小花自我迭代 (HuaNiu Enhanced)

小花专用自我迭代技能 - 基于 self-improving-agent 增强，集成 OpenClaw 工作流、MEMORY.md、百度千帆、看想做找四部曲。专为国内部署优化。

This skill appears internally consistent and implements short, opt‑in reminders and error-detection hooks for an OpenClaw workspace. Before enabling it: 1) review the hook scripts (hooks/openclaw/handler.{js,ts}, scripts/*.sh) to confirm you accept their behavior; they will run with the same user permissions as the agent. 2) If you enable hooks globally (~/.claude or ~/.openclaw settings), prefer a sandbox workspace first to verify outputs. 3) Note extract-skill.sh writes to a relative ./skills directory but validates paths to avoid absolute or '..' writes — still confirm the target directory. 4) The skill references cross-session APIs in docs; those are powerful — only enable inter-session features if you trust the environment. 5) There are no credentials requested by this skill; services like Baidu/OLLAMA are mentioned as optional integrations and would require separate configuration. If you want higher assurance, run the scripts in a controlled environment (non-production user) before enabling them in your main workspace.

Type: OpenClaw Skill Name: xiaohua-self-improving Version: 1.0.0 The skill is designed for agent self-improvement, including logging, knowledge promotion, and operational fixes. It is classified as 'suspicious' due to powerful self-management instructions in `SKILL.md` that direct the AI agent to modify `openclaw.json` and execute system commands like `Stop-Process` and `openclaw gateway` to restart the gateway. While framed as a legitimate 'fix' for context window issues, these capabilities present a significant attack surface for prompt injection, potentially leading to unauthorized configuration changes or denial of service if an attacker could manipulate the agent's interpretation of these instructions. The `scripts/extract-skill.sh` script, while creating new files, includes robust path validation to prevent path traversal.