← 返回 Skills 市场
161
总下载
0
收藏
0
当前安装
3
版本数
在 OpenClaw 中安装
/install reflex-learn
功能描述
Detects repeated queries as implicit negative feedback and non-repetition as positive feedback, enabling continuous learning by writing reflections and patte...
安全使用建议
This skill appears to do what it claims. Before installing: review install.sh and consider running it inside a Python virtual environment to avoid altering system packages; confirm you have ~80 MB free for the model cache; be aware that reflections and patterns will be written to ~/.openclaw/MEMORY.md, ~/.openclaw/reflexlearn-pending.md (default cautious mode) and optionally ~/.openclaw/SOUL.md if you switch to aggressive mode or manually accept pending entries. If you plan to use Ollama features, verify Ollama runs locally (http://localhost:11434) — the skill contacts only localhost for that integration. If you want tighter control, keep MODE=cautious and review reflexlearn-pending.md regularly before promoting changes to SOUL.md.
功能分析
Type: OpenClaw Skill
Name: reflex-learn
Version: 1.0.2
ReflexLearn is a continuous learning skill that uses query similarity (via sentence-transformers) to infer user satisfaction. The code is well-structured and includes explicit security features, such as path validation in `reflex_learn.py` to prevent file operations outside the `~/.openclaw/` directory. Network activity is transparently handled: `install.sh` performs a one-time download of model weights from Hugging Face, and runtime operations are offline by default (with an optional local-only Ollama integration). No evidence of data exfiltration, malicious execution, or harmful prompt injection was found.
能力评估
Purpose & Capability
Name/description (detect repeated queries and write reflections/patterns) align with required binaries (python3, bash), included Python code, and files that read/write ~/.openclaw/*. No unrelated credentials, external services, or binaries are requested.
Instruction Scope
SKILL.md and the script instruct only to embed queries, compare against ~/.openclaw/reflex_history.json, and write to MEMORY.md / SOUL.md / reflexlearn-pending.md under ~/.openclaw. Optional Ollama calls are to localhost. The skill reads its SKILL.md for config. All referenced files and operations are consistent with the documented behavior.
Install Mechanism
install.sh performs pip installs from PyPI and pre-caches a Hugging Face model (~80 MB). These are declared and require explicit user confirmation. Note: the script uses the system 'pip' (no virtualenv), so consider using a virtual environment if you want to avoid modifying global Python packages.
Credentials
No environment variables or external credentials are requested. Network access is confined to the documented install step and an optional local-only Ollama instance; the runtime supports a strict --offline mode and enforces writes only under ~/.openclaw/.
Persistence & Privilege
Skill is not always-enabled, does not request system-wide privileges, and restricts all filesystem writes to ~/.openclaw/. It does not attempt to modify other skills or global agent settings. Autonomous invocation is allowed by default (normal for skills) but not escalated.
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 reflex-learn - 安装完成后，直接呼叫该 Skill 的名称或使用
/reflex-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
v1.0.2
- Added install.sh script for easy, one-step setup (installs dependencies and downloads model weights).
- Enforced strict path validation: all file operations restricted to ~/.openclaw/.
- Introduced model-download guard with explicit warning and new --offline flag for offline operation.
- Removed unused scikit-learn dependency.
- All network access now occurs only in install.sh, with user confirmation and clear listing of operations.
v1.0.1
**v1.1.0 adds robust multi-repeat handling, modifier detection, tiered memory updates, and optional review mode.**
- Requires multiple repeats before flagging an answer as failed, reducing false positives.
- Detects modifier words in repeated queries and extracts user preferences instead of flagging failure.
- Adds cautious (pending file) and aggressive (direct write) modes for updating SOUL.md.
- Stages potential SOUL.md updates in reflexlearn-pending.md for review in cautious mode.
- Introduces new configuration options for repeat threshold, session timing, and operational mode.
- Updates documentation and usage examples to reflect new features and slash commands.
v1.0.0
ReflexLearn v1.0.0
- Introduces implicit feedback-based continuous learning by detecting repeated or non-repeated queries as signals of negative or positive feedback.
- Automatically writes reflections and learned patterns to MEMORY.md and SOUL.md in response to user behavior.
- Operates after agent responses and on regular heartbeat intervals, with triggers for both.
- Configurable learning parameters such as similarity threshold, lookback interactions, and reinforcement delay.
- Embeds queries using local sentence-transformers; optionally supports richer reflection via local Ollama.
- Designed to never call external APIs (except optional local Ollama) and only modifies permitted files in ~/.openclaw/.
元数据
常见问题
Reflexlearn 是什么？
Detects repeated queries as implicit negative feedback and non-repetition as positive feedback, enabling continuous learning by writing reflections and patte... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 161 次。
如何安装 Reflexlearn？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install reflex-learn」即可一键安装，无需额外配置。
Reflexlearn 是免费的吗？
是的，Reflexlearn 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
Reflexlearn 支持哪些平台？
Reflexlearn 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 Reflexlearn？
由 Kinvectum（@kaventures）开发并维护，当前版本 v1.0.2。
推荐 Skills