Writing Pipeline

内容再创作管线：输入参考资料 → AI 生成中文初稿 → 人工改稿 → 系统对比学习写作偏好 → 多平台格式输出。 支持自动路由——根据输入材料特征自动匹配最合适的写作风格。

This skill mostly does what it says (draft → review → publish) but contains two noteworthy risks: (1) it instructs the agent to modify other 'writing-style' skills after learning from edits, and (2) it orchestrates many helper skills (deepreader, x_search, infographic-gen, etc.) without declaring their credentials or access requirements. Before installing: 1) Confirm you trust the owner/source (unknown here). 2) Verify how the platform enforces user confirmation — ensure the agent cannot auto-confirm updates that modify other skills. 3) Back up the skill directories and any writing-style skills that could be changed. 4) Audit the helper skills (deepreader, x_search, etc.) to see whether they require API keys or have network/exfil behavior. 5) If you allow updates to other skills, require explicit, per-change user approval and maintain an audit log. If you cannot verify those controls, treat this skill as higher risk and consider running it in a restricted environment or refusing permission to modify other skills.

Type: OpenClaw Skill Name: writing-pipeline Version: 1.0.0 The writing-pipeline skill bundle orchestrates a complex multi-stage workflow involving content generation, style adaptation, and multi-platform publishing. It exhibits high-risk behaviors, specifically the capability to modify other writing-style skill files based on an AI-driven 'Diff-and-Learn' process (detailed in SKILL.md and references/diff-learn.md), which could be targeted by indirect prompt injection. While the logic appears aligned with its stated purpose, the combination of broad file system access, orchestration of numerous external skills, and self-modification of the agent's own skill library warrants a suspicious classification.