← 返回 Skills 市场
6917
总下载
24
收藏
0
当前安装
10
版本数
在 OpenClaw 中安装
/install hippocampus-memory
功能描述
Persistent memory system for AI agents. Automatic encoding, decay, and semantic reinforcement — just like the hippocampus in your brain. Based on Stanford Generative Agents (Park et al., 2023).
安全使用建议
Install only after reviewing the privacy and scripting risks. Avoid enabling cron or whole-history processing unless you explicitly want historical conversations summarized into persistent plaintext memory. Do not use this around secrets or sensitive personal data until the code-injection, prompt-injection, dashboard escaping, retention, permissions, approval, and deletion controls are fixed.
功能分析
Type: OpenClaw Skill
Name:
Developer:
Version:
Description: OpenClaw Agent Skill
Suspicious High-Entropy/Eval files: 10
The OpenClaw Hippocampus skill is designed to provide persistent memory for AI agents by processing conversation history. Its core functionality involves reading all session transcripts (`preprocess.sh`), storing and managing memories in `index.json` within the agent's workspace, and running cron jobs for encoding and decay. While accessing conversation history is a sensitive operation, it is fundamental to the skill's stated purpose. The instructions for the AI agent (found in `SKILL.md`, `install.sh`'s `--agent-turn` arguments, `summarize-pending.sh`, and `agents/hippocampus-agent.md`) are detailed but consistently focused on memory management tasks, without any evidence of intentional harmful behavior, data exfiltration to external endpoints, or unauthorized system manipulation. All observed capabilities, including file access and agent instructions, are plausibly and transparently aligned with the benign objective of building an AI memory system.
能力评估
Credentials
The skill reads all JSONL session transcripts for the selected OpenClaw agent, writes plaintext intermediate and long-term memory files under the workspace, and generates a dashboard from memory and related brain-skill state without documented retention, deletion, filtering, or file-permission safeguards.
Install Mechanism
The manifest install path recommends running install.sh with cron, and the installer can add recurring encoding and decay cron jobs, chmod scripts, initialize memory files, generate dashboard output, and recommend automatic RAG loading; no remote payload retrieval was found.
Instruction Scope
The instructions disclose memory capture and recall, but raw transcript-derived signals are presented adjacent to agent instructions for summarization, creating a prompt-injection path into persistent memory; self-reflection prompts also encourage durable agent self-concept formation.
Persistence & Privilege
Persistence is central to the purpose, but it is high impact: cron jobs continue processing over time, HIPPOCAMPUS_CORE.md is intended for future context loading, unquoted heredocs allow Python code injection through parameters such as recall queries or thresholds, and dashboard content is rendered unsafely with innerHTML.
Purpose & Capability
The package coherently implements a persistent OpenClaw memory system, but it captures user, relationship, emotional, and agent-identity facts into long-lived memory and core context; the claimed default limit of the last 100 signals is not enforced by the encoding pipeline on a fresh install.
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 hippocampus-memory - 安装完成后，直接呼叫该 Skill 的名称或使用
/hippocampus-memory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
v3.8.6
- Skill name updated to "hippocampus-memory" and title clarified.
- Description rewritten for clarity and conciseness.
- Metadata version bumped to 3.8.6.
- Documentation restructured for easier reading; headings clarified and formatting improved.
- No changes to scripts, features, or usage—documentation improvements only.
v3.8.5
- Renamed skill from "hippocampus-memory" to "hippocampus" and simplified the title and description.
- Updated `metadata.openclaw.version` from 3.8.3 to 3.8.4.
- Clarified that the skill runs separately from the main agent.
- Minor streamlining and restructuring throughout documentation for clarity and conciseness.
- No functional or implementation changes—documentation update only.
v3.8.3
v3.8.3 - Multi-session memory support
- FIXED: preprocess.sh now scans ALL session files (not just one)
- NEW: Datetime watermark (lastProcessedTimestamp) for cross-session tracking
- NEW: CHANGELOG.md added
- Renamed to "Hippocampus - Memory System" for better discoverability
v3.7.3
Fix: Dashboard generator was using wrong variable for amygdala data (HIPPOCAMPUS_FILE defined twice, second line overwrote first)
Fix: Removed hardcoded Europe/Warsaw timezone from consolidation-guide.md
v3.7.2
- Fixed the skill version number in metadata (now matches release version 3.7.2).
- No other user-facing changes.
v3.8.0
- Adds scripts/generate-dashboard.sh to generate a unified brain dashboard (HTML) for all installed brain skills.
- Introduces a "Brain Dashboard" feature showcasing memory and cognitive skills, with auto-detection of related skills and install guidance for missing ones.
- Dashboard reads avatar and name from IDENTITY.md and is regenerated during install and cron runs.
- Updated documentation to describe the new dashboard and include notification instructions for agents after installation.
v3.7.0
- Adds semantic encoding pipeline: new memories are scored, checked for duplication, and either reinforced or summarized using LLM.
- Replaces manual `reinforce.sh` with automatic semantic reinforcement during encoding (no more manual boosts).
- Updates and expands available scripts: adds `encode-pipeline.sh`, removes `reinforce.sh` and duplicate checking scripts.
- Introduces new install options for controlling signal processing and enables whole-history encoding.
- Shifts cron workflow: encoding now runs periodically with reinforcement logic; decay and consolidation remain scheduled.
- Improves documentation—clarifies memory lifecycle, file structure, and agent integration for OpenClaw.
v3.1.3
- Added prompt for encoding memory signals (`prompts/encode-signals.md`)
- Added duplicate memory checker script (`scripts/check-duplicate.py`)
- Updated documentation to introduce the "AI Brain Series" section and list related skills
- Bumped version to 3.4.0
v3.1.2
Docs: Explain background processing concept
v3.1.1
Memory system with importance scoring, decay, reinforcement
元数据
常见问题
Hippocampus 是什么？
Persistent memory system for AI agents. Automatic encoding, decay, and semantic reinforcement — just like the hippocampus in your brain. Based on Stanford Generative Agents (Park et al., 2023). 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 6917 次。
如何安装 Hippocampus？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install hippocampus-memory」即可一键安装，无需额外配置。
Hippocampus 是免费的吗？
是的，Hippocampus 完全免费（开源免费），可自由下载、安装和使用。
Hippocampus 支持哪些平台？
Hippocampus 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 Hippocampus？
由 ImpKind（@impkind）开发并维护，当前版本 v3.8.6。
推荐 Skills