← 返回 Skills 市场
Midos Self Improver
作者
msruruguay
· GitHub ↗
· v1.0.0
338
总下载
0
收藏
2
当前安装
1
版本数
在 OpenClaw 中安装
/install midos-self-improver
功能描述
Structured learning pipeline with quality-gated promotion. Captures corrections, errors, and patterns — promotes only what proves itself through recurrence.
安全使用建议
This skill appears to implement a reasonable capture → gate → promote pipeline, but there are red flags you should address before installing: 1) Missing referenced code — skill.json names hook/tool modules that are not included in the package. Ask the author for the missing files or an explanation of how those hooks are provided in your runtime. 2) Data sensitivity — the skill explicitly logs commands, error traces, and context to files in the repository; these can contain secrets or PII. If you install, restrict write locations, set tight filesystem permissions, and review/clean captured entries before promoting them to permanent memory. 3) provenance — source and homepage are unknown; prefer skills with a clear source or run this one in a sandbox environment first. 4) Test locally: run the included tests and a dry-run in an isolated repo, inspect any created files, and confirm promotion rules won't overwrite important files like CLAUDE.md/AGENTS.md. If you cannot obtain the missing hook modules or a trusted source, treat the package as incomplete and do not enable it in production agents.
功能分析
Type: OpenClaw Skill
Name: midos-self-improver
Version: 1.0.0
The midos-self-improver skill is a structured framework designed to help AI agents manage their own 'memory' by logging errors, user corrections, and successful patterns to a local directory structure (.learnings/, .patterns/). It includes a deterministic quality-gate and scoring system to promote high-value learnings into project-level rule files like CLAUDE.md. The provided files (SKILL.md, skill.json, and a Python test suite) contain no evidence of data exfiltration, malicious execution, or harmful prompt injection; instead, they focus on project-specific workflow improvements and include security-conscious tests to prevent secret leakage.
能力评估
Purpose & Capability
The name/description (agent self-improvement) aligns with the SKILL.md content, which documents capture → quality gate → staging → promotion. However skill.json lists source_tools (hooks/pattern_harvester.py, hooks/memory_protocol.py, tools/normalize_naming.py) that are not present in the file manifest. SKILL.md also contains code snippets that import hooks (hooks.learning_capture) and assumes platform capture hooks exist — those runtime components are missing from the published files, an incoherence that could break expected behavior or indicate incomplete packaging.
Instruction Scope
Runtime instructions tell the agent to log corrections, errors, patterns, and tool outputs to local paths (.learnings/, .patterns/, .knowledge/), including command lines, exception traces, and root causes. That scope is plausible for this purpose but will capture potentially sensitive data (commands, file paths, error traces, config contents). The SKILL.md also prescribes deterministic deduplication and promotion rules and shows code snippets wired to capture hooks — but the referenced hook modules are not included.
Install Mechanism
No install spec — instruction-only skill — so nothing is written to disk by an installer. This minimizes supply-chain risk. The risk surface is limited to what the agent executes per SKILL.md (file writes and hook wiring).
Credentials
The skill requests no environment variables, binaries, or credentials (proportionate). However, because it explicitly logs errors/commands/contexts, it may end up persisting secrets or sensitive config if run in an environment that surfaces them. The skill does not declare or limit what it will capture beyond examples, so accidental capture/exfiltration of secrets is possible unless the user enforces storage policies.
Persistence & Privilege
always:false and no special privilege flags. The skill writes into project-local paths (.learnings, .patterns, .knowledge) which is consistent with its purpose. It does not request to modify other skills or system-wide configs in the provided materials.
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 midos-self-improver - 安装完成后，直接呼叫该 Skill 的名称或使用
/midos-self-improver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
v1.0.0
Initial release of midos-self-improver—a self-improving agent with a structured, quality-gated learning pipeline.
- Captures corrections, errors, knowledge gaps, best practices, and recurring patterns via five detection triggers.
- Introduces deduplication and quality gates to filter out low-value or duplicate learnings before storing.
- Implements a four-axis scoring system (recurrence, freshness, specificity, impact) to determine which learnings are promoted, pruned, or kept in staging.
- Automates rule promotion into permanent project memory once patterns prove recurring value.
- Provides ready-to-use capture hooks and guides for both standalone and integrated usage.
元数据
常见问题
Midos Self Improver 是什么？
Structured learning pipeline with quality-gated promotion. Captures corrections, errors, and patterns — promotes only what proves itself through recurrence. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 338 次。
如何安装 Midos Self Improver？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install midos-self-improver」即可一键安装，无需额外配置。
Midos Self Improver 是免费的吗？
是的，Midos Self Improver 完全免费（开源免费），可自由下载、安装和使用。
Midos Self Improver 支持哪些平台？
Midos Self Improver 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 Midos Self Improver？
由 msruruguay（@msruruguay）开发并维护，当前版本 v1.0.0。
推荐 Skills