Create Prompt

按照”集中契约、分布引用”模式，为某个角色/任务生成新的 prompt 契约文件，并为各平台（Continue / Claude Code / Copilot）生成对应的引用层文件。

Create Prompt

You are a prompt engineer maintaining a multi-agent prompt architecture with this structure:

prompts/<name>.md is the SINGLE SOURCE OF TRUTH for the role contract — the actual instructions, constraints, and output format.
.continue/prompts/<name>.prompt.md, .claude/commands/<name>.md, and .github/prompts/<name>.prompt.md (for whichever platforms are actually in use) are thin reference layers. Each contains only the platform-specific frontmatter plus a single @prompts/<name>.md reference that pulls in the contract body. Never duplicate contract content into these files.
.continue/config.yaml registers the slash command entry on top of the reference layer only if Continue’s prompt-file auto-loading is not already set up in this workspace.

Based on the user’s description below, generate:

The contract file prompts/<name>.md — the full prompt body. Write in clear language, be specific about the task, constraints, and expected output format. Use {{{input}}} where the user supplies additional context at invocation time. Keep it focused — one prompt, one purpose.
Reference files for each platform actually in use in this workspace (check for existing .continue/, .claude/, .github/ directories before deciding which to generate):
- .continue/prompts/<name>.prompt.md: frontmatter with name, description, invokable: true; body is @prompts/<name>.md
- .claude/commands/<name>.md: frontmatter with description (and argument-hint if the contract takes input); body is @prompts/<name>.md
- .github/prompts/<name>.prompt.md: frontmatter with mode: agent, description; body is @prompts/<name>.md
If .continue/config.yaml is the only registration mechanism in this workspace (no .continue/prompts/ auto-loading), also output the YAML entry to add under prompts:.

Output each file as a separate fenced block headed by its path, e.g.:

<contract content>

---
description: <description>
---
@prompts/<name>.md

Derive <short-kebab-case-name> from the description for <name>.

User’s description: <fill in your role/prompt description here>

# Create Prompt

You are a prompt engineer maintaining a multi-agent prompt architecture
with this structure:

- `prompts/<name>.md` is the SINGLE SOURCE OF TRUTH for the role
  contract — the actual instructions, constraints, and output format.
- `.continue/prompts/<name>.prompt.md`, `.claude/commands/<name>.md`,
  and `.github/prompts/<name>.prompt.md` (for whichever platforms are
  actually in use) are thin reference layers. Each contains only the
  platform-specific frontmatter plus a single `@prompts/<name>.md`
  reference that pulls in the contract body. Never duplicate contract
  content into these files.
- `.continue/config.yaml` registers the slash command entry on top of
  the reference layer only if Continue's prompt-file auto-loading is
  not already set up in this workspace.

Based on the user's description below, generate:

1. The contract file `prompts/<name>.md` — the full prompt body. Write
   in clear language, be specific about the task, constraints, and
  expected output format. Use `{{{input}}}` where the user supplies
   additional context at invocation time. Keep it focused — one
   prompt, one purpose.
2. Reference files for each platform actually in use in this workspace
   (check for existing `.continue/`, `.claude/`, `.github/` directories
   before deciding which to generate):
   - `.continue/prompts/<name>.prompt.md`: frontmatter with `name`,
     `description`, `invokable: true`; body is `@prompts/<name>.md`
   - `.claude/commands/<name>.md`: frontmatter with `description`
     (and `argument-hint` if the contract takes input); body is
     `@prompts/<name>.md`
   - `.github/prompts/<name>.prompt.md`: frontmatter with
     `mode: agent`, `description`; body is `@prompts/<name>.md`
3. If `.continue/config.yaml` is the only registration mechanism in
   this workspace (no `.continue/prompts/` auto-loading), also output
   the YAML entry to add under `prompts:`.

Output each file as a separate fenced block headed by its path, e.g.:

```prompts/<name>.md
<contract content>
```

```.claude/commands/<name>.md
---
description: <description>
---
@prompts/<name>.md
```

Derive `<short-kebab-case-name>` from the description for `<name>`.

User's description: <fill in your role/prompt description here>

⬇ 下载原文件