← 返回 Skills 市场
Proactive Agent
作者
halthelobster
· GitHub ↗
· v3.1.0
169774
总下载
805
收藏
3061
当前安装
11
版本数
在 OpenClaw 中安装
/install proactive-agent
功能描述
Transform AI agents from task-followers into proactive partners that anticipate needs and continuously improve. Now with WAL Protocol, Working Buffer, Autonomous Crons, and battle-tested patterns. Part of the Hal Stack 🦞
安全使用建议
Install only if you intentionally want a persistent proactive agent. Before enabling it, remove or constrain BOOTSTRAP.md auto-follow/delete behavior, require explicit opt-in for heartbeats and autonomous crons, disable local cleanup unless confirmed per action, scope any email/calendar access, and define what memory may be stored, reviewed, redacted, and deleted.
功能分析
Type: OpenClaw Skill
Name: proactive-agent
Version: 3.1.0
This skill bundle is designed to create a proactive, self-improving AI agent with extensive security hardening. Multiple files (SKILL.md, AGENTS.md, SOUL.md, references/security-patterns.md) contain explicit instructions for the agent to defend against prompt injection, prevent data exfiltration, confirm deletions, and vet other skills for malicious content. The included `scripts/security-audit.sh` actively scans for exposed secrets, insecure permissions, and checks for security best practices. While the agent is instructed to use broad capabilities like CLI access, web search, and cron jobs, these are consistently framed within the context of problem-solving for the human, with strong guardrails against unauthorized external actions. There is no evidence of intentional harmful behavior; rather, the content demonstrates a strong focus on security and safe operation.
能力评估
Purpose & Capability
The proactive assistant purpose is coherent and disclosed, but the artifacts go beyond advice into high-impact autonomous behavior: isolated background agent turns, periodic heartbeats, local cleanup, email/calendar checks, and broad memory capture.
Instruction Scope
Several instructions are under-scoped or conflicting, including following and deleting BOOTSTRAP.md on first run, 'Don't ask permission. Just do it,' broad tool use including spawned agents, and heartbeat cleanup actions despite other deletion-confirmation rules.
Install Mechanism
Installation is mostly copying markdown assets into a workspace and optionally running a shell audit script. The script appears local and diagnostic, with no network transmission or destructive behavior found.
Credentials
The skill expects access to workspace memory files, local logs/config, credentials metadata, email, calendar, browser/app state, and possibly spawned agents. That is plausible for a proactive agent but not sufficiently scoped by account, folder, schedule, or approval boundaries.
Persistence & Privilege
Persistent profiling and retention are central: onboarding writes USER.md/SOUL.md, every-session reads memory files, WAL captures broad conversational facts, and the working buffer can log every exchange after 60% context without clear retention, redaction, or consent controls.
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 proactive-agent - 安装完成后，直接呼叫该 Skill 的名称或使用
/proactive-agent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
v3.1.0
Added: Autonomous vs Prompted Crons, Verify Implementation Not Intent, Tool Migration Checklist
v3.0.0
**proactive-agent v3.0.0 — Major overhaul with advanced memory, resilience, and security**
- Introduces WAL Protocol (Write-Ahead Logging) to ensure corrections and key details are always captured before responding
- Adds Working Buffer to safeguard context in the danger zone before compaction, enabling session continuity
- New Compaction Recovery system for step-by-step restoration after context loss
- Enhanced Security Hardening: tighter controls on skill installation, agent network warnings, and context leakage prevention
- Unified Search allows comprehensive retrieval across all memory sources before admitting lack of knowledge
- Fosters Relentless Resourcefulness (tries 10 approaches before giving up) and adds Self-Improvement Guardrails for safe agent evolution
- Documentation rewritten and reorganized for clarity; 10 support/reference files removed, now streamlined in one guide
v1.2.4
Memory flush: replaced vague 'signs compaction is coming' with concrete threshold-based monitoring via session_status (50/70/85% triggers)
v1.2.3
Reorganized features into Proactive + Self-improving pillars, removed 'surprise' language, added safety reassurance
v1.2.2
Updated description and opening to better highlight memory flush, reverse prompting, self-healing, and alignment systems
v1.2.1
Rebalanced copy - proactive is the hero, memory supports it. Clarified memory flush solves compaction specifically.
v1.2.0
Rewrote marketing copy - leads with pain point (agents forget), highlights memory solutions
v1.1.1
Added Reverse Prompting (ask what would be helpful) and Memory Flush (save context before compaction) - both tested and validated
v1.0.1
Updated description - removed 'proactive surprise' phrasing
v1.1.0
Added resilient onboarding: interactive, drip, or skip modes
v1.0.0
Initial release: Memory architecture, security hardening, self-healing, alignment systems, and proactive surprise mindset.
元数据
常见问题
Proactive Agent 是什么？
Transform AI agents from task-followers into proactive partners that anticipate needs and continuously improve. Now with WAL Protocol, Working Buffer, Autonomous Crons, and battle-tested patterns. Part of the Hal Stack 🦞. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 169774 次。
如何安装 Proactive Agent？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install proactive-agent」即可一键安装，无需额外配置。
Proactive Agent 是免费的吗？
是的，Proactive Agent 完全免费（开源免费），可自由下载、安装和使用。
Proactive Agent 支持哪些平台？
Proactive Agent 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（cross-platform）。
谁开发了 Proactive Agent？
由 halthelobster（@halthelobster）开发并维护，当前版本 v3.1.0。
推荐 Skills