Agent助手
- 作者仓库星标 2,235
- 作者更新于 实时读取
- 作者仓库 claude-code-plugins-plus-skills
- 领域
- 效率工具
- 兼容 Agent
-
- Claude Code
- Cursor
- Cline
- Codex
- Windsurf
- Gemini CLI
- +20
- 信任分
- 88 / 100 · 社区维护
- 作者 / 版本 / 许可
- @jeremylongshore · 未声明 license
- Token 消耗评级
- 低消耗
- 接入复杂程度
- 即装即用
- 是否需要外部 API Key
- 不需要
- 兼容的系统
- 未声明(默认跨平台)
- 底层运行要求
- 无特殊要求
- 文件与系统权限
-
- 只读
- 允许写入 / 修改
- 网络行为
- 仅限本地
- 安装命令数
- 26 条
档案由构建时根据 SKILL.md 与安装命令自动衍生,可能与作者实际意图存在差异。
需要注意: 未限定 allowed-tools,默认拥有全部工具权限。
---
name: agent-context-loader
description: 'Execute proactive auto-loading: automatically detects and loads agents.md Automatic discovery a…
category: 效率工具
runtime: 无特殊运行时
---
# agent-context-loader 输出预览
## PART A: 任务判断
- 适用问题:日常重复事务、资料整理、邮件或工作流提效。
- 输入要求:目标材料、限制条件、期望输出和验收方式。
- 证据边界:围绕“Overview / Prerequisites / Instructions”读取原文规则,不把推断写成作者承诺。
## PART B: 执行结果
- **01** 任务判断:确认你的需求是否属于日常重复事务、资料整理、邮件或工作流提效,并标出输入、限制和预期结果。
- **02** 执行计划:优先按“Overview / Prerequisites / Instructions”拆成步骤,说明每一步会读取什么、修改什么、产出什么。
- **03** 交付结果:给出可复制的命令、文件改动、检查清单或内容草稿,并说明如何继续迭代。
- **04** 风险边界:结合 读取文件、写入/修改文件、主要在本地完成、通常不需要额外 API Key 给出执行前确认项。
## Running Rules
- 读取文件、写入/修改文件;主要在本地完成;通常不需要额外 API Key。
- 先小样例验证,再放大到真实任务。
- 交付时同时给结果、检查口径和下一步迭代建议。 原文没有稳定的斜杠命令要求。安装验证后通常全局生效,直接在对话里点名这个 Skill 并描述任务即可。
告诉 Agent 目标文件或材料、期望结果、不可改范围、是否允许联网或执行命令。本 Skill 的权限画像是:读取文件、写入/修改文件。
先用一个小任务确认它会围绕“Overview / Prerequisites / Instructions”工作;涉及文件或命令时,先看 diff、日志、预览或测试结果。
检查最终产物是否包含明确结果、必要证据和下一步动作;如果输出泛泛而谈,就补充输入、边界和验收标准后重跑。
---
name: agent-context-loader
description: 'Execute proactive auto-loading: automatically detects and loads agents.md Automatic discovery a…
category: 效率工具
source: jeremylongshore/claude-code-plugins-plus-skills
---
# agent-context-loader
## 什么时候使用
- 把日常协作方向的常用动作沉淀成 Agent 可调用的技能 适合处理重复事务、信息整理、邮件和日常工作流提效,核心价值是把输入、判断、执行、验证和交付边界固定下来,避免 Agent 泛泛回答。 把任务拆成可执行、可检查、可继续迭代的步骤…
- 面向日常重复事务、资料整理、邮件或工作流提效,优先处理能明确输入、步骤和验收标准的工作。
## 需要提供什么
- 目标材料、目录范围、期望结果和不可改动内容。
- 是否允许联网、执行命令、读写文件或调用外部服务。
## 执行规则
- 围绕「Overview / Prerequisites / Instructions」组织步骤,不把推断写成作者事实。
- 读取文件、写入/修改文件;主要在本地完成;通常不需要额外 API Key。
- 先跑小样例,确认结果可检查后再扩大任务范围。
## 输出要求
- 给出最终产物、关键证据、验证方式和下一步动作。
- 信息不足时标记 unknown,不编造命令、平台或依赖。 作者原文负责流程事实;仓库文件负责来源和命令;流狐只补充适用场景、限制和质量判断。
skill "agent-context-loader" {
输入层 -> 用户目标 + 目标文件 + 禁止范围 + 验收标准
上下文层 -> Overview / Prerequisites / Instructions
规则层 -> SKILL.md 触发条件 / 执行顺序 / 输出格式
运行层 -> 无特殊运行时 | 读取文件、写入/修改文件 | 主要在本地完成
安全层 -> 通常不需要额外 API Key + 小任务验证 + diff / 日志复核
输出层 -> 可复制结果 + 检查清单 + 下一步迭代
} Agent Context Loader
Overview
Automatic discovery and loading of AGENTS.md files across project hierarchies for AI coding agents. This skill scans the current workspace and its parent directories to locate agent instruction files, then surfaces their contents so the active agent session has full operational context.
Prerequisites
- A project workspace containing one or more
AGENTS.mdfiles at any directory depth - Read and Glob permissions to traverse the directory tree
- Grep access for searching file contents when multiple candidates exist
Instructions
- Scan the current working directory and all ancestor directories (up to the filesystem root or repository root) for files named
AGENTS.mdoragents.md. - Search subdirectories of the current workspace for additional
AGENTS.mdfiles that may apply to sub-projects or modules. - Determine load order: files closer to the repository root load first (global context), files in deeper directories load later (override or supplement).
- Read each discovered
AGENTS.mdfile and extract its instruction blocks, workflow definitions, and constraint declarations. - Merge instructions into a unified context, noting any conflicts between levels (e.g., a subdirectory agent overriding a root-level rule).
- Present the loaded context as a structured summary: source file path, instruction count, and any detected conflicts.
- Cache the discovery results for the current session to avoid redundant filesystem scans on subsequent activations.
Output
- Ordered list of discovered
AGENTS.mdfile paths with their directory depth - Merged instruction set combining all discovered agent contexts
- Conflict report highlighting any contradictory directives between levels
- Session cache status indicating whether results are fresh or reused
Error Handling
| Error | Cause | Solution |
|---|---|---|
No AGENTS.md found |
Workspace has no agent instruction files | Confirm the correct working directory; create an AGENTS.md at the project root if needed |
| Permission denied on directory traversal | Restricted parent directories above the workspace | Limit scan depth to the repository root (detected via .git/ presence) |
| Circular symlink detected | Symlinked directories create an infinite traversal loop | Skip symlinked directories during the scan; log a warning with the symlink path |
| Conflicting instructions across levels | A subdirectory AGENTS.md contradicts the root-level file |
Flag the conflict in the output; apply the more specific (deeper) instruction by default |
| File encoding error | AGENTS.md uses a non-UTF-8 encoding |
Attempt latin-1 fallback; report the file path and encoding issue |
Examples
Example 1: Monorepo with per-package agent instructions
- Structure: Root
AGENTS.mdsets global conventions;packages/api/AGENTS.mdadds API-specific tooling rules. - Result: Both files loaded in order. The API-specific rules supplement the global context without conflict.
Example 2: Nested workspace with conflicting commit policies
- Structure: Root
AGENTS.mdsays "always sign commits";services/legacy/AGENTS.mdsays "skip commit signing." - Result: Conflict detected and reported. The deeper (legacy) directive is applied with a warning noting the override.
Example 3: First-time setup with no agent files
- Structure: A freshly cloned repository with no
AGENTS.mdanywhere. - Result: Scan completes with zero files found. Output includes a suggestion to create
AGENTS.mdat the repository root with a minimal template.
Resources
- Claude Code AGENTS.md specification: https://docs.anthropic.com/en/docs/agents
- Monorepo workspace patterns for agent configuration: https://docs.anthropic.com/en/docs/claude-code
- File discovery best practices for hierarchical configuration loading
先判断是否适合
作者设计意图
作者的方法与取舍
边界和复核