一键导入
file-prompts
Load prompts from external files with path expansion. Use for version-controlled prompts outside config.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Load prompts from external files with path expansion. Use for version-controlled prompts outside config.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Tool-agnostic search — query construction, tool selection, source trust hierarchy.
Auto-continue through todos with idle detection and safety gates. Use for multi-step orchestration.
Level 2 — Pantheon-native context compression with priority scoring, semantic summarization, downstream-aware compression, budget allocation, and cross-references
Automated visual review pipeline — Playwright screenshots, self-analysis, fix loop, escalation. Used by Aphrodite for UI verification.
Multi-agent orchestration with model routing, category delegation, and sprint management. Use for coordinating Pantheon agents.
MCP security hardening — credential leakage prevention, input sanitization, and tool access control. Use for reviewing agent MCP configurations.
| name | file-prompts |
| description | Load prompts from external files with path expansion. Use for version-controlled prompts outside config. |
| context | fork |
| globs | [] |
| alwaysApply | false |
Use this skill to load agent system prompts from external files instead of embedding them in config. Enables version control, sharing, and cleaner configuration.
Instead of embedding prompts in JSON:
// ❌ Embedded prompt (hard to version, hard to share)
{
"agents": {
"hermes": {
"prompt": "You are Hermes, the backend specialist. Your expertise is..."
}
}
}
Load from files:
// ✅ File-based prompt (versioned, shareable)
{
"agents": {
"hermes": {
"prompt": "file://./prompts/hermes-prompt.md"
}
}
}
| Format | Example | Resolves To |
|---|---|---|
| Relative | file://./prompts/hermes.md | <project>/prompts/hermes.md |
| Home expansion | file://~/prompts/hermes.md | ~/.config/opencode/prompts/hermes.md |
| Absolute | file:///opt/prompts/hermes.md | /opt/prompts/hermes.md |
project/
├── opencode.json # Config (prompt references)
├── prompts/
│ ├── hermes-prompt.md # Backend specialist prompt
│ ├── aphrodite-prompt.md # Frontend specialist prompt
│ └── themis-prompt.md # Review specialist prompt
Prompts can be versioned independently from config.
~/.config/opencode/prompts/
├── hermes-prompt.md # Shared across all projects
├── aphrodite-prompt.md
└── themis-prompt.md
{
"agents": {
"hermes": {
"prompt": "file://~/prompts/hermes-prompt.md"
}
}
}
{
"categories": {
"deep": {
"prompt_append": "file://./prompts/deep-category-append.md"
}
}
}
Appends additional context to the category's prompt without duplicating the base prompt.
Prompt files are plain markdown:
# Hermes — Backend Specialist
You are the **BACKEND TASK IMPLEMENTER** (Hermes) called by Zeus to implement
FastAPI endpoints, services, and routers.
## Core Principles
- TDD first: write tests that fail, write minimal code to pass
- Async/await on all I/O operations
- Type hints on all parameters
- Max 300 lines per file
## Your Tools
- FastAPI, SQLAlchemy, Pydantic, Redis
- pytest for testing
- ruff for linting
## When to Delegate
- @apollo for codebase discovery
- @themis for code review
- @demeter for database migrations
| Benefit | Description |
|---|---|
| Version control | Prompts in git, track changes over time |
| Sharing | Same prompt across multiple projects |
| Cleaner config | Config files stay concise |
| Easier editing | Edit prompts in any text editor |
| Category context | Append category-specific instructions without duplication |
This is a configuration feature, not a runtime skill. The file is loaded once at agent startup and injected into the system prompt.