← 返回 Skills 市场

Autonomous Cascade

Name: Autonomous Cascade
Author: hupmann86-cell

作者 hupmann86-cell · GitHub ↗ · v1.0.0 · MIT-0

cross-platform ⚠ suspicious

总下载

当前安装

版本数

在 OpenClaw 中安装

/install autonomous-cascade

功能描述

Führt mehrstufige Tasks autonom durch Planen, Ausführen und Bewerten in Schleifen, bis Ziel erreicht oder Budget erschöpft ist.

使用说明 (SKILL.md)

Autonomous Cascade - Plan→Act→Evaluate Loop

Inspiriert von MCP-God-Mode/autonomous_cascade.ts, angepasst für OpenClaw.

Wann nutzen

Wenn eine Aufgabe mehrere Schritte braucht und kein User-Ping nötig ist. Beispiel: "Optimiere Bot-Params bis WR > 55%", "Diagnostiziere Gateway-Problem", "Update Workspace nach OC-Reset"

Loop-Struktur

PLAN:
  goal: \x3CZiel in einem Satz>
  max_rounds: \x3C3-10>
  budget_seconds: \x3C60-600>
  risk: read_only | write_fs | networked
  stop_when: \x3Cklare Bedingung>

ROUND N:
  THINK: Was ist der aktuelle Stand?
  ACT: \x3CTool-Call>
  EVALUATE: Bin ich naehergekommen? Neue Erkenntnisse?
  CONTINUE oder HALT:SOFT (done) oder HALT:HARD (budget/blocked)

Stop-Signale

HALT:SOFT - Ziel erreicht, sauber abschliessen
HALT:HARD - Budget, Fehler oder User needed - sofort stoppen
HALT:BLOCKED - Weiterarbeit benoetigt User-Input

Beispiel: Gateway diagnostizieren

PLAN:
  goal: Gateway auf Port 18789 zum Laufen bringen
  max_rounds: 5
  risk: write_fs
  stop_when: netstat zeigt 18789 LISTENING + /health antwortet

ROUND 1: THINK: Port pruefen
  ACT: netstat -ano | findstr :18789
  EVALUATE: Port down → weiter

ROUND 2: THINK: tmp-Dateien und doppelte Instanzen
  ACT: taskkill /F /IM node.exe
  EVALUATE: gekillt → neu starten

ROUND 3: THINK: Gateway neu starten
  ACT: Start gateway.cmd
  EVALUATE: 18789 LISTENING → HALT:SOFT

Kontext-Budget

max_rounds gilt pro Cascade
Nie mehr als 3 Tool-Calls pro Round
Bei HALT:HARD: User informieren was blockiert ist

Octopaminergic Override (2024 FlyWire)

Falls System-Stress-Level hoch (USDT \x3C $5, EPERM, Loop dead):

Risk downgrade: write_fs → read_only
max_rounds halbieren
User sofort informieren bevor ACT

安全使用建议

This skill implements a powerful autonomous loop that may run local commands and change system state (kill processes, start scripts, write files, use network). Before installing: 1) Confirm which host tools and commands the skill is allowed to run and restrict or whitelist them. 2) Ask the author to remove or explain the 'Octopaminergic Override' (references to USDT < $5 and EPERM are unclear and could cause surprising behavior). 3) Require explicit user approval for any write_fs or network actions, or run the skill in a sandboxed/test environment first. 4) Log all ACT steps and make a dry-run mode available so you can review planned actions before they execute. If the author cannot clarify the external triggers and safeguards, treat the skill as high-risk and avoid granting it access to production systems.

能力评估

ℹ Purpose & Capability

Name and description claim an autonomous multi-round planner, and the SKILL.md implements that loop (PLAN → ROUNDS with THINK/ACT/EVALUATE). The requested capabilities (running system/network/process checks and optionally writing to filesystem or network) are consistent with diagnostic/automation tasks. Minor note: the skill declares no required binaries or credentials even though examples call platform tools (netstat, taskkill, Start gateway.cmd) — that's plausible for an instruction-only skill but should be documented so deployers know what host tools must be available.

⚠ Instruction Scope

The instructions explicitly direct the agent to run system-changing commands (e.g., taskkill, Start *.cmd) and to allow write_fs/networked risk levels. That can be appropriate for automation/diagnostics, but the SKILL.md gives wide discretion about what 'ACT' can run and when to change system state. It also contains an incoherent 'Octopaminergic Override' section that mixes unrelated signals (e.g., 'USDT < $5' as a system-stress indicator, reference to EPERM and 'Loop dead'). Those elements are unclear and could lead to unpredictable or unintended actions if the agent uses them as triggers.

✓ Install Mechanism

No install spec and no code files (instruction-only). This lowers supply-chain risk because nothing is downloaded or written by an installer. Behavior is entirely driven by runtime instructions.

✓ Credentials

The skill does not request credentials, environment variables, or config paths. That is proportionate to an instruction-only planner. Note: the SKILL.md references external signals (e.g., a USDT price check) but provides no guidance about how to obtain that data or what credentials/APIs to use; if the implementation attempts to call external services, those calls and any required keys should be explicitly declared.

✓ Persistence & Privilege

The skill is not always-enabled and does not request elevated persistent privileges. Autonomous invocation is allowed (platform default) — combined with the skill's ability to run system commands, this increases blast radius but is an expected property of autonomous skills.

如何使用

确保已安装 OpenClaw（本地或 Docker 部署）
在对话框中输入安装命令：/install autonomous-cascade
安装完成后，直接呼叫该 Skill 的名称或使用 /autonomous-cascade 触发
根据 Skill 的参数说明提供必要输入，即可获得结构化输出

版本历史

v1.0.0

Initial release of autonomous-cascade: a structured Plan→Act→Evaluate loop for multi-step, unattended tasks. - Enables autonomous multi-step task handling, inspired by MCP-God-Mode, tailored for OpenClaw. - Defines a clear loop structure: PLAN, repeated ROUNDS with THINK, ACT, EVALUATE, and CONTINUE/HALT. - Explicit stop conditions (`HALT:SOFT`, `HALT:HARD`, `HALT:BLOCKED`) for safe operation. - Includes example workflow for diagnosing a gateway issue. - Context-aware safeguards: round, budget, and risk controls, plus "Octopaminergic Override" for high-stress situations.

元数据

Slug autonomous-cascade

版本 1.0.0

许可证 MIT-0

累计安装 1

当前安装数 1

历史版本数 1

常见问题