Agent 生成器
- 作者仓库星标 435
- 作者更新于 实时读取
- 作者仓库 nvim
- 领域
- AI 智能
- 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
- @benbrastmckie · 未声明 license
- Token 消耗评级
- 低消耗
- 接入复杂程度
- 需简单配置
- 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
- Shell 执行
- 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: skill-filetypes
description: File format conversion routing with specialized sub-agent dispatch Thin wrapper that routes file…
category: AI 智能
runtime: 无特殊运行时
---
# skill-filetypes 输出预览
## PART A: 任务判断
- 适用问题:提示词、Agent 工作流、模型评估或自动化推理。
- 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“Context Pointers / Trigger Conditions / Direct Invocation”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提示词、Agent 工作流、模型评估或自动化推理,并标出输入、限制和预期结果。
- **02** 执行计划:优先按“Context Pointers / Trigger Conditions / Direct Invocation”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、执行终端命令、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、执行终端命令;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。 原文出现了 `/convert` 这类斜杠命令;如果你的 Agent 支持命令触发,优先用命令开场,再补充目标和边界。
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、执行终端命令。
先用一个小任务确认它会围绕“Context Pointers / Trigger Conditions / Direct Invocation”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: skill-filetypes
description: File format conversion routing with specialized sub-agent dispatch Thin wrapper that routes file…
category: AI 智能
source: benbrastmckie/nvim
---
# skill-filetypes
## 什么时候使用
- 把 AI / Agent方向的常用动作沉淀成 Agent 可调用的技能 适合处理AI Agent、提示词、模型评估与自动化推理,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…
- 面向提示词、Agent 工作流、模型评估或自动化推理,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「Context Pointers / Trigger Conditions / Direct Invocation」组织步骤,不把推断写成作者事实。
- 读取文件、写入/修改文件、执行终端命令;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 "skill-filetypes" {
输入层 -> 用户目标 + 目标文件 + 禁止范围 + 验收标准
上下文层 -> Context Pointers / Trigger Conditions / Direct Invocation
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、执行终端命令 | 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
} Filetypes Skill
Thin wrapper that routes file format operations to the filetypes-router-agent, which then delegates to specialized sub-agents based on file type.
Context Pointers
Reference (do not load eagerly):
- Path:
.opencode/context/core/formats/subagent-return.md - Purpose: Return validation
- Load at: Subagent execution only
Note: This skill is a thin wrapper. Context is loaded by the delegated agents, not this skill.
Trigger Conditions
This skill activates when:
Direct Invocation
- User explicitly runs
/convertcommand - User requests file format conversion in conversation
Implicit Invocation (during task implementation)
When an implementing agent encounters any of these patterns:
Plan step language patterns:
- "Extract text from [file].pdf"
- "Extract content from [file]"
- "Convert [file] to markdown"
- "Convert [file] to PDF"
- "Generate PDF from [documentation/file]"
- "Read content from [file].docx"
- "Create PDF version of [file]"
- "Parse [file].pdf for content"
- "Import [file] content"
File extension detection:
- Source file has extension:
.pdf,.docx,.xlsx,.pptx,.html - Target mentions: "markdown", ".md", "PDF", ".pdf"
Task description keywords:
- "document conversion"
- "format transformation"
- "extract from PDF"
- "generate PDF"
When NOT to trigger
Do not invoke for:
- Reading source code files (.py, .js, .lean, etc.)
- Viewing images without extraction
- Operations that don't involve format conversion
- Files already in the target format
- Spreadsheet-specific operations (use skill-spreadsheet)
- Presentation-specific operations (use skill-presentation)
Execution
1. Input Validation
Validate required inputs:
source_path- Must be provided and file must existoutput_path- Optional, defaults to source dir with appropriate extension
# Validate source exists
if [ ! -f "$source_path" ]; then
return error "Source file not found: $source_path"
fi
# Determine output path if not provided
if [ -z "$output_path" ]; then
source_dir=$(dirname "$source_path")
source_base=$(basename "$source_path" | sed 's/\.[^.]*$//')
source_ext="${source_path##*.}"
# Infer target extension
case "$source_ext" in
pdf|docx|xlsx|pptx|html) output_path="${source_dir}/${source_base}.md" ;;
md) output_path="${source_dir}/${source_base}.pdf" ;;
*) return error "Cannot infer output format for .$source_ext" ;;
esac
fi
2. Context Preparation
Prepare delegation context:
{
"source_path": "/absolute/path/to/source.pdf",
"output_path": "/absolute/path/to/output.md",
"metadata": {
"session_id": "sess_{timestamp}_{random}",
"delegation_depth": 1,
"delegation_path": ["orchestrator", "convert", "skill-filetypes"]
}
}
3. Invoke Router Agent
CRITICAL: You MUST use the Task tool to spawn the router agent.
Required Tool Invocation:
Tool: Task (NOT Skill)
Parameters:
- subagent_type: "filetypes-router-agent"
- prompt: [Include source_path, output_path, metadata]
- description: "Convert {source_path} to {output_path}"
DO NOT use Skill(filetypes-router-agent) - this will FAIL.
Agents live in .opencode/agents/ or extension agent directories, not .opencode/skills/.
The Skill tool can only invoke skills from .opencode/skills/.
The router will:
- Detect source and target formats
- Select appropriate sub-agent (document, spreadsheet, presentation)
- Delegate to sub-agent for actual conversion
- Return standardized JSON result
4. Return Validation
Validate return matches subagent-return.md schema:
- Status is one of: converted, extracted, partial, failed
- Summary is non-empty and <100 tokens
- Artifacts array present with output file path
- Metadata contains session_id, agent_type, delegation info
5. Return Propagation
Return validated result to caller without modification.
Return Format
See .opencode/context/core/formats/subagent-return.md for full specification.
Expected successful return:
{
"status": "converted",
"summary": "Successfully converted document.pdf to document.md using markitdown",
"artifacts": [
{
"type": "implementation",
"path": "/absolute/path/to/document.md",
"summary": "Converted markdown document"
}
],
"metadata": {
"session_id": "sess_...",
"agent_type": "document-agent",
"delegation_depth": 2,
"delegation_path": ["orchestrator", "convert", "skill-filetypes", "filetypes-router-agent", "document-agent"],
"tool_used": "markitdown"
},
"next_steps": "Review converted document"
}
Error Handling
Input Validation Errors
Return immediately with failed status if source file not found.
Unsupported Format
Return failed status with clear message about supported formats.
Router/Subagent Errors
Pass through the router/subagent's error return verbatim.
Tool Not Available
Return failed status with installation instructions.
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