← 返回 Skills 市场
Memory Bench Designer
作者
TatsuKo Tsukimi
· GitHub ↗
· v0.1.0
· MIT-0
88
总下载
0
收藏
0
当前安装
1
版本数
在 OpenClaw 中安装
/install memory-bench-designer
功能描述
Designs a custom agent-memory benchmark for the user's specific use case. Activate when the user asks which memory strategy fits their agent, how to evaluate...
安全使用建议
Before installing or enabling this skill, verify the following: (1) Where does the 'memory-bench' runner come from? The skill expects to run 'memory-bench' but the registry gives no install or binary source — ask the publisher for an official install/instruction or a trusted release URL. (2) Templates referenced in SKILL.md (templates/scenario.yaml.tmpl, templates/weights.yaml.tmpl) are not included in the package; request those or confirm how they are created. (3) Running the skill will write files into your current working directory and will run a local CLI that may download ~90 MB of model data from external hosts (Hugging Face or similar). If you are uncomfortable with filesystem writes or network/model downloads, do not enable autonomous runs; prefer manual invocation. (4) Because the skill's source and homepage are unknown, exercise extra caution: ask the owner for provenance, an install guide, and a checksum/verified release of the runner. If those clarifications are provided (runner origin, templates included, or an explicit install spec), the coherence issues would be resolved; otherwise, treat the skill as suspicious and avoid giving it autonomous execution privileges.
功能分析
Type: OpenClaw Skill
Name: memory-bench-designer
Version: 0.1.0
The memory-bench-designer skill is a legitimate tool for evaluating AI memory strategies (e.g., BM25, ACT-R, Embeddings). It follows a structured four-stage process to elicit user requirements, generate YAML configuration files, and execute a local benchmarking CLI tool (`memory-bench`). The instructions in SKILL.md and the supporting documentation in the references/ and examples/ directories are consistent, transparent, and strictly aligned with the stated purpose. There is no evidence of data exfiltration, malicious prompt injection, or unauthorized system access; the mentioned model download (~90MB for sentence-transformers) is standard for the semantic retrieval tasks described.
能力标签
能力评估
Purpose & Capability
SKILL.md clearly expects to run an external runner (memory-bench run ...) and to use template files (templates/scenario.yaml.tmpl, templates/weights.yaml.tmpl). The registry metadata declares no required binaries and includes no templates or install instructions. Either the metadata is incomplete or the skill assumes tools/files that are not provided — that is an incoherence between claimed functionality and declared requirements.
Instruction Scope
Runtime instructions tell the agent to (a) conduct multi-turn elicitation, (b) write scenario-<name>.yaml and weights-<name>.yaml into the user's current working directory, (c) invoke a CLI command that produces results/<name>/results.md and results.json, and (d) read and interpret results.md. The instructions reference template files that are not present in the file manifest. They also note that the runner will download a ~90 MB sentence-transformers model on first run. These behaviors involve filesystem writes, executing a local CLI, and network downloads — all beyond what's declared in the metadata.
Install Mechanism
There is no install spec (instruction-only), which is low-risk in itself. However, the skill assumes the 'memory-bench' CLI exists on PATH and will cause a model download (Hugging Face / sentence-transformers) when invoked with --embedding. The absence of an install step or a declared source for the runner binary means it's unclear how that binary would be obtained or whether it is safe/trusted.
Credentials
The skill declares no required environment variables, credentials, or config paths, and SKILL.md does not ask for secrets. That is proportionate. Note: network activity (model download) and reading/writing local files will still occur; no explicit credential access is requested.
Persistence & Privilege
always:false (good) and no install-time persistence is requested. The default autonomous-invocation setting is enabled (disable-model-invocation:false) — normal for skills — but combined with the instruction to execute a CLI and write files, this increases the impact if the agent runs the skill without clear user approval. The skill does not request to modify other skills or system-wide settings.
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 memory-bench-designer - 安装完成后，直接呼叫该 Skill 的名称或使用
/memory-bench-designer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
v0.1.0
Initial release: 4-stage elicitation skill + companion Python runner benchmarking 5 memory strategies (Recency, BM25, ACT-R, Embedding, Composite) across 8 dimensions in 4 families (Exploration, Ranking, Adaptation, Maintenance). Three worked examples (game-AI, NPC cognition, coding agent) demonstrate different use cases produce different winners. Runner at https://github.com/TatsuKo-Tsukimi/memory-bench-designer
元数据
常见问题
Memory Bench Designer 是什么？
Designs a custom agent-memory benchmark for the user's specific use case. Activate when the user asks which memory strategy fits their agent, how to evaluate... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 88 次。
如何安装 Memory Bench Designer？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install memory-bench-designer」即可一键安装，无需额外配置。
Memory Bench Designer 是免费的吗？
是的，Memory Bench Designer 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
Memory Bench Designer 支持哪些平台？
Memory Bench Designer 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 Memory Bench Designer？
由 TatsuKo Tsukimi（@tatsuko-tsukimi）开发并维护，当前版本 v0.1.0。
推荐 Skills