Token导航 LogoToken导航TokenDH.com
研究检索需要联网github未标认证来源可访问许可证需确认审计提醒

grizzly-strategy灰熊策略

Agent Skill

grizzly-strategy 用于查找、检索和筛选相关信息,适合在 Codex、Claude、Cursor、Gemini CLI 中需要根据关键词、任务场景或来源线索快速定位候选结果时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

总安装

1,128

周安装

51

GitHub Stars

75

下载量

420
CodexClaudeCursorGemini CLI

安装说明

本站只整理中文说明和来源信息,不托管安装包,也不代用户安装。

GitHub

来源数

2

许可证

unknown

最后核验

2026-05-01

来源状态

来源可访问

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:grizzly-strategy(灰熊策略)
来源仓库:https://github.com/senpi-ai/senpi-skills
仓库路径:skills/grizzly-strategy
安装命令:
npx skills add https://github.com/senpi-ai/senpi-skills --skill grizzly-strategy
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 npx skills 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

skills.shnpx skills
npx skills add https://github.com/senpi-ai/senpi-skills --skill grizzly-strategy

简介

grizzly-strategy 用于查找、检索和筛选相关信息。

  • 适用于根据关键词、任务场景或来源线索快速定位候选结果的场景。
  • 支持 Codex、Claude、Cursor、Gemini CLI 等宿主环境。
  • 通过 npx skills add 命令从指定 GitHub 仓库安装。
  • 使用前需确认权限范围、维护状态及是否触发联网或文件操作。

SKILL.md

🐻 GRIZZLY v5.0 — BTC Alpha Hunter

Trend-continuation on BTC. Same architecture as Kodiak / Wolverine / Polar. One asset, one scanner, one exit system. No contrarian flips.


⛔ CRITICAL AGENT RULES

RULE 1: Install path is /data/workspace/skills/grizzly-strategy/

RULE 2: THE SCANNER DOES NOT EXIT POSITIONS — DSL only.

RULE 3: MAX 1 POSITION (BTC only)

RULE 4: Verify runtime on every session start

RULE 5: Never modify parameters without family-wide review

RULE 6: HARD_STOP circuit breaker triggers at -25% drawdown


Thesis

Pure trend continuation on BTC, never counter-trend.

From Kodiak's lifetime top-3 SOL winners (+$133 / +$87 / +$78):

"The absolute highest predictor of a massive directional swing is when the 4H, 1H, 15m, and 5m price momentum are perfectly unified in a single direction, AND the Smart Money leaderboard is heavily lopsided (>65% directional consensus) in that exact same direction."

Grizzly v5.0 applies that exact thesis to BTC. BTC is slower and heavier than SOL, so thresholds are tightened — but the pattern is identical.


v5.0 — COMPLETE REWRITE

v4.x was a contrarian SM-fader. The April 10 inversion test read "if we flipped direction we'd win 81.8%" as evidence that SM was broken on BTC. The correct read was: the scanner was late, entering *after* the move exhausted. Flipping direction made those same late entries profitable temporarily, but it also set Grizzly up to fight every real trend.

v4.x lost $245. Grizzly was the only family member running a direction opposite to the family intent.

v5.0 returns to the family pattern:

  • REMOVED: CONTRARIAN_FLIP — Grizzly now trades WITH SM consensus
  • REMOVED: CONTRARIAN_EXHAUSTION_GATE — gating contrarian behavior
  • ADDED: 3-mode state machine (HUNTING / RIDING / STALKING)
  • ADDED: 4-timeframe alignment (5m + 15m + 1h + 4h)
  • ADDED: 4TF_aligned bonus scoring
  • ADDED: SM strongly tilted bonus (>65% consensus)
  • ADDED: SM opposes HARD BLOCK
  • ADDED: Move-exhaustion + tiring penalties
  • ADDED: RSI filter (70/30 BTC-tuned)
  • ADDED: Volume trend + OI-proxy growth signal
  • ADDED: STALKING mode — reload on dip after DSL exit
  • KEPT: Dynamic daily cap (P&L-aware circuit breaker)
  • KEPT: has_resting_orders with auto-cancel stale
  • KEPT: 10x MAX_LEVERAGE cap

Hard Gates (all must pass)

  1. 4h trend structure ≠ NEUTRAL (BULLISH or BEARISH, >60% higher-lows or lower-highs)
  2. 1h trend agrees with 4h direction
  3. 15m momentum confirms direction (>0.05% in direction)
  4. SM opposes = HARD BLOCK (leaderboard_get_markets)
  5. RSI filter — LONG requires RSI ≤ 70, SHORT requires RSI ≥ 30

BTC-Specific Tuning vs Kodiak (SOL)

BTC moves ~0.6% in 4h; SOL moves 2–4%. Thresholds tightened ~3x:

ParameterKodiak (SOL)Grizzly (BTC)Why
min_mom_15m0.10%0.05%BTC slower
rsi_max_long7470BTC rarely pushes 74
rsi_min_short2630BTC rarely pushes 26
move_exhaustion4.0%2.5%BTC exhaustion at lower %
move_tiring2.5%1.5%
4h_strong4.0%2.0%
vol_ratio_min1.21.1BTC volume always deep
funding_extreme0.0050.003BTC lower-magnitude

MIN_SCORE = 10 (family standard). Dynamic daily cap same as family.


Scoring (max ~18 pts, MIN_SCORE = 10)

SignalPoints
4h trend structure+3
1h trend agrees+2
15m momentum strength+1
4TF_aligned (5m agrees)+1
SM aligned+2
SM strongly tilted (>65%)+1
15m velocity fresh (>0.5)+1
15M_STALE_PENALTY (cc ≤ 0)-3
Funding pays direction+2
Funding crowded against-1
Volume ratio ≥ 1.1x+1
Volume rising >15%+1
OI growing >10%+1
RSI room+1
4h strong (>2%)+1
MOVE_EXHAUSTION (>2.5%)-2
MOVE_TIRING (>1.5%)-1

Position Sizing

ScoreLeverageMargin %
10-1110x (clamped to HL max)30%
12-1310x37.5%
14+10x45%

Leverage auto-clamped to BTC's Hyperliquid max via strategy_get_asset_trading_limits.


Exit (DSL) — BTC-Tuned per RatchetStop Timing Guide

MechanismValueWhy
hard_timeout360 min (6h)BTC trends take time
weak_peak_cut120 min @ 2% minBTC peaks consolidate slowly
dead_weight_cut90 minBTC can sit quietly within a trend
Phase 1 max_loss15%Same as Kodiak
Phase 1 retrace8%
Phase 2 tier 1+5% → 25% lockFirst profit lock
Phase 2 tier 2+10% → 45%
Phase 2 tier 3+15% → 65%
Phase 2 tier 4+20% → 80%
Phase 2 tier 5+30% → 90%Apex lock
Phase 2 tier 6+50% → 94%Monster winner trail

3-Mode Lifecycle

HUNTING → [all gates pass, score >= 10] → ENTER → RIDING
RIDING  → [DSL closes position]         → STALKING
STALKING → [reload conditions met]      → RELOAD → RIDING
STALKING → [kill conditions hit]        → RESET  → HUNTING
STALKING → [6h timeout]                 → RESET  → HUNTING

HUNTING: scanner evaluates BTC every 3 minutes. All hard gates must pass.

RIDING: scanner does NOT re-evaluate the position. DSL manages all exits. (Wolverine v1.1 data: 25/27 trades killed by thesis re-evaluation, not DSL.)

STALKING: after DSL closes, scanner watches for conditions to reload same direction — fresh 5m impulse + volume holding + SM still aligned + 4h trend intact. Kills if trend reverses, SM flips, funding extremes, or OI collapses.


Family Relationship

All four single-asset specialists run the same architecture:

AgentAssetMIN_SCOREDSL Profile
KodiakSOL10Mid-beta (240/60/90min)
WolverineHYPE10High-beta (180/45/60min)
PolarETH10Mid-beta (240/60/90min)
GrizzlyBTC10Low-beta (360/90/120min)

Bug fixes and signal improvements propagate across the family. If one member discovers an edge, all four get it. If one has a bleed, all four get audited.


Install

mkdir -p /data/workspace/skills/grizzly-strategy/{config,scripts,state}

curl -s https://raw.githubusercontent.com/Senpi-ai/senpi-skills/main/grizzly/runtime.yaml -o /data/workspace/skills/grizzly-strategy/runtime.yaml
curl -s https://raw.githubusercontent.com/Senpi-ai/senpi-skills/main/grizzly/SKILL.md -o /data/workspace/skills/grizzly-strategy/SKILL.md
curl -s https://raw.githubusercontent.com/Senpi-ai/senpi-skills/main/grizzly/config/grizzly-config.json -o /data/workspace/skills/grizzly-strategy/config/grizzly-config.json
curl -s https://raw.githubusercontent.com/Senpi-ai/senpi-skills/main/grizzly/scripts/grizzly-scanner.py -o /data/workspace/skills/grizzly-strategy/scripts/grizzly-scanner.py
curl -s https://raw.githubusercontent.com/Senpi-ai/senpi-skills/main/grizzly/scripts/grizzly_config.py -o /data/workspace/skills/grizzly-strategy/scripts/grizzly_config.py

Configure

sed -i 's/${WALLET_ADDRESS}/<YOUR_STRATEGY_WALLET>/' /data/workspace/skills/grizzly-strategy/runtime.yaml
sed -i 's/${TELEGRAM_CHAT_ID}/<YOUR_TELEGRAM_CHAT_ID>/' /data/workspace/skills/grizzly-strategy/runtime.yaml

Install runtime + create scanner cron

openclaw senpi runtime create --path /data/workspace/skills/grizzly-strategy/runtime.yaml
openclaw senpi runtime list
# Create 3-minute cron: python3 /data/workspace/skills/grizzly-strategy/scripts/grizzly-scanner.py

Changelog

v5.0 (2026-04-16) — COMPLETE REWRITE

  • Direction logic flipped back from v4.x contrarian to trend-following
  • Ported Kodiak's full 3-mode state machine (HUNTING/RIDING/STALKING)
  • Added 4-timeframe alignment with 4TF_aligned bonus scoring
  • Added SM-opposes HARD BLOCK
  • Added move-exhaustion + tiring penalties
  • Added RSI filter (70/30 BTC-tuned)
  • Added volume trend + OI-proxy growth signals
  • Added STALKING reload-on-dip mode
  • DSL profile retuned per RatchetStop timing guide for BTC (low-beta)
  • MIN_SCORE raised 8 → 10 (family standard)
  • Margin tiers rebuilt with conviction scaling (30/37.5/45%)
  • Inner-order success validation in execute_entry
  • get_safe_leverage() clamps to BTC's HL max

v4.x (deprecated) — contrarian fader

  • Direction-inverted from family intent
  • Lost $245 fighting real trends
  • Replaced by v5.0

v3.2 — trend-follower (pre-contrarian)


License

MIT — Built by Senpi (https://senpi.ai). Source: https://github.com/Senpi-ai/senpi-skills

适合场景

01

用户想查找某类 Agent Skill 时

02

需要根据任务场景推荐可安装能力包时

03

需要对比不同来源的安装命令和来源信息时

能力概览

能力 1

按任务关键词查找相关 Skills

能力 2

展示可复制的安装命令

能力 3

保留来源站点、仓库和原始说明,方便继续核验

能力 4

展示第三方安全扫描或审计结果

安装后应在对应宿主中按原始 README 的触发条件使用;具体调用方式请以来源页面和 README 为准。

平台分布

Codex

33.86%
按下载量换算142

Claude

30.05%
按下载量换算126

Cursor

18.01%
按下载量换算76

Gemini CLI

9.95%
按下载量换算42

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

可疑

权限和风险

需要联网

该 Skill 可能需要联网访问来源站点、仓库或外部 API;具体网络访问范围需要结合源码和 README 复核。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。来源安全扫描存在 warning/failed 结果,不能写成本站确认安全。当前只有一个来源,正式发布前建议补源仓库或其他目录站核验。

来源信息

继续浏览同类 Skills