Agent Integrations Guide

When you run prompts-gpt sync, agent-specific files are generated for each configured target. This guide explains what gets generated and how each agent uses the files.

Supported Agent Targets

Configuring Targets

Sync all targets

prompts-gpt sync --agent all

Sync specific targets

prompts-gpt sync --agent cursor,codex,claude-code

In config.json

{
  "agentTargets": ["codex", "cursor", "claude-code"]
}

Generated Files by Target

Codex (codex)

File: AGENTS.md

Codex reads AGENTS.md at the project root for agent instructions. A managed block is inserted:

<!-- prompts-gpt:start -->
# Prompts-GPT Agent Instructions

Prompts synced by `prompts-gpt sync` live in `.prompts-gpt/`.
Start with [.prompts-gpt/manifest.json](.prompts-gpt/manifest.json).

## Available Prompt Packs
- [Security Review](.prompts-gpt/security-review.md)
- [Code Quality](.prompts-gpt/code-quality.md)
<!-- prompts-gpt:end -->

Any content you add outside the managed block is preserved on re-sync.

Claude Code (claude-code)

File: CLAUDE.md

Same managed block pattern as AGENTS.md:

<!-- prompts-gpt:start -->
# Prompts-GPT Agent Instructions
...
<!-- prompts-gpt:end -->

Cursor (cursor)

Files:

• .cursor/rules/prompts-gpt-*.mdc — One rule file per prompt pack
• .cursor/commands/prompts-gpt-*.md — One command file per prompt pack

Rule files use Cursor's .mdc format with frontmatter:

---
description: Security Review - Review code for security vulnerabilities
globs: "**/*.{ts,tsx,js,jsx}"
---

Review the codebase for security vulnerabilities...

VS Code (vscode)

Files:

• .github/copilot-instructions.md — Copilot chat instructions
• .github/instructions/prompts-gpt.instructions.md — Detailed instructions
• .vscode/prompts-gpt.code-snippets — Code snippets

GitHub Copilot (copilot)

Files:

• .github/prompts/prompts-gpt-*.prompt.md — One prompt per pack

Continue (continue)

Files:

• .continue/rules/prompts-gpt-*.md — Rule files

Gemini CLI (gemini-cli)

File: GEMINI.md

Managed block pattern:

<!-- prompts-gpt:start -->
# Prompts-GPT Agent Instructions
...
<!-- prompts-gpt:end -->

Windsurf (windsurf)

Files:

• .windsurf/rules/prompts-gpt-*.md — Rule files

Cline (cline)

Files:

• .clinerules/prompts-gpt-*.md — Rule files

Junie (junie)

File: .junie/guidelines.md

Managed block in guidelines file.

Amp (amp)

File: AGENT.md

Managed block pattern.

Managed Block Mechanism

Files using managed blocks are updated idempotently. The sync engine:

1. Reads the existing file
2. Finds <!-- prompts-gpt:start --> and <!-- prompts-gpt:end --> markers
3. Replaces only the content between markers
4. Preserves all content outside the markers

If the file doesn't exist, it creates it with the managed block.

If the file exists without markers, the managed block is appended.

Selective Sync

Skip certain targets

# Only sync Cursor and Codex
prompts-gpt sync --agent cursor,codex

Re-sync after prompt changes

prompts-gpt sync --overwrite

Sync without pulling from API

prompts-gpt sync --skip-pull

.gitignore Recommendations

Most agent files should be committed to version control:

# DO commit these (agent instructions)
# AGENTS.md
# CLAUDE.md
# .cursor/rules/
# .cursor/commands/

# DO NOT commit credentials
.prompts-gpt/.credentials.json

Verifying Sync

prompts-gpt status

Shows which agent targets are configured and which files exist.

See Also

• Sync CLI
• Sync API
• Configuration Reference