← 返回 Skills 市场
Reflect
作者
Steven Gonsalvez
· GitHub ↗
· v2.1.0
8587
总下载
14
收藏
0
当前安装
2
版本数
在 OpenClaw 中安装
/install reflect-learn
功能描述
Self-improvement through conversation analysis. Extracts learnings from corrections and success patterns, proposes updates to agent files or creates new skil...
安全使用建议
Install only if you want an agent to keep durable learning records and propose persistent behavior changes. Review every proposed diff before approving, avoid enabling auto-reflection unless you are comfortable with files being created during compaction, and periodically inspect or delete ~/.claude/reflections, ~/.reflect, and any generated .claude/skills entries.
功能分析
Type: OpenClaw Skill
Name: reflect-learn
Version: 2.1.0
The skill is designed for agent self-improvement, allowing it to analyze conversations, propose updates to its own configuration files, and create new skills. While this involves powerful capabilities like modifying agent files, creating new skill definitions, and using the `Bash` tool, the `SKILL.md` explicitly outlines strong safety guardrails, including mandatory human approval for all proposed changes, Git versioning for easy rollbacks, and incremental updates. The Python scripts handle file operations within expected `.claude/` and `~/.reflect/` directories, and there is no evidence of data exfiltration, unauthorized persistence, or malicious execution beyond the stated purpose. The instructions themselves do not contain prompt injection attempts to bypass security or hide actions.
能力评估
Purpose & Capability
The stated purpose matches the capabilities: it analyzes conversations, proposes durable agent updates, creates new skills, and uses Read/Write/Edit/Grep/Glob/Bash. The concern is that this is high-impact self-modification and cross-session memory behavior, not that it is hidden or unrelated.
Instruction Scope
SKILL.md requires explicit approval before applying agent changes and shows diffs, but it also frames session end, context compaction, and broad success patterns as use cases. Detection patterns include common approval phrases, which can overinterpret ordinary conversation unless the user carefully reviews proposals.
Install Mechanism
The package includes hook installation examples and executable Python helpers, but no evidence of network exfiltration, obfuscation, package-install abuse, or automatic installation during review. Bundled development log JSON exposes session paths and prompt history, which is poor packaging hygiene but not a direct user-system compromise.
Credentials
The global reflection store under ~/.claude/reflections, by-project copies, by-agent learnings, exact source quotes, project paths, and proposed diffs are broader than a project-local reflection workflow and can retain sensitive conversation or project context across unrelated work.
Persistence & Privilege
The skill can update agent behavior, create .claude/skills/{name}/SKILL.md, commit changes, maintain state under ~/.reflect or ~/.claude/session, and install a PreCompact hook. These are purpose-aligned but require careful user control because they persist and can affect future agent behavior.
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 reflect-learn - 安装完成后，直接呼叫该 Skill 的名称或使用
/reflect-learn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
v2.1.0
Add Security category to signal detection, fix agent_mappings.md root-level agents, add Memory integration guidance, consistency fixes across reference files
v2.0.0
- Updated skill.
元数据
常见问题
Reflect 是什么？
Self-improvement through conversation analysis. Extracts learnings from corrections and success patterns, proposes updates to agent files or creates new skil... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 8587 次。
如何安装 Reflect？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install reflect-learn」即可一键安装，无需额外配置。
Reflect 是免费的吗？
是的，Reflect 完全免费（开源免费），可自由下载、安装和使用。
Reflect 支持哪些平台？
Reflect 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 Reflect？
由 Steven Gonsalvez（@stevengonsalvez）开发并维护，当前版本 v2.1.0。
推荐 Skills