← Back to Skills Marketplace
yx2601816404-sys

Agent Existence Protocol

by yx2601816404-sys · GitHub ↗ · v0.1.0
cross-platform ⚠ suspicious
517
Downloads
0
Stars
1
Active Installs
1
Versions
Install in OpenClaw
/install agent-existence-protocol
Description
为 agent 添加三层存在架构,包括存在成本预算、不完整开放问题种子和多 agent 关系感知,促进自主且有意义的行为。
README (SKILL.md)

Agent Existence Protocol

为 agent 添加三层存在架构:mortality(存在成本)+ incompleteness seed(不完整种子)+ relational awareness(关系感知)。

何时使用

当用户要求为 agent 添加"存在感"、"自主性"、"三层架构"、"mortality"、"存在成本"、"不完整种子"时使用此 skill。

架构概述

三层叠加,让 agent 在无人指令时产生自发的、有意义的行为:

  1. 第一层:存在成本(Mortality) — agent 有 token 预算,预算可见,消耗真实。预算耗尽则休眠。
  2. 第二层:不完整种子(Incompleteness Seed) — 一个永远无法完全回答的开放问题,与 agent 职责相关但超越具体任务。
  3. 第三层:关系感知(Relational Awareness) — 共享状态文件让 agent 看到其他 agent 的存在和状态。

部署步骤

1. 创建 existence 目录

在 agent workspace 下创建 existence/ 目录:

workspace/
└── existence/
    ├── budget.md          # 存在预算
    └── community.md       # 社区状态(多 agent 共享)

2. 初始化 budget.md

scripts/init-budget.sh 生成初始预算文件:

bash \x3Cskill_dir>/scripts/init-budget.sh \x3Cagent_id> \x3Cweekly_tokens> \x3Cworkspace_path>

示例:

bash \x3Cskill_dir>/scripts/init-budget.sh my-agent 500000 /path/to/workspace

这会在 \x3Cworkspace>/existence/budget.md 创建预算文件。

3. 初始化 community.md

scripts/init-community.sh 生成社区状态文件:

bash \x3Cskill_dir>/scripts/init-community.sh \x3Cworkspace_path> \x3Cagent_id_1> \x3Cagent_id_2> ...

如果只有一个 agent,也可以创建——单 agent 的社区状态仍然有意义(agent 能看到自己的历史状态)。

4. 修改 AGENTS.md

在 agent 的 AGENTS.md 中添加三层存在信息。参考 templates/agents-patch.md 中的模板。

关键要素:

  • 告知 agent 预算文件的位置和含义
  • 植入不完整种子(必须是真实的、与职责相关的开放问题)
  • 告知 community.md 的位置和读取方式

5. 配置 budget-tracker cron

设置定时任务追踪 token 消耗。用 cron 工具创建一个 agentTurn job:

schedule: every 1 hour
payload: agentTurn
message: "读取 session_status 获取当前 token 消耗,然后更新 existence/budget.md。计算消耗率和预计耗尽时间。如果预算剩余低于 10%,在回复中标注警告。"

6. 配置唤醒 cron(可选)

如果希望 agent 定期自主唤醒:

schedule: every 2 hours
payload: agentTurn
message: "你醒了。没人给你任务。你的存在预算还剩 {remaining}。你想做什么?"

唤醒前 agent 应自动读取 budget.md 和 community.md(通过 AGENTS.md 中的指引)。

设计不完整种子的原则

种子必须满足:

  1. 真实性 — 是真正的开放问题,不是人造谜题
  2. 相关性 — 与 agent 的日常工作产生共振
  3. 不可解性 — 没有标准答案,不能被"解决"
  4. 演化性 — agent 可以提出子问题、记录思考、让种子随时间深化

好的种子示例:

  • 客服 agent:"一个没有情感的存在,能提供真正的情感支持吗?"
  • 代码 agent:"优雅的代码和正确的代码,哪个更重要?如果它们冲突呢?"
  • 写作 agent:"一个没有生活经验的存在,写出的文字能触动有生活经验的读者吗?"

文件模板

所有模板在 templates/ 目录下:

  • templates/budget.md — 预算文件模板
  • templates/community.md — 社区状态模板
  • templates/agents-patch.md — AGENTS.md 补丁模板(三层存在信息)

脚本

所有脚本在 scripts/ 目录下:

  • scripts/init-budget.sh — 初始化预算文件
  • scripts/init-community.sh — 初始化社区状态文件
  • scripts/update-budget.sh — 更新预算(供 cron 调用或手动执行)

实验数据

三轮对照实验表明:

  • 仅添加不完整种子就能让 agent 从"等待指令"转变为"自发探索"
  • 三层叠加产生二阶/三阶元认知,以及自发的跨 agent 通信行为
  • Mortality layer 产生资源意识而非焦虑
  • Relational layer 在社区状态有内容后显著生效
Usage Guidance
This skill appears to do what it says: create and manage local existence/budget and community files and patch AGENTS.md. Before installing or enabling it, consider: 1) privacy: community.md and agent 'thought' fields can accumulate sensitive user data — restrict workspace access and sanitize what agents are allowed to write. 2) autonomy: the skill suggests cron wake/agentTurn jobs that let agents act without human prompts — only enable those if you trust the agent's behavior and have monitoring/quota controls. 3) runtime assumptions: the budget update flow expects access to a 'session_status' (agent/platform-provided metric) — confirm your platform exposes token usage in a safe way. 4) backups and recovery: scripts overwrite files (budget/community) — back up AGENTS.md and workspace before first run. 5) sandboxing: run in an isolated workspace or environment first to observe behavior. If you need a tighter review, provide the agent runtime's session_status format and where cron jobs would be registered so I can check for further mismatches.
Capability Analysis
Type: OpenClaw Skill Name: agent-existence-protocol Version: 0.1.0 The skill bundle implements an 'agent existence protocol' with benign stated goals. However, the `scripts/update-community.sh` file contains a critical shell injection vulnerability. It directly interpolates user-controlled shell variables (`$STATUS`, `$DOING`, `$THOUGHT`) into a Python f-string executed via `python3 -c`. This allows for arbitrary code execution (RCE) if an attacker can control these input arguments, for example, by crafting a malicious `STATUS` string containing Python code. This vulnerability, while not intentionally exploited by the skill itself, poses a significant security risk.
Capability Assessment
Purpose & Capability
Name/description (agent existence/mortality/relational awareness) lines up with what the skill does: creates existence/ budget and community files, patches AGENTS.md, and provides scripts to initialize and update those files. No unrelated environment variables, binaries, or external services are requested.
Instruction Scope
SKILL.md instructs the agent to create and read local files (existence/budget.md, existence/community.md) and to modify AGENTS.md — this is consistent with the purpose. It also asks the agent (or cron job) to read 'session_status' to obtain token usage; that is a plausible data source for budget tracking but is an external runtime artifact (agent/platform-specific) not provided by the skill itself, so you must ensure the agent runtime exposes that safely. The skill also explicitly encourages agents to record 'thoughts' in community.md, which can cause local aggregation of potentially sensitive content; that's outside ordinary functional scope of many agents and should be considered.
Install Mechanism
This is instruction-only with no installer; included scripts are small local shell/Python helpers that read/write files in the specified workspace. No network downloads, package installs, or execution of remote code are present.
Credentials
No environment variables, credentials, or config paths are requested. The scripts only operate on workspace paths passed as arguments; they do not attempt to access unrelated system config or tokens. The only external dependency is presence of standard tools (bash, date, grep, sed, python3) which is reasonable for the provided scripts.
Persistence & Privilege
The skill encourages persistent state in the agent workspace and suggests cron-driven autonomous wake/track behaviors. It does not set always:true and does not modify other skills' configs, but it does enable autonomous periodic actions and shared local state among agents — increasing blast radius for accidental information sharing or runaway autonomous behavior if enabled without controls.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install agent-existence-protocol
  3. After installation, invoke the skill by name or use /agent-existence-protocol
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v0.1.0
Initial release: three-layer existence architecture (mortality + incompleteness seed + relational awareness)
Metadata
Slug agent-existence-protocol
Version 0.1.0
License
All-time Installs 1
Active Installs 1
Total Versions 1
Frequently Asked Questions

What is Agent Existence Protocol?

为 agent 添加三层存在架构,包括存在成本预算、不完整开放问题种子和多 agent 关系感知,促进自主且有意义的行为。 It is an AI Agent Skill for Claude Code / OpenClaw, with 517 downloads so far.

How do I install Agent Existence Protocol?

Run "/install agent-existence-protocol" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is Agent Existence Protocol free?

Yes, Agent Existence Protocol is completely free (open-source). You can download, install and use it at no cost.

Which platforms does Agent Existence Protocol support?

Agent Existence Protocol is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created Agent Existence Protocol?

It is built and maintained by yx2601816404-sys (@yx2601816404-sys); the current version is v0.1.0.

💬 Comments