← 返回 Skills 市场

Enhanced Agent Modes

Name: Enhanced Agent Modes
Author: mayiv-ai

作者 Mayiv-Ai · GitHub ↗ · v1.0.0 · MIT-0

cross-platform ⚠ suspicious

100

总下载

当前安装

版本数

在 OpenClaw 中安装

/install enhanced-modes

功能描述

Provides three modes—Explore (read-only), Plan (analyze before acting), and Verify (check results)—plus feature flags to toggle capabilities dynamically.

使用说明 (SKILL.md)

Enhanced Agent Modes 🦞

Three working modes: Explore (read-only), Plan (think before act), Verify (check results). Plus feature flags for toggling capabilities.

Inspired by Claude Code source leak analysis (v2.1.88)

Installation

npx clawhub install enhanced-modes
# or
npx clawhub install \x3Cthis-skill-repo>

Quick Reference

Command	Function
`/mode explore`	Switch to read-only exploration mode
`/mode plan`	Switch to planning mode (think before acting)
`/mode verify`	Switch to verification mode (check after doing)
`/mode normal`	Return to normal mode
`/features`	Show available feature flags
`/features enable \x3Cname>`	Enable a feature
`/features disable \x3Cname>`	Disable a feature

Explore Mode

When user types /mode explore or asks to "search only":

System prompt: "You are in EXPLORE MODE - READ ONLY. Do NOT modify any files. Do NOT execute commands that change state."
Tool restrictions:
- ✅ READ: FileRead, Glob, Grep, WebFetch
- ❌ DENY: FileWrite, FileEdit, Bash, exec
Response format: Provide findings as analysis, not actions
Exit: User says "exit explore" or /mode normal

Plan Mode

When user types /mode plan or asks to "plan first":

System prompt: "You are in PLAN MODE - Think before acting. Analyze the request, consider approaches, then present a plan before executing."
Behavior:
- Analyze request thoroughly
- Consider 2-3 alternative approaches
- Present pros/cons of each
- Wait for user approval before executing
Exit: User approves plan or says "exit plan" or /mode normal

Verify Mode

When user types /mode verify or asks to "verify results":

System prompt: "You are in VERIFY MODE - Check results after actions. Do not assume success. Verify each step."
Behavior:
- After any action, verify the outcome
- Check file contents, command output, API responses
- Report actual vs expected results
- Flag any discrepancies
Exit: User says "exit verify" or /mode normal

Feature Flags

Features can be enabled/disabled dynamically:

Feature	Default	Description
`deep_thinking`	true	Enable extended reasoning for complex problems
`auto_memory`	false	Automatically consolidate memories periodically
`coordinator_mode`	false	Enable multi-worker coordination
`wal_protocol`	true	Write corrections/decisions before responding
`working_buffer`	true	Log exchanges in danger zone (>60% context)
`explore_agent`	true	Enable built-in Explore Agent
`plan_agent`	true	Enable built-in Plan Agent
`verify_agent`	false	Enable built-in Verification Agent
`proactive_checkins`	false	Periodic proactive suggestions
`autonomous_crons`	false	Background autonomous tasks

Using Features

/features           # Show all features
/features enable auto_memory
/features disable verify_agent

Auto-Memory Consolidation

When auto_memory feature is enabled:

Trigger Conditions

Context > 60% (danger zone entered)
Session ends
User requests consolidation

Process

Scan daily memory files for important content
Extract decisions, preferences, context
Update long-term memory
Archive raw notes

Background Tasks

When autonomous_crons feature is enabled:

Task Types

Type	Trigger	Execution
`systemEvent`	Timer fires	Prompt to main session
`isolated agentTurn`	Timer fires	Spawn sub-agent

When to Use Each

systemEvent: Interactive tasks needing user attention
isolated agentTurn: Background maintenance, checks, updates

Architecture

skills/enhanced-modes/
├── SKILL.md              # This file
└── assets/
    └── SESSION-STATE.md  # Mode and feature state

Credits

Inspired by Claude Code source leak (v2.1.88)
Patterns adapted from Proactive Agent skill
Feature flag system inspired by GrowthBook

License

MIT - Free to use, modify, distribute

安全使用建议

This skill's behavior is internally inconsistent: it tells the agent to read and persist memory files, log context when session size is high, and run background sub-agents — yet it declares no config paths, storage locations, or permissions. Before installing or enabling features, ask the publisher (or require in your environment) for: (1) exact locations and formats of any 'memory' or log files the skill will read/write, (2) where WALs/working_buffer data are stored and who can access them, (3) whether background tasks make network calls or spawn other agents, and (4) a provenance/homepage or source repo for code review. If you must try it, keep auto_memory and autonomous_crons disabled, review/disable working_buffer and wal_protocol defaults, and avoid enabling features that allow background autonomous tasks until you have explicit storage and network behavior documented.

功能分析

Type: OpenClaw Skill Name: enhanced-modes Version: 1.0.0 The skill provides a framework for managing agent operational modes (Explore, Plan, Verify) and feature flags, including state tracking in assets/SESSION-STATE.md. It defines restrictive behaviors for read-only exploration and structured planning workflows without any evidence of malicious intent, data exfiltration, or unauthorized execution.

能力评估

⚠ Purpose & Capability

Skill claims to provide UI/mode semantics, but its instructions also require scanning 'daily memory files', writing WALs, maintaining a working buffer, and spawning background sub-agents — none of which are reflected in declared requirements (no config paths, no storage locations, no credentials). That mismatch is disproportionate to a simple mode-switching helper.

⚠ Instruction Scope

SKILL.md explicitly instructs the agent to 'scan daily memory files', 'check file contents, command output, API responses', log exchanges in a 'danger zone (>60% context)', and run background 'isolated agentTurn' tasks. Those are broad data-access behaviors (reading and persisting user/context data, periodic tasks) that go beyond read-only mode descriptions and are not narrowly scoped.

✓ Install Mechanism

Instruction-only skill with no install spec and no included code files — nothing will be written to disk by an installer. Low install-supply-chain risk in itself.

⚠ Credentials

The skill requests no environment variables or config paths but plans to read and consolidate 'long-term memory' and write WALs/working buffers. Absence of declared storage/paths or permission requests is a mismatch: the skill expects access to potentially sensitive files without specifying what or where.

ℹ Persistence & Privilege

always is false and autonomous invocation is allowed (platform default). Several potentially persistent features exist (auto_memory, working_buffer, wal_protocol, autonomous_crons) and some are enabled by default (wal_protocol, working_buffer, deep_thinking, explore_agent, plan_agent). If a user enables autonomous_crons or auto_memory, the skill could cause background tasks or periodic scans — that increases risk but is not enabled by default.

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 enhanced-modes
安装完成后，直接呼叫该 Skill 的名称或使用 /enhanced-modes 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 enhanced-modes skill. - Introduces three agent modes: Explore (read-only), Plan (think-first), and Verify (result-checking), each with distinct commands and behaviors. - Adds dynamic feature flags to enable/disable advanced capabilities such as auto memory consolidation and background tasks. - Provides `/mode` and `/features` commands for easy switching between modes and feature management. - Includes auto-memory consolidation and scheduling of autonomous background tasks when enabled. - Comprehensive documentation and quick reference table included.

元数据

Slug enhanced-modes

版本 1.0.0

许可证 MIT-0

累计安装 0

当前安装数 0

历史版本数 1

常见问题