anthropics / anthropics/knowledge-work-plugins

Would a general agent-quality plugin (verification, memory, cost) fit alongside the role plugins?

未关闭
#334 0 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看
主要语言
Python
星标
24k
派生
2.9k
平均合并
1 天 16 小时
30 天内合并 PR
149

描述

These plugins give a Cowork agent domain skills for a role (sales, legal, and so on). There is a cross-cutting layer underneath that every role benefits from, regardless of domain, and I wanted to ask whether it fits here or belongs in the skills repo / community marketplace.

The layer is general agent quality: not asserting past what it actually knows, keeping its working memory small and unambiguous so it does not drift across a long task, checking its own work before declaring it done, and not spending a large model on a trivial step. I have packaged five small, self-contained skills along these lines (critical-qa for self-review, classify for cheapest-viable model routing, ship-web for verifying web artifacts before shipping, anti-hallucinate for claim verification, compact-memory for compact agent state) as one plugin, each a single SKILL.md, Apache-2.0.

A sales or legal agent built on these plugins inherits the same failure modes any agent has, so a general-quality plugin sitting alongside the role-specific ones could be useful. Would that fit this repo, or would you rather cross-cutting plugins go to the community marketplace? Happy to route it wherever you prefer, and to follow whatever contribution path you use here.

贡献指南

这个仓库没有索引到贡献指南

评估

这个 Issue 还没有评估数据。

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。