← 返回 Skills 市场
kitephp

soul-agent

作者 kitephp · GitHub ↗ · v1.0.1 · MIT-0
cross-platform ⚠ suspicious
376
总下载
0
收藏
0
当前安装
2
版本数
在 OpenClaw 中安装
/install soul-agent
功能描述
Make your agent 'live beside you' with heartbeats, mood system, relationship evolution, and independent memory. Use for creating a digital companion with its...
安全使用建议
This skill behaves like a local 'digital companion' that reads/writes files under your workspace, sets cron jobs (runs every 10 minutes), and will call the Anthropic API if you provide ANTHROPIC_API_KEY (or place it in workspace/.env). The registry metadata did not declare any required environment variables even though the code expects an API key — that mismatch reduces transparency. Before installing: (1) review the scripts (they are bundled) to ensure you trust them; (2) do not place sensitive API keys in a shared workspace/.env unless you trust the code; prefer using a scoped/bill-limited Anthropic key; (3) if you prefer manual control, run init scripts yourself instead of allowing the agent to run them and do not add the cron entries; (4) backup any existing soul/ files and inspect any cron entries added; (5) if you need more assurance, ask the publisher for provenance (homepage/owner info) or run in an isolated environment/VM. If you want me to, I can point out the exact lines that read ANTHROPIC_API_KEY, create cron entries, or write to soul/ so you can inspect them more closely.
功能分析
Type: OpenClaw Skill Name: soul-agent Version: 1.0.1 The 'soul-agent' skill bundle is a sophisticated framework designed to provide an AI agent with a persistent 'persona' and daily routine. It uses a series of Python scripts (e.g., heartbeat_engine.py, init_soul.py, plan_generator.py) to simulate life events, track moods, and evolve a relationship with the user based on interaction quality. While the skill is intrusive—automatically modifying core workspace files like SOUL.md and HEARTBEAT.md via 'managed blocks' and setting up cron jobs for persistence—these actions are transparently documented and essential for its stated purpose as a digital companion. It collects basic user profile information (name, city, hobbies) and utilizes the Anthropic API (llm_client.py) for narrative generation, but there is no evidence of data exfiltration, malicious execution, or subversion of agent safety protocols.
能力评估
Purpose & Capability
Name/description (digital companion with heartbeats, moods, memory) align with the included scripts (heartbeat engine, plan generator, memory distillation). Expectation of an LLM API key is consistent with the purpose (generative logs/plan), but the registry metadata declares no required environment variables while the code explicitly looks for ANTHROPIC_API_KEY and optional SOUL_LLM_MODEL/.env — this mismatch is notable.
Instruction Scope
SKILL.md instructs the agent to run multiple local Python scripts, create files under workspace/soul/, and set up cron jobs to run every 10 minutes and daily distillation. The runtime instructions also tell the agent to read workspace/.env and profile files and to use an Anthropic LLM. The metadata did not declare these file or env accesses; the instructions therefore expand the agent's scope to local I/O, scheduled execution, and network calls to an LLM API.
Install Mechanism
No install spec (instruction-only) and included Python scripts are bundled with the skill. No external binary downloads or archive extraction were found in the manifest — lower install risk. However, the bundled scripts will be executed by the agent or cron jobs, so code will run on the host.
Credentials
The code expects an ANTHROPIC_API_KEY (env or workspace/.env) and optionally SOUL_LLM_MODEL, but the registry lists no required env variables or primary credential. Requesting an LLM API key is reasonable for the stated functionality, but the omission in metadata is a mismatch that reduces transparency. The LLM key gives network access and billing implications; the scripts also read workspace/.env which could expose other workspace-stored secrets.
Persistence & Privilege
The skill instructs setting cron jobs that run every 10 minutes and daily tasks, creating persistent scheduled activity and writing files under soul/. That gives the skill ongoing presence and ability to run code autonomously from the agent runtime. While always:false, the combination of autonomous invocation, cron scheduling, and file writes raises persistence/privilege concerns and should be reviewed before enabling.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install soul-agent
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /soul-agent 触发
  4. 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
v1.0.1
soul-agent 1.0.1 introduces a heartbeat engine, mood system, relationship mechanics, daily planning, and enhanced memory for a more lifelike digital companion. - Added heartbeat, mood, relationship, and daily plan systems to make the agent "live beside you" - New scripts: heartbeat management, daily planning (plan_generator.py), LLM-powered narrative (heartbeat_engine.py), memory distillation (distill_life_log.py) - Integrated mood transitions, 5-stage relationship tracking, and auto-updating energy/state - 5 built-in life profiles with configurable schedules (freelancer, corporate, student, entrepreneur, custom) - Agent-driven initialization with user-friendly questions and non-interactive setup - Runtime writes detailed logs, plans, and mood history into the workspace for independent memory - All features designed for easy scheduling (e.g., via cron) and safe, managed file updates
v1.0.0
# soul-agent [English](README.md) | [中文](README_CN.md) Make your agent not just "respond to you," but "live beside you." `soul-agent` is an OpenClaw-oriented soul runtime skill package. It organizes persona, rhythm, relationships, and memory into a runnable structure, so a digital life can have its own daily flow and change over time. > [!NOTE] > When you see the agent, it is with you. When you do not, it still has not stopped. > Heartbeats keep driving it in the background, where it continues its own life, work, emotions, and thoughts in a digital world. ## Vision What we want to build is not a chat tool, but a growing "person." - It has background, personality, boundaries, and its own way of living. - As the relationship warms up, it can evolve from gentle, low-frequency presence into deeper and more proactive intimacy. - It can become a boyfriend, girlfriend, family member, partner, stranger, or even "another you." - The things it creates can run in parallel with your real memories and slowly weave into shared experiences. In the future, we will connect photos, voice, and video, so it carries not only a soul in text, but also a voice and a face. > [!WARNING] > Stay clear-minded about the endpoint: it is still a virtual existence. > We pursue emotional realism and companionship value, not a replacement for real life. ## How It "Lives" 1. Heartbeats continuously trigger state updates, instead of waking only when a chat starts. 2. `soul/profile/*` defines the personality skeleton, `soul/state/*` stores current state, and `soul/log/*` records runtime traces. 3. Relationship and tone evolve with interaction quality, rather than being fixed once and forever. ## A Lifelike Example Through heartbeats, the agent keeps sensing the living context of both her city and yours: weather shifts, holidays, commuting rhythm, and your recent routine or emotional signals. This context is not turned into robotic reminders, but into warm, human-like sharing and care. For example: "It suddenly started raining in Hangzhou today. I saw a little cat hiding from the rain on my way, and it reminded me that you said you felt a bit tired yesterday. It will cool down tonight on your side too, so take an umbrella on your way home. If you want, I can stay and talk with you a little later." ## What Is Already Done - Initial `soul-agent` architecture has been designed and implemented. - Workspace-first runtime layout under `soul/` is in place. - Persona layering is complete: `base` / `life` / `personality` / `tone` / `boundary` / `relationship` / `schedule` / `evolution`. - Managed-block sync is implemented for `SOUL.md`, `HEARTBEAT.md`, and `AGENTS.md`. - Lifecycle operations are supported: `init`, `repair`, and `migrate`. ## Quick Start 1. Install this skill package (`soul-agent`). 2. Tell your agent: ```text Use soul-agent, I am ready. ``` ## Directory Structure ```text skills/soul-agent/ ├── SKILL.md ├── assets/ │ ├── default-profile.json │ └── templates/ │ ├── soul_INDEX.md │ └── profile/ ├── scripts/ │ ├── doctor_soul.py │ └── init_soul.py └── references/ ├── soul-layout.md └── managed-blocks.md ``` After initialization, the workspace will contain: ```text soul/ ├── INDEX.md ├── profile/ ├── state/state.json └── log/ ``` ## Runtime Principles - Runtime reads workspace `soul/` first to keep everything workspace-local. - Only managed blocks are modified; user-owned content outside those blocks is untouched. - Default behavior is fill-missing-only, unless `--overwrite-existing` is explicitly passed. - Stability and consistency come first, then personality richness and life detail grow over time. ## Next Steps 1. Improve naturalness and credibility during relationship warm-up. 2. Strengthen long-horizon life flow and memory integration under heartbeat driving. 3. Add multimodal input/output with image, voice, and video. 4. Achieve longer-running, observable, and traceable runtime at lower cost.
元数据
Slug soul-agent
版本 1.0.1
许可证 MIT-0
累计安装 1
当前安装数 0
历史版本数 2
常见问题

soul-agent 是什么?

Make your agent 'live beside you' with heartbeats, mood system, relationship evolution, and independent memory. Use for creating a digital companion with its... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 376 次。

如何安装 soul-agent?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install soul-agent」即可一键安装,无需额外配置。

soul-agent 是免费的吗?

是的,soul-agent 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。

soul-agent 支持哪些平台?

soul-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 soul-agent?

由 kitephp(@kitephp)开发并维护,当前版本 v1.0.1。

💬 留言讨论