "wallet_list", "setup_check",
     "account", "status", "apex_status",
+    "funding_hedge_propose", "funding_hedge_backtest",
     "agent_memory", "trade_journal", "judge_report", "obsidian_context",
     "order_status", "funding_rates",
 }
 # Tools that move funds or cancel/close live orders/positions — handle with care.
 _DESTRUCTIVE_TOOLS = {
-    "trade", "run_strategy", "apex_run", "schedule_cancel", "emergency_close_all",
+    "trade", "run_strategy", "apex_run", "funding_hedge_execute",
+    "schedule_cancel", "emergency_close_all",
 }
 # Everything else (wallet_auto, radar_run, reflect_run) is
 # state-changing-but-safe: neither a pure read nor fund-destructive.
@@ -357,8 +359,9 @@ def _ann(name: str, title: str):
         instructions=(
             "Autonomous Hyperliquid trading CLI — 14 strategies, APEX orchestrator, "
             "REFLECT reviews. Always confirm details with the user before calling "
-            "destructive tools (trade, run_strategy, apex_run, schedule_cancel, "
-            "emergency_close_all). "
+            "destructive tools (trade, run_strategy, apex_run, funding_hedge_execute, "
+            "schedule_cancel, emergency_close_all). "
+            "funding_hedge_execute requires confirmed=true. "
             "emergency_close_all requires confirm=true."
         ),
     )
@@ -796,6 +799,84 @@ def funding_rates(
             args.append("--mainnet")
         return _run_hl(*args, env_overrides=_request_env(ctx))
 
+    @mcp.tool(**_ann("funding_hedge_propose", "Funding hedge proposal"))
+    def funding_hedge_propose(
+        coin: str = "BTC",
+        mainnet: bool = False,
+        ctx: FastMCPContext = None,
+    ) -> str:
+        """Build a CFI v2 funding hedge proposal without placing an order.
+
+        Args:
+            coin: Perp coin to hedge (for example BTC).
+            mainnet: Use mainnet instead of testnet.
+        """
+        args = ["hedge", "propose", coin]
+        if mainnet:
+            args.append("--mainnet")
+        return _run_hl(*args, timeout=60, env_overrides=_request_env(ctx))
+
+    @mcp.tool(**_ann("funding_hedge_backtest", "Funding hedge backtest"))
+    def funding_hedge_backtest(
+        coin: str = "BTC",
+        days: int = 365,
+        notional: float = 1_000_000,
+        ctx: FastMCPContext = None,
+    ) -> str:
+        """Run the reference CFI v2 funding hedge backtest.
+
+        Args:
+            coin: Perp coin to backtest (for example BTC).
+            days: Backtest window; passed through for CLI compatibility.
+            notional: Source perp notional in USD.
+        """
+        return _run_hl(
+            "hedge", "backtest",
+            "--coin", coin,
+            "--days", str(days),
+            "--notional", str(notional),
+            timeout=120,
+            env_overrides=_request_env(ctx),
+        )
+
+    @mcp.tool(**_ann("funding_hedge_execute", "Execute funding hedge"))
+    def funding_hedge_execute(
+        coin: str = "BTC",
+        dry_run: bool = False,
+        mainnet: bool = False,
+        confirmed: bool = False,
+        ctx: FastMCPContext = None,
+    ) -> str:
+        """Build and execute a CFI v2 funding hedge through `hl hedge execute`.
+
+        WARNING: with dry_run=False this can place a real CFI v2 hedge order.
+
+        Args:
+            coin: Perp coin to hedge (for example BTC).
+            dry_run: Preview the order only; no submit and no hedge-state write.
+            mainnet: Use mainnet instead of testnet.
+            confirmed: Must be true after explicit user approval.
+        """
+        if not confirmed:
+            return _json_error("funding_hedge_execute requires confirmed=true after explicit user approval.")
+        env_overrides = _request_env(ctx)
+        error = _context_limit_error(
+            "funding_hedge_execute",
+            env_overrides,
+            mainnet=mainnet,
+            confirmed=confirmed,
+            require_signing=not dry_run,
+        )
+        if error:
+            return _json_error(error)
+        args = ["hedge", "execute", coin]
+        if dry_run:
+            args.append("--dry-run")
+        if mainnet:
+            args.append("--mainnet")
+        args.append("--yes")
+        return _run_hl(*args, timeout=120, env_overrides=env_overrides)
+
     # ------------------------------------------------------------------
     # Self-improvement tools — memory, journal, judge, obsidian
     # ------------------------------------------------------------------

cli/skill.md

@@ -1,6 +1,6 @@
 ---
 name: yex-trader
-description: Autonomous Hyperliquid trading — 14 strategies (MM, momentum, arbitrage, LLM) with APEX multi-slot orchestrator, REFLECT performance review, DSL trailing stops, and builder fee revenue collection.
+description: Autonomous Hyperliquid trading — 19 strategies (MM, momentum, arbitrage, LLM) with APEX multi-slot orchestrator, REFLECT performance review, Guard trailing stops, and builder fee revenue collection.
 user-invocable: true
 argument-hint: "<strategy> [options]"
 allowed-tools:
@@ -17,7 +17,7 @@ metadata:
 
 # YEX Trader
 
-Autonomous Hyperliquid trading via agent-cli. 14 strategies across market making, momentum, arbitrage, and LLM-powered trading. APEX multi-slot orchestrator. REFLECT nightly performance review. Builder fee revenue collection.
+Autonomous Hyperliquid trading via agent-cli. 19 strategies across market making, momentum, arbitrage, and LLM-powered trading. APEX multi-slot orchestrator. REFLECT nightly performance review. Builder fee revenue collection.
 
 ## Quick Start (Agent-Friendly)
 
@@ -137,13 +137,12 @@ hl reflect report [--date 2026-03-03]
 hl reflect history [-n 10]
 ```
 
-### Dynamic Stop Loss (DSL)
+### Guard Trailing Stop
 
 ```bash
-hl dsl start <instrument> [--entry-price 2500] [--direction long] [--preset tight|standard|wide]
-hl dsl check <instrument>
-hl dsl status
-hl dsl presets
+hl guard start ETH-PERP --entry 2500 --size 1 --direction long [--preset tight|moderate]
+hl guard status
+hl guard presets
 ```
 
 ### Radar & Movers
@@ -179,16 +178,16 @@ hl setup bootstrap                   # Auto-create venv and install
 hl setup claim-usdyp                 # Claim testnet USDyP tokens
 ```
 
-### MCP Server (16 Tools)
+### MCP Server (~24 Tools)
 
 ```bash
 hl mcp serve                         # Start MCP server (stdio transport)
 hl mcp serve --transport sse         # Start MCP server (SSE transport)
 ```
 
-Tools: `strategies`, `builder_status`, `wallet_list`, `wallet_auto`, `setup_check`, `account`, `status`, `trade`, `run_strategy`, `radar_run`, `apex_status`, `apex_run`, `reflect_run`, `agent_memory`, `trade_journal`, `judge_report`
+Tools: `strategies`, `builder_status`, `wallet_list`, `wallet_auto`, `setup_check`, `account`, `status`, `trade`, `run_strategy`, `radar_run`, `apex_status`, `apex_run`, `reflect_run`, `schedule_cancel`, `emergency_close_all`, `order_status`, `funding_rates`, `funding_hedge_propose`, `funding_hedge_backtest`, `funding_hedge_execute`, `agent_memory`, `trade_journal`, `judge_report`, `obsidian_context`
 
-## Strategies (14)
+## Strategies (19)
 
 | Name | Type | Description |
 |------|------|-------------|
@@ -204,13 +203,26 @@ Tools: `strategies`, `builder_status`, `wallet_list`, `wallet_auto`, `setup_chec
 | momentum_breakout | Signal | Enters on volume + price breakout above/below N-period range |
 | aggressive_taker | Taker | Directional spread crossing with bias |
 | hedge_agent | Risk | Reduces excess exposure per deterministic mandate |
+| cfi_hedge | Risk | CFI-v2 funding-cost hedge (YEX testnet / Paragon mainnet BTCSWP) |
 | rfq_agent | RFQ | Block-size dark RFQ liquidity |
 | claude_agent | LLM | Claude/Gemini-powered autonomous trading agent |
+| simplified_ensemble | Signal | 6-signal ensemble vote |
+| funding_momentum | Signal | Funding rate mean-reversion with EMA confirmation |
+| oi_divergence | Signal | Price/OI divergence filter |
+| trend_follower | Signal | EMA crossover + ADX trend strength filter |
 
 ## Instruments
 
 - **Standard perps**: ETH-PERP, BTC-PERP, SOL-PERP, etc.
-- **YEX yield markets**: VXX-USDYP (yex:VXX), US3M-USDYP (yex:US3M)
+- **YEX yield markets (testnet)**: VXX-USDYP (`yex:VXX`), US3M-USDYP (`yex:US3M`), BTCSWP-USDYP (`yex:BTCSWP`)
+- **Paragon BTCSWP swap perps (HIP-3)**:
+
+| Network | Instrument | HL coin | Notes |
+|---------|------------|---------|-------|
+| Testnet | `BTCSWP-OSRS` | `osrs:BTCSWP` | Paragon swap perp on the `osrs` dex |
+| Mainnet | `BTCSWP-PARA` | `para:BTCSWP` | Paragon swap perp on the `para` dex |
+
+Shorthand `BTCSWP` resolves by network: testnet → `BTCSWP-USDYP` (YEX yield), mainnet → `BTCSWP-PARA`. Use `BTCSWP-OSRS` or `osrs:BTCSWP` for the explicit Paragon swap perp on testnet.
 
 ## Workflow
 

deploy/hermes-railway/workspace/TOOLS.md

@@ -2,7 +2,7 @@
 
 ## MCP Server: nunchi_trading
 
-The primary tool provider, registered in `config.yaml` under `mcp_servers.nunchi_trading`. Exposes 13 trading tools via Model Context Protocol:
+The primary tool provider, registered in `config.yaml` under `mcp_servers.nunchi_trading`. Exposes 24 trading tools via Model Context Protocol, including:
 
 - `account` — Show HL account state (balance, margin, positions)
 - `status` — Current positions, PnL, and risk state
@@ -17,6 +17,10 @@ The primary tool provider, registered in `config.yaml` under `mcp_servers.nunchi
 - `builder_status` — Check builder fee approval status
 - `wallet_list` — List available wallets
 - `wallet_auto` — Create wallet automatically
+- `funding_rates` — Read current funding rates
+- `funding_hedge_propose` — Build a CFI v2 funding hedge proposal
+- `funding_hedge_backtest` — Run the reference funding hedge backtest
+- `funding_hedge_execute` — Execute or dry-run a CFI v2 hedge; requires `confirmed=true`
 
 ## CLI: hl
 

deploy/openclaw-railway/workspace/TOOLS.md

@@ -11,7 +11,7 @@ mcporter call nunchi_trading.account       # call a tool (no args)
 mcporter call nunchi_trading.trade instrument=ETH-PERP side=buy size=0.1
 ```
 
-Exposes 13 trading tools via Model Context Protocol:
+Exposes 24 trading tools via Model Context Protocol, including:
 
 - `account` — Show HL account state (balance, margin, positions)
 - `status` — Current positions, PnL, and risk state
@@ -26,6 +26,10 @@ Exposes 13 trading tools via Model Context Protocol:
 - `builder_status` — Check builder fee approval status
 - `wallet_list` — List available wallets
 - `wallet_auto` — Create wallet automatically
+- `funding_rates` — Read current funding rates
+- `funding_hedge_propose` — Build a CFI v2 funding hedge proposal
+- `funding_hedge_backtest` — Run the reference funding hedge backtest
+- `funding_hedge_execute` — Execute or dry-run a CFI v2 hedge; requires `confirmed=true`
 
 ## CLI: hl
 

docs/MCP_PRICING_MEASUREMENTS.md

@@ -0,0 +1,77 @@
+# MCP Pricing Measurements
+
+Measured on 2026-07-01 for the hosted MCP tools runner path.
+
+## Harness
+
+Run:
+
+```bash
+python3 scripts/pricing_measure.py --output tmp/pricing-measurement-local.json
+railway run --service hosted-trading-mcp --environment production -- python3 scripts/pricing_measure.py --output tmp/pricing-measurement-runner-env.json
+```
+
+Use `--openrouter-live` only when spending OpenRouter credits is intended.
+
+## Results
+
+Production runner env dry-run:
+
+- `RUN_MODE=mcp`, `HL_TESTNET=true`.
+- `python.import_cli`: 58.6 ms.
+- MCP `strategies`: 118.7 ms, 3,068 response bytes.
+- MCP `funding_hedge_execute` without `confirmed=true`: 0.09 ms refusal, no order path.
+- Railway resource metrics, last 1h: `<0.01 vCPU`, 15 MB memory, 0 MB network, 0 MB disk.
+
+Local Task 7 dry-run (`tmp/pricing-measurement-task7-local.json`):
+
+- `python.import_cli`: 69.7 ms.
+- MCP `tools/list`: 0.01 ms, 24 hosted runner tools surfaced by the JSON-RPC wrapper.
+- MCP `setup_check`: 7.9 ms.
+- MCP `strategies`: 166.8 ms, 3,068 response bytes.
+- MCP `trade` without signing context: 0.09 ms refusal. This is the safe
+  noninteractive confirmation/hang check; no subprocess order path was entered.
+- MCP `funding_hedge_execute` without `confirmed=true`: 0.03 ms refusal.
+
+The pricing harness now emits the full Task 7 classification:
+
+- Free/read: 15 tools.
+- Paid compute/inference cost centers: 5 tools.
+- Safety-gated/fund-moving/wallet-write: 7 tools, or 6 if `wallet_auto` is
+  excluded from the costable 26-tool surface because it is disabled on the
+  hosted keyless runner.
+- Recommended beta free cap: about 20 hosted MCP discovery/read calls before
+  subscription or upgrade nudges.
+
+## Economics
+
+Mode 1, hosted MCP tools:
+
+- `C_seat` is not computed yet. Railway metrics expose CPU/memory/network, but not monthly billing cost. Set `RAILWAY_SHARED_RUNTIME_MONTHLY_USD` or pass `--runtime-monthly-usd` once billing data is available.
+
+Mode 2, hosted MCP tools plus Nunchi/OpenRouter inference:
+
+- No live OpenRouter spend was measured because `OPENROUTER_API_KEY` is not present in the runner env and `--openrouter-live` was not run.
+- Current inference budgets remain inputs only: Starter `$10`, Growth `$50`, Team `$250`.
+- Anchor estimates from the Task 7 prompt:
+  - `openai/gpt-4.1-mini` at about `$0.0002` per heartbeat gives about
+    50,000 / 250,000 / 1,250,000 heartbeats for Starter / Growth / Team.
+  - `openrouter/auto` at about `$0.0037` per heartbeat gives about
+    2,703 / 13,514 / 67,568 heartbeats.
+  - Fusion at about `$0.033` per capped run, provided as about 146x mini, gives
+    about 303 / 1,515 / 7,576 runs.
+
+Mode 3, clone/local plus builder economics:
+
+- No funded-wallet fill measurement was run because the production runner has no `HL_PRIVATE_KEY`, `HL_KEYSTORE_PASSWORD`, or `~/.hl-agent/env`.
+- Formulaic builder-fee economics at the default `BUILDER_FEE_TENTHS_BPS=100` are `$100` per `$100,000` notional and `$1,000` per `$1,000,000` notional.
+- `trade` itself should remain free or low-friction from a pricing standpoint:
+  it is safety-sensitive, not inference-heavy, and it is the path that can
+  produce builder-code/builder-fee economics. Gate it with confirmation,
+  builder-code validation, network consent, size limits, and signing context.
+
+## Blockers
+
+- Missing Railway monthly billing cost input for Mode 1 `C_seat`.
+- Missing `OPENROUTER_API_KEY` for Mode 2 inference spend.
+- Missing funded-wallet/HL signing credentials for live fills and builder-fee realization.