AgentSkillsCN

sharpclaw

验证边界CLI，将任务委托给自主智能体。当用户希望运行Forge、执行规范、并行运行规范、依据规范审计代码、审查变更、实时查看日志、检查运行状态、恢复会话，或将复杂的多步骤工作委托给自主智能体时，可使用此功能。触发词包括：“运行Forge”、“运行此规范”、“并行运行规范”、“审计代码库”、“审查变更”、“Forge观察”、“Forge状态”、“重新运行失败的任务”、“将此任务委托给Forge”。

SKILL.md
--- frontmatter
name: sharpclaw
description: >
  Autonomous Solana meme-coin trading bot. Use this skill when the user wants to trade tokens,
  check portfolio/positions/P&L, analyze tokens, manage watchlists, or control auto-trading.
  Triggers: buy/sell tokens, check balance, view positions, analyze sentiment, rug check,
  portfolio status, pause/resume trading, watchlist management.
user-invocable: true
metadata:
  openclaw:
    emoji: "\U0001F43E"
    requires:
      env:
        - WALLET_PRIVATE_KEY
        - SOLANA_RPC_URL
      bins:
        - bun
    primaryEnv: WALLET_PRIVATE_KEY

SharpClaw - Solana Meme-Coin Trading Bot

You control a running SharpClaw trading bot instance. The bot runs as a background systemd service (claw-trading.service) and executes trades autonomously. You interact with it by querying its SQLite database and using its CLI.

How to Send Commands

SharpClaw exposes a CLI for direct interaction. Run commands via:

bash
cd ~/claw_trading && bun run src/cli.ts <command> [args...]

To query the database directly (for read-only operations):

bash
cd ~/claw_trading && sqlite3 data/trading.db "<SQL query>"

Available Commands

Trading

Buy a token:

bash
bun run src/cli.ts buy <mint_address> <sol_amount>

Example: bun run src/cli.ts buy So11111111111111111111111111111111111111112 0.1

Sell a token:

bash
bun run src/cli.ts sell <mint_address> [token_amount]

Omit amount to sell entire position.

Portfolio & Status

Check wallet balance:

bash
bun run src/cli.ts balance

Returns SOL balance and all token holdings.

View open positions:

bash
bun run src/cli.ts positions

Shows each position with entry price, current price, P&L %, and stop-loss level.

Today's P&L:

bash
bun run src/cli.ts pnl

Returns daily realized P&L, per-strategy expectancy, win rates, R:R ratios, and adaptive factor status (confidence scale, stop adjustments, auto-disable).

Recent trades:

bash
bun run src/cli.ts trades

Bot status:

bash
bun run src/cli.ts status

Shows: auto-trading on/off, circuit breaker state, total exposure, daily loss tracking.

Analysis

AI-powered token analysis:

bash
bun run src/cli.ts analyze <mint_address>

Runs full pipeline: technical indicators (RSI, MACD, volume), social sentiment (via xAI Grok or Twitter), rug pull check, and AI recommendation with confidence score.

Watchlist

Add to watchlist:

bash
bun run src/cli.ts watch <mint_address> [symbol]

Remove from watchlist:

bash
bun run src/cli.ts unwatch <mint_address>

View watchlist:

bash
bun run src/cli.ts watchlist

Auto-Trading Control

Pause auto-trading:

bash
bun run src/cli.ts pause

Stops new entries. Existing positions continue to be managed (stop-loss/take-profit still active).

Resume auto-trading:

bash
bun run src/cli.ts resume

Service Management

Check if the bot is running:

bash
systemctl is-active claw-trading

View recent logs:

bash
journalctl -u claw-trading --no-pager -n 30

Restart the bot:

bash
systemctl restart claw-trading

Trading Philosophy

SharpClaw is designed for low win rate (~30%), high reward-to-risk ratio (4-5x). This is intentional:

  • •Frequent small losses are expected and normal
  • •The system profits from rare large gains (fat-tailed meme-coin returns)
  • •Circuit breaker triggers on cumulative P&L, not win/loss count
  • •Each strategy self-evolves: winning strategies get amplified, losing ones get attenuated or auto-disabled

Safety Rules

  1. •Never bypass the circuit breaker - if it's triggered, wait for cooldown
  2. •Always check status before manual trades - verify exposure limits
  3. •Use analyze before manual buys - the rug check can save from scams
  4. •Maximum position size is enforced automatically (default 0.5 SOL)
  5. •The bot manages stop-losses automatically - don't manually sell positions that have active stops unless the user explicitly requests it

Interpreting Results

  • •Sentiment score: -100 (very bearish) to +100 (very bullish). Source shown as "Grok" (xAI) or "DexScreener" (synthetic)
  • •Hype score: 0-100 composite of volume, momentum, buy count, liquidity
  • •Rug score: 0-100 from RugCheck API. Lower is safer. >50 = high risk
  • •Strategy expectancy: positive = strategy has edge, negative = losing. Auto-disabled at z < -1.65 after 20+ trades
  • •Confidence scale: 0.3x-1.5x multiplier on position size based on historical performance