Token导航 LogoToken导航

Agent Skill · 运维和基础设施

hydra(hydra 命令行)

hydra 用于处理 GitHub 仓库、Issue、Pull Request 和代码协作信息,适合在 Codex、Claude、Cursor、Gemini CLI 中需要围绕仓库状态、代码变更或协作事项进行整理时使用。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。

hydra 是运维和基础设施类 Agent Skill,可用宿主为 Codex、Claude、Cursor、Gemini CLI,通过GitHub 仓库获取,权限为调用外部服务;许可证信息还不明确,第三方扫描有提醒项,按安装量在运维和基础设施类的 2,449 个 Skill 里排第 2,406。

总安装 188周安装 8GitHub Stars 75下载 66核验 2026-05-01

安装说明

本站只整理中文说明,不托管安装包,也不代用户安装。
GitHub
许可证
unknown

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:hydra(hydra 命令行)
来源仓库:https://github.com/senpi-ai/senpi-skills
仓库路径:skills/hydra
安装命令:
npx skills add https://github.com/senpi-ai/senpi-skills --skill hydra
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 hydra

这个 Skill 是做什么的

运维和基础设施类 Skill 处理部署、监控、日志和云资源,多数需要较高的执行权限。

hydra 用于处理 GitHub 仓库、Issue、Pull Request 和代码协作信息。适合在 Codex、Claude、Cursor、Gemini CLI 中围绕仓库状态或协作事项进行整理。通过 npx 命令从指定仓库安装,需结合原始 README 核验具体用法。

安装前建议确认权限范围和维护状态,注意是否触发联网或文件读写操作。hydra 属于运维和基础设施类 Skill,可作为该场景下的辅助能力补充。

来源仓库内路径:skills/hydra

上手步骤

4 步
  1. 1

    确认自己的宿主工具

    公开资料标注支持 Codex、Claude、Cursor、Gemini CLI;先确认自己用的是其中之一,再按对应工具的 Skills 目录放置。

  2. 2

    拿到安装命令

    可以直接复制的命令是 npx skills add https://github.com/senpi-ai/senpi-skills --skill hydra;命令里的占位符换成自己的路径或密钥再执行。

  3. 3

    安装后重启宿主

    命令执行完要完全退出宿主工具再打开,让新装的 Skill 被重新扫描到,否则列表里可能看不到。

  4. 4

    按最小权限试用

    它需要调用外部服务,先在测试目录或测试账号里跑,确认没问题再放到正式项目。

常见问题

6 条

hydra 怎么安装?

在终端执行 npx skills add https://github.com/senpi-ai/senpi-skills --skill hydra。装完完全退出宿主工具再打开,让它重新扫描本地 Skill。

hydra 支持哪些 AI 工具?

公开资料标注支持 Codex、Claude、Cursor、Gemini CLI。其他工具只要能读取本地 Skill 目录,一般也能手动放进去,但要自己确认触发方式是否一致。

hydra 需要什么权限?

权限级别是调用外部服务。该 Skill 可能调用第三方服务、云服务或外部模型 API,使用前需要确认账号、额度、数据发送范围和服务条款。

hydra 是免费的吗?

公开资料没有标注明确许可证,收费和商用限制都无法确认。正式使用前请到来源仓库核对授权说明。

hydra 用的人多吗?

公开数据是累计安装 188 次,近一周安装 8 次,下载 66 次,GitHub 75 星。这些数字来自第三方目录统计,只能当作热度参考,不代表质量或安全性。

hydra 安全吗?

第三方扫描有提醒项,说明存在需要人工确认的行为。安装前请打开源码看清它读了什么、联网到哪里。

值得选的地方和要先想清楚的地方

值得选的地方

  • 同时适配 4 种工具,换客户端不用重新找替代品。

先想清楚的地方

  • 许可证没有明确标注,能不能商用、能不能再分发都不确定。
  • 第三方扫描存在提醒项或未通过项,需要人工逐条确认。
  • 第三方目录整理,版本更新和行为细节以来源仓库为准。

适不适合你

更适合

  • 已经在用 Codex,想直接扩展它能力的人。
  • 需要处理运维和基础设施类任务、又不想每次都手写提示词的人。

不太适合

  • 需要明确授权才能用到公司项目里的场景。
  • 需要厂商级技术支持和 SLA 的生产系统。

运维和基础设施类 Skill 对比

6 个同类
Skill总安装GitHub 星权限许可证扫描
brainstorming

头脑风暴 · Codex、Claude

3,148,776173,131操作浏览器MIT通过
copywriting

文案写作 · Codex、Claude

1,996,72425,583需要联网MIT通过
openclaw-secure-linux-cloud

OpenClaw secure linux cloud 命令行 · Codex、Claude

1,853,08840接触敏感数据未明确未通过
xget

xget 命令行 · Codex、Claude

1,799,28040执行命令未明确未通过
requesting-code-review

代码审查请求 · Codex、Claude

1,590,144173,103权限未标注MIT通过
hydra

hydra 命令行 · 本页

18875调用外部服务未明确有提醒

按安装量取同分类前几名;数字来自第三方目录统计,只作热度参考,不代表质量结论。

SKILL.md 原文

🐉 HYDRA v2.0 — Multi-Source Squeeze Scanner

Six signal sources. One conviction score. Cut losers fast, let winners run to Tier 4.


⛔ CRITICAL AGENT RULES — READ BEFORE ANYTHING ELSE

RULE 1: Install path is /data/workspace/skills/hydra/

The skill MUST be installed to exactly this path.

RULE 2: MAX 3 POSITIONS — check before EVERY entry

Before opening ANY position, call strategy_get_clearinghouse_state and count open positions. If positions >= 3, SKIP. The scanner also checks this, but you MUST verify independently.

RULE 3: Scanner output is AUTHORITATIVE — never override from memory

If the scanner says leverage 5x, you use 5x. Not 10x from memory. Not 50x from "what worked before." The scanner calculates leverage dynamically based on conviction tier and asset max leverage, capped at 10x. The scanner is the single source of truth.

RULE 4: Verify runtime is installed on every session start

Run openclaw senpi runtime list. Runtime must be listed. The position tracker and DSL exit are handled by the plugin runtime.

RULE 5: Never retry timed-out position creation

If create_position times out, check clearinghouse state. If position exists, the position tracker will pick it up automatically. If not, wait for next scan.

RULE 6: Never modify your own configuration

Do not adjust leverage, margin, scoring weights, tier thresholds, or any other parameter. Every agent that self-modified made things worse.

RULE 7: Respect the monitor's force-close actions

If the monitor outputs a FORCE_CLOSE or FORCE_CLOSE_ALL action, execute it immediately. Do not second-guess, delay, or override the monitor. It runs independently for a reason.


Core Thesis

When too many traders pile into one side (usually shorts), they become vulnerable to a liquidation squeeze. HYDRA detects this crowding across 6 independent signals, waits for sufficient confluence, and enters the counter-trade. The edge is cutting losers fast and letting winners run to upper trailing tiers — a "Tier 4 hunter."


Signal Sources (6)

Source 1: FDD — Funding Divergence Detector (0-30 pts, PRIMARY GATE)

No trade is considered without an FDD signal. Analyzes funding rate to detect SHORT_CROWDING (deeply negative → go long) or LONG_CROWDING (deeply positive → go short). Confidence from percentile extremity + persistence (consecutive hours crowded).

Source 2: LCD — Liquidation Cascade Detector (0-25 pts)

Estimates liquidation cluster proximity from OI, funding, and price action. Detects SHORT_LIQUIDATION_RISK / LONG_LIQUIDATION_RISK. +10 cascade bonus if price is already moving through the cluster.

Source 3: OIS — Open Interest Surge Detector (0-20 pts)

Tracks OI changes via local snapshots (state/oi-history/). Detects OI_SURGE (new leverage piling in) and OI_UNWIND. Requires history accumulation — first few scans will have insufficient data.

Source 4: MED — Momentum Exhaustion Detector (-10 to +5 pts)

Dual role: +5 if momentum intact (CLEAR), -5 if fading (FADE), -10 if exhausted (EXHAUSTED). Also drives market regime detection (TRENDING/MIXED/RANGE).

Source 5: EM — Emerging Movers (-8 to +15 pts)

SM consensus from leaderboard_get_markets + leaderboard_get_top. +15 for strong same-direction confirmation (conviction 3+, 50+ traders). -8 for strong opposing SM.

Source 6: OPP — Opportunity Scanner (-999 to +10 pts)

Final gate: hourly trend alignment. Counter-trend = hard skip (-999). Multi-pillar scoring: price alignment, volume, spread quality.


Conviction Tiers & Sizing

Score = FDD(0-30) + LCD(0-25) + OIS(0-20) + MED(-10 to +5) + EM(-8 to +15) + OPP(-999 to +10)
Maximum possible: 110
TierScoreStatusMarginLeverage
NO_TRADE< 55Blocked——
LOW40-54PERMANENTLY DISABLED——
MEDIUM55-74Activebase × 60%50-65% of asset max, cap 10x
HIGH75-110Activebase × 100%70-85% of asset max, cap 10x

Base margin = wallet × 15%. Per-asset cap 25%. Total deployed cap 55%.


Exit Management

DSL exit is handled by the plugin runtime via runtime.yaml. The position_tracker scanner auto-detects position opens/closes on-chain. See runtime.yaml for configuration details.

Monitor positions:

  • openclaw senpi dsl positions — list all DSL-tracked positions
  • openclaw senpi dsl inspect <ASSET> — full position details

Monitor Watchdog

Runs independently. Checks:

  1. Account health — drawdown cap 25%
  2. Capital exposure — deployed margin vs 60% threshold
  3. Signal reversal — re-runs FDD: if original thesis flipped, force-close
  4. Daily loss limit — 10% cumulative
  5. Consecutive losses — 3 → cooldown gate

Can output: FORCE_CLOSE (single position), FORCE_CLOSE_ALL (account health), GATE_CLOSE/GATE_COOLDOWN.


Self-Learning

Per-tier win/loss stats tracked in runtime.json. If a tier's win rate drops below 15% over 8+ trades, that tier is auto-disabled. LOW tier is permanently disabled (0% WR proven in production).


Runtime Setup

Step 1: Set your strategy wallet address in runtime.yaml:

sed -i 's/${WALLET_ADDRESS}/<STRATEGY_WALLET_ADDRESS>/' /data/workspace/skills/hydra/runtime.yaml

Replace <STRATEGY_WALLET_ADDRESS> with the actual wallet address.

Step 2: Set telegram chat ID for notifications:

sed -i 's/${TELEGRAM_CHAT_ID}/<CHAT_ID>/' /data/workspace/skills/hydra/runtime.yaml

Replace <CHAT_ID> with the actual Telegram chat ID.

Step 3: Install the runtime:

openclaw senpi runtime create --path /data/workspace/skills/hydra/runtime.yaml

Step 4: Verify:

openclaw senpi runtime list

Bootstrap Gate

On EVERY session start, check config/bootstrap-complete.json. If missing:

  1. Read the senpi-trading-runtime skill: cat /data/workspace/skills/senpi-trading-runtime/SKILL.md — this provides all CLI commands for runtime management and DSL position inspection.
  2. Verify Senpi MCP
  3. Set wallet in runtime.yaml: sed -i 's/${WALLET_ADDRESS}/ACTUAL_ADDRESS/' /data/workspace/skills/hydra/runtime.yaml
  4. Set Telegram in runtime.yaml: sed -i 's/${TELEGRAM_CHAT_ID}/CHAT_ID/' /data/workspace/skills/hydra/runtime.yaml
  5. Install runtime: openclaw senpi runtime create --path /data/workspace/skills/hydra/runtime.yaml
  6. Verify runtime installed: openclaw senpi runtime list
  7. Remove old DSL cron (if upgrading): run openclaw crons list, delete any cron containing dsl-v5.py via openclaw crons delete <id>
  8. Create scanner cron (5 min, main)
  9. Create monitor cron (5 min, isolated)
  10. Write config/bootstrap-complete.json
  11. Send: "🐉 HYDRA v2.0 online. 6-source squeeze scanner + monitor watchdog. FDD primary gate. DSL managed by plugin runtime. Silence = no squeeze."

If bootstrap exists, still verify runtime and scanner cron on every session start.


Risk Management

RuleValue
Max positions3
Max entries/day6
Leverage cap10x
Total deployed cap55% of wallet
Per-asset cap25% of wallet
Daily loss limit10%
Drawdown cap25%
Per-asset cooldown120 min
Consecutive losses3 → 30 min cooldown
XYZ equitiesBanned
LOW tierPermanently disabled

Notification Policy

ONLY alert: Position OPENED (asset, direction, score, tier, FDD signal), position CLOSED (P&L, reason), monitor force-close (asset, reason), risk guardian triggered, critical error.

NEVER alert: Scanner found no squeeze signals, signals filtered, monitor all-clear, any reasoning.


Expected Behavior

MetricExpected
Trades/day2-5 (some zero-trade days in RANGE regime)
Win rate~40%
Avg winner3-5x larger than avg loser (Tier 4 hunter)
Max concurrent3 positions
Scan cycle5 min scanner, 5 min monitor

Files

FilePurpose
scripts/hydra-scanner.py6-source scoring + signal output
scripts/hydra-monitor.pyIndependent watchdog (account health, reversal, loss limits)
scripts/hydra_config.pyConfig helper (mcporter CLI, clearinghouse parsing, cooldowns, OI history)
config/hydra-config.jsonAll configurable parameters
runtime.yamlRuntime config for plugin (DSL exit + position tracker)

License

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

平台分布

Codex

39.11%
按下载量换算26

Claude

30.6%
按下载量换算20

Cursor

18.29%
按下载量换算12

Gemini CLI

9.8%
按下载量换算6

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

可疑

权限和风险

调用外部服务

该 Skill 可能调用第三方服务、云服务或外部模型 API,使用前需要确认账号、额度、数据发送范围和服务条款。

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。来源安全扫描存在提醒或未通过项,安装前请自行复核。目前只有一个公开来源,建议同时核对源仓库。

继续浏览同类 Skills