agent-skills

agent-skills copied to clipboard

Fix: Remove deprecated Option 3 (Notepads) from cursor setup documentation

1

Fixes #44 ## Changes - Remove deprecated Option 3 (Notepads) section from cursor setup documentation - Update Phase-Specific Skills section to use on-demand rule files instead of notepads - Update...

sbusanelli

feat: introduce evaluation framework for skill testing

4

## Summary Introduce an evaluation framework for testing agent skills effectiveness. The repo had no way to measure whether skills actually improve agent performance - this addresses that gap. Follows...

Ic3b3rg

Docs inconsistency: conflicting skill description and anatomy guidance

1

I found a documentation inconsistency that makes first-time skill contributions ambiguous. `CONTRIBUTING.md` says the `description` should start with "Use when ...", while `docs/skill-anatomy.md` says the description should start with what...

MarkADom

Fixes #35 ## Changes - Add context: fork to all 7 slash commands in .claude/commands/ - Add context: fork to all 20 skill files in skills/*/SKILL.md - Add agent: Plan...

sbusanelli

feat: use content: fork in Claude Code slash commands for isolated skill execution

3

## Background Skills and slash commands currently inject their instructions directly into the active conversation context. Claude Code supports a `context: fork` frontmatter field that runs them in an isolated...

nucliweb

## Summary - **`scripts/validate-skills.js`** — zero-dependency Node.js script that validates every skill on every PR. Checks: SKILL.md exists, frontmatter has `name` + `description`, `name` matches directory, description ≤ 1024 chars,...

googlarz

example : https://github.com/shashankswe2020-ux/whoop-mcp/blob/main/.github/agents/issue-orchestrator.agent.md Here I combined your pre-existing agents to be used as sub-agents Next I think we can create Phase level agents with required skills and agents access as...

shashankswe2020-ux

Add retrieval safety guard to source-driven-development skill

5

**Summary** Proposal to add a retrieval safety guard to the source-driven-development skill. The skill has no guidance for what happens after docs are fetched, fetched pages can contain directives that...

KoolP

## Observed behavior When using Claude Code with the agent-skills plugin to scaffold projects, the agent sometimes runs build verification commands (e.g. `npx tsc --noEmit`, `npx vite build`) multiple times...

federicobartoli

Hi addyosmani, 关注项目有一阵子了！ 在尝试将大语言模型工作流集成到本地测试时，我发现很多官方给的 Free Tier API 最近极易遭遇限制或反爬风控，尤其是做高频请求测试的时候经常掉线。 后来和几个做独立开发的朋友研究了一套比较稳定的策略：**用官方活动通道的长期号，配合反重力模式池，可以极大地规避区域风控。** 目前多并发跑下来，稳定性大幅提升。 因为我们在大批量拿海外的活动名额，如果你这边的团队做持续开发测试也有这种 **高频/长期请求大模型接口** 的痛点需求，我们这边刚好手里有一些闲置的安全高级号渠道可以分摊出来： 🚀 **Gemini Pro 1年高级稳定号 (已配置反重力通过)** - 支持用于自动化并发测试验证。 - 考虑到稳定性，我们将长期获取与分摊地址放在了这里（单测试大概十几块钱）：https://gist.github.com/df31d806647d07307f05971d1a20f59a - 或者直接加入TG频道看最新稳定货源：https://t.me/pixelPku 希望这个分享对正在搭建 AI 工作流的朋友有帮助。如果大家有关于绕过接口风控更好的思路，也欢迎下面交流！打扰了，不需要可随时 Close。 📱 TG单独技术/资源交流:...

zcjcoder