← 返回 Skills 市场

Agent Memory Journal

Name: Agent Memory Journal
Author: misolith

作者 misolith · GitHub ↗ · v0.2.3 · MIT-0

cross-platform ✓ 安全检测通过

141

总下载

当前安装

版本数

在 OpenClaw 中安装

/install agent-memory-journal

功能描述

Durable episodic memory for agents — file-based, inspectable, and easy to review. Use when an agent needs to keep a working journal of what happened, what ma...

使用说明 (SKILL.md)

Skill: Agent Memory Journal

Manage and retrieve long-term memory using a 3-tier structure (Hot, Warm, Cold).

Primary integration surface: the Python Journal API. The CLI is a convenience layer for shell and cron workflows.

Tools

agent-memory-journal note \x3Ctext>: Add episodic note.
agent-memory-journal remember \x3Ctext> --category \x3Ccat>: Add core memory.
agent-memory-journal search --query \x3Cq>: Search memory.
agent-memory-journal forget \x3Cid>: Supersede a memory.
agent-memory-journal ingest: Run promotion and rebuild cycle.

Core Categories

decision: Strategic choices.
constraint: Hard rules or limitations.
gotcha: Lessons learned or bugs.
preference: User or agent preferences.
capability: New skills or tools.

Guidelines

Episodic first: Use note for observations.
Explicit curation: Use remember for facts that must persist.
Atomic pinning: Use --pinned for items that MUST be in the configured hot file.
Verification: Run doctor periodically to ensure integrity.

Agent installation

See docs/agent_install.md for:

workspace installation
CLI symlink setup
startup contract for agents
hot promotion target override via .memory/config.json

安全使用建议

This package implements a local, file-backed memory system: it will create and update files under the configured memory root (default .memory) and a hot file (default AGENT.md). There is no network activity or secret access in the code. Before installing, confirm you trust code that will write to your workspace because it will: create/append to files in .memory (episodic/core/sessions/archive/index) and may update a hot file in the workspace or the workspace parent (config's hot_path can point to AGENT.md in parent). If you want tighter containment, set the memory root to a directory you control and review .memory/config.json (hot_path) so the skill cannot write files outside that directory. If you need extra assurance, inspect the remaining truncated files locally; the visible code uses atomic temp-file writes and path validation, and there are no network calls or credential accesses.

功能分析

Type: OpenClaw Skill Name: agent-memory-journal Version: 0.2.3 The bundle is a legitimate and well-architected memory management system for AI agents, implementing a tiered storage structure (Hot, Warm, and Cold). It includes proactive security features such as prompt-injection filtering in `agent_memory/security.py`, path validation to prevent directory traversal when configuring the 'hot' memory file, and SHA256-based integrity verification in `agent_memory/doctor_v2.py`. The code follows safe development practices, such as using atomic file writes via temporary files, and lacks any indicators of data exfiltration, remote execution, or malicious intent.

能力评估

✓ Purpose & Capability

The name/description (durable episodic memory) matches the provided code and SKILL.md. All required behavior (Journal API, CLI, file-based hot/warm/cold tiers) is implemented in the package; nothing in the package asks for unrelated capabilities (no cloud, no unrelated service credentials).

✓ Instruction Scope

SKILL.md instructs use of the Journal API and CLI (note/remember/search/forget/ingest) and references local workspace installation and a config file. The runtime instructions and code limit actions to filesystem reads/writes under the configured memory root and related review/doctor/ingest flows; there are no instructions to read unrelated system files or contact external endpoints.

✓ Install Mechanism

No install spec is declared (instruction-only at metadata level), but the skill bundle includes Python source files. There are no remote download URLs or install scripts in the package. The code will be executed locally as Python code if the agent loads it; nothing in the package pulls code from untrusted hosts.

✓ Credentials

The skill does not request environment variables, credentials, or config paths outside its memory root. The CLI uses AGENT_MEMORY_ROOT only as a default for the root path; no secrets or unrelated tokens are required.

✓ Persistence & Privilege

The skill has standard local persistence (writes files under a memory root, creates an index/manifest and AGENT.md hot file). always:false and agent autonomy flags are default. It does not modify other skills or global agent settings beyond creating files in the workspace and an index under the configured root.

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 agent-memory-journal
安装完成后，直接呼叫该 Skill 的名称或使用 /agent-memory-journal 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

v0.2.3

Various fixes and enhancements

v0.2.1

Initial migration to Python module format with expanded features and new CLI. - Migrated to a modular Python package structure. - Added multiple new components: episodic recall, ingestion, review, decay, categorization, analytics, normalization, storage, and security modules. - Replaced previous config and example files with `docs/agent_install.md` and a project-level `pyproject.toml`. - Introduced a 3-tier memory model (Hot, Warm, Cold) with curated categories. - Updated CLI for more granular operations: add notes, core memories, search, forget, and ingest. - New Python API (`Journal`) available for integration and automation.

v0.1.0

Initial release of agent-memory-journal. - Provides a file-based, inspectable, and reviewable episodic memory journal for agents. - CLI supports adding notes, duplicate protection, searching, topic/cadence tracking, and extracting key lines. - Organizes daily notes and long-term memory with tools for digesting and elevating important information. - Includes configuration support (e.g., extraction triggers), plus sample and recommended workflows. - No external database required; operates directly on the filesystem.

元数据

Slug agent-memory-journal

版本 0.2.3

许可证 MIT-0

累计安装 0

当前安装数 0

历史版本数 3

常见问题