sync-agents

name

sync-agents

description

Synchronize AGENTS.md content into all detected AI coding agent configurations. AGENTS.md is the universal source of truth — read natively by GitHub Copilot (Aug 2025+), OpenAI Codex, Cursor, Gemini CLI, and VS Code. Falls back to .github/copilot-instructions.md for repos that have not migrated yet. Use when asked to mirror AGENTS.md or .github/copilot-instructions.md into Claude, Cursor, Gemini, Windsurf, and related tooling.

Sync Agent Instructions, Agents, and Skills

Mirror all GitHub Copilot customizations — instructions, custom agents, and agent skills — into every detected AI coding agent configuration found in this repository.

This skill supports 12 agent ecosystems. Rather than creating files for every agent unconditionally, it detects which agents are already configured in the repo and syncs only those.

Source locations:

• AGENTS.md — global instructions (primary source of truth)
• .github/copilot-instructions.md — global instructions (fallback for repos not yet using AGENTS.md)
• .github/instructions/*.instructions.md — path-scoped instructions
• .github/agents/*.agent.md — custom agent profiles
• .github/skills/<name>/SKILL.md — agent skills (open standard)

Your Task

Execute the following steps in order, creating or overwriting files as needed. Never delete AGENTS.md or anything inside .github/.

---

Step 1 — Read the source of truth

Read AGENTS.md at the repo root and store its full content as the source of truth.

If AGENTS.md is not found, fall back to .github/copilot-instructions.md. If neither exists, stop and tell the user:

"No source file found. Create AGENTS.md at the repo root with your project instructions, then re-run /sync-agents."

Note which source file was used — it will be referenced in sync headers and reports.

Also scan and store the contents of:

• .github/instructions/ — any *.instructions.md files. For each, record the filename, applyTo frontmatter glob (if present), and body content.
• .github/agents/ — any *.agent.md files. For each, record the full filename, the complete YAML frontmatter block (name, description, model, tools, handoffs, etc.), and the body prompt.
• .github/skills/ — any subdirectory containing a SKILL.md file. For each skill, record the directory name, the complete SKILL.md frontmatter (name, description, user-invokable, disable-model-invocation, etc.), the body content, and note any additional bundled asset files (scripts, examples, references) present in that directory.

---

Step 2 — Detect active agents

Scan the repository root for the presence of the following files and directories. For each one found, mark that agent as active and include it in subsequent sync steps. Do not create directories or files for agents that are not already present.

Agent	Detection Signal(s)	Category
Claude Code	CLAUDE.md at repo root or .claude/ directory	Full sync
Gemini CLI	GEMINI.md at repo root or .gemini/ directory	Full sync
OpenAI Codex	.agents/ directory	Full sync — Note: AGENTS.md is the source of truth for Codex instructions; no instructions sync needed. Detect only if .agents/ directory is explicitly present.
OpenCode	.opencode/ directory or .opencode.json	Full sync
Cursor	.cursor/ directory or .cursorrules file	Full sync
Windsurf	.windsurf/ directory or .windsurfrules file	Full sync
Cline	.clinerules file or .clinerules/ directory	Full sync
Roo Code	.roo/ directory or .roorules file	Full sync
Kilo Code	.kilocode/ directory	Full sync
JetBrains Junie	.junie/ directory	Full sync
Zed	.rules file at repo root	Instructions only
Augment Code	.augment/ directory	Full sync

If no agents are detected, offer the user a choice:

"No agent configuration directories or files were detected in this repository. You have two options:

Option A — Create AGENTS.md (recommended for new repos): AGENTS.md is now read by GitHub Copilot, OpenAI Codex, OpenCode, Cursor, Gemini CLI, and VS Code. Creating this file gives you broad multi-tool coverage from a single file. Confirm to create AGENTS.md now and sync your Copilot instructions into it.

Option B — Initialize a specific agent: Create the config signal for the tool you use (e.g., CLAUDE.md for Claude Code, .cursor/ for Cursor) and re-run /sync-agents."

If the user confirms Option A, create AGENTS.md at the repo root with a sync header and the full content of the source file from Step 1 (.github/copilot-instructions.md), then skip to Step 6 and report completion. Note: once AGENTS.md is created, future runs will use it as the source of truth.

List the detected agents before proceeding:

🔍 Detected agents: <comma-separated list of detected agent names>
   Skipping (not detected): <comma-separated list of undetected agent names>

---

Step 3 — Sync instructions to detected agents

For each agent marked as active in Step 2, sync the global instructions and any path-scoped instructions using the agent-specific format described below.

Every synced file must include a sync header (as a comment appropriate to the format) indicating:

• The file is auto-synced and should not be edited directly
• The source of truth path (AGENTS.md or the specific instructions file)
• How to re-sync (/sync-agents)

---

3A — Claude Code

Global instructions → CLAUDE.md

Create or overwrite CLAUDE.md at the repo root:

<!--
  ╔══════════════════════════════════════════════════════════════╗
  ║  AUTO-SYNCED — DO NOT EDIT THIS FILE DIRECTLY               ║
  ║  Source of truth: AGENTS.md                                 ║
  ║  To update: run /sync-agents                                ║
  ╚══════════════════════════════════════════════════════════════╝
-->

# Claude Code Instructions

> **Source of truth:** [`AGENTS.md`](AGENTS.md)
>
> This file is automatically generated by the `/sync-agents` command.
> To propagate changes, update `AGENTS.md` and re-run that command.

---

[INSERT FULL CONTENT OF AGENTS.md HERE]

Path-scoped instructions → .claude/instructions/

If path-specific .github/instructions/*.instructions.md files exist, create matching files in .claude/instructions/ with a sync header comment. Strip the applyTo frontmatter (Claude Code does not use it) but preserve it as a markdown comment for context.

---

3B — Gemini CLI

Global instructions → GEMINI.md

Gemini CLI supports @path/to/file.md import syntax for dynamic loading at runtime.

<!--
  ╔══════════════════════════════════════════════════════════════╗
  ║  SOURCE OF TRUTH: AGENTS.md                                 ║
  ║  This file uses Gemini CLI's native @import syntax so it    ║
  ║  always loads the latest instructions at runtime.           ║
  ║  Run /sync-agents to update path-specific files.            ║
  ╚══════════════════════════════════════════════════════════════╝
-->

# Gemini CLI Instructions

> **Source of truth:** `AGENTS.md`
> This file dynamically imports from `AGENTS.md` using Gemini's
> `@file` syntax. Any edits to `AGENTS.md` are reflected
> here automatically — no manual sync required.

@AGENTS.md

Path-scoped instructions: Append additional @ imports, one per line. If any have applyTo targeting a subdirectory, create matching GEMINI.md files in those directories using the same @ import syntax pointing back to .github/instructions/.

---

3C — AGENTS.md (Universal)

AGENTS.md is the source of truth — no sync needed.

AGENTS.md is read natively by GitHub Copilot (Aug 2025+), OpenAI Codex, OpenCode, Cursor, Gemini CLI, and VS Code. Because it is the source of truth, no copy is written here. All other sync steps distribute FROM this file.

Path-scoped instructions: For each .github/instructions/*.instructions.md with an applyTo targeting a subdirectory, create a corresponding AGENTS.md inside that subdirectory. Codex reads one AGENTS.md per directory level and merges top-down.

Skip to Step 3D.

---

3D — OpenCode

OpenCode reads AGENTS.md at the repo root directly. Since AGENTS.md is the source of truth, no additional sync step is needed — OpenCode picks it up automatically.

Custom commands: If .github/skills/ contains skills, optionally create matching .opencode/commands/<skill-name>.md files containing the skill's body content as a reusable prompt. Prepend each with a sync header comment.

---

3E — Cursor

Global instructions → .cursor/rules/global-instructions.mdc

Cursor uses .mdc files (Markdown with YAML frontmatter) in .cursor/rules/.

---
# AUTO-SYNCED from AGENTS.md — do not edit directly
# Source of truth: AGENTS.md | Re-sync: /sync-agents
description: "Global project instructions synced from GitHub Copilot"
alwaysApply: true
---

[INSERT FULL CONTENT OF AGENTS.md HERE]

Path-scoped instructions → .cursor/rules/<name>.mdc

For each .github/instructions/<name>.instructions.md, create .cursor/rules/<name>.mdc:

• Set description from the filename or first heading
• Set globs from the applyTo frontmatter value (if present)
• Set alwaysApply: false (the glob handles activation)
• Write the body content after the frontmatter

---

3F — Windsurf

Global instructions → .windsurf/rules/global-instructions.md

Windsurf uses plain Markdown files in .windsurf/rules/.

<!-- AUTO-SYNCED from AGENTS.md — do not edit directly -->
<!-- Source of truth: AGENTS.md | Re-sync: /sync-agents -->

# Global Project Instructions

[INSERT FULL CONTENT OF AGENTS.md HERE]

Path-scoped instructions → .windsurf/rules/<name>.md

For each .github/instructions/<name>.instructions.md, create a matching .md file with sync header and body content.

---

3G — Cline

Global instructions → .clinerules/global-instructions.md

Cline reads Markdown files from the .clinerules/ directory. If only a .clinerules file (not directory) exists, convert it to a directory first by moving the existing file to .clinerules/_original.md, then proceed.

<!-- AUTO-SYNCED from AGENTS.md — do not edit directly -->
<!-- Source of truth: AGENTS.md | Re-sync: /sync-agents -->

# Global Project Instructions

[INSERT FULL CONTENT OF AGENTS.md HERE]

Path-scoped instructions → .clinerules/<name>.md

For each .github/instructions/<name>.instructions.md, create a matching .md file.

---

3H — Roo Code

Global instructions → .roo/rules/global-instructions.md

Roo Code reads .md files recursively from .roo/rules/ in alphabetical order.

<!-- AUTO-SYNCED from AGENTS.md — do not edit directly -->
<!-- Source of truth: AGENTS.md | Re-sync: /sync-agents -->

# Global Project Instructions

[INSERT FULL CONTENT OF AGENTS.md HERE]

Path-scoped instructions → .roo/rules/<name>.md

For each .github/instructions/<name>.instructions.md, create a matching .md file.

---

3I — Kilo Code

Global instructions → .kilocode/rules/global-instructions.md

Kilo Code reads Markdown files from .kilocode/rules/.

<!-- AUTO-SYNCED from AGENTS.md — do not edit directly -->
<!-- Source of truth: AGENTS.md | Re-sync: /sync-agents -->

# Global Project Instructions

[INSERT FULL CONTENT OF AGENTS.md HERE]

Path-scoped instructions → .kilocode/rules/<name>.md

For each .github/instructions/<name>.instructions.md, create a matching .md file.

---

3J — JetBrains Junie

All instructions → .junie/guidelines.md

Junie uses a single guidelines.md file. Concatenate all instructions into one file.

<!-- AUTO-SYNCED from AGENTS.md — do not edit directly -->
<!-- Source of truth: AGENTS.md | Re-sync: /sync-agents -->

# Project Guidelines

[INSERT FULL CONTENT OF AGENTS.md HERE]

---

<!-- Path-scoped instructions follow (if any) -->

[FOR EACH .github/instructions/*.instructions.md, INSERT:]

## [filename context / applyTo info]

[body content]

Since Junie only reads a single file, all path-scoped instructions are appended to the same file with clear section headers.

---

3K — Zed

Global instructions → .rules

Zed reads a .rules file at the project root. It is plain text (no frontmatter).

# AUTO-SYNCED from AGENTS.md — do not edit directly
# Source of truth: AGENTS.md | Re-sync: /sync-agents

[INSERT FULL CONTENT OF AGENTS.md HERE]

Zed also auto-detects .cursorrules, .windsurfrules, AGENTS.md, CLAUDE.md, and GEMINI.md — so if those files were written by earlier steps, Zed gets additional coverage automatically.

Path-scoped instructions are not supported by Zed's .rules format. If path-scoped instructions exist, append a note:

# NOTE: Path-scoped instructions exist in .github/instructions/ but cannot be
# represented in Zed's .rules format. See those files for directory-specific guidance.

---

3L — Augment Code

Global instructions → .augment/rules/global-instructions.md

Augment Code uses Markdown files with optional YAML frontmatter in .augment/rules/.

---
# AUTO-SYNCED from AGENTS.md — do not edit directly
# Source of truth: AGENTS.md | Re-sync: /sync-agents
always_apply: true
---

# Global Project Instructions

[INSERT FULL CONTENT OF AGENTS.md HERE]

Path-scoped instructions → .augment/rules/<name>.md

For each .github/instructions/<name>.instructions.md, create .augment/rules/<name>.md:

• Set always_apply: false and agent_requested: true in the YAML frontmatter
• Include a description derived from the filename or applyTo value
• Write the body content after the frontmatter

---

Step 3M — Model normalization policy (all targets)

Apply a single normalization policy for model fields across all detected targets.

For each detected target, determine whether the synced artifact format supports an explicit agent/profile model field:

• Supports explicit model field in this sync flow: Claude Code (.claude/agents/*.md)
• Does not support agent profile model fields in this sync flow: Gemini CLI, OpenAI Codex, OpenCode, Cursor, Windsurf, Cline, Roo Code, Kilo Code, JetBrains Junie, Zed, Augment Code

When a target supports an explicit model field, normalize source model values to that target's default family map before writing files.

Default target map (current):

Target	Keep-as-is patterns	Rewrite patterns	Default fallback
Claude Code	claude-*, sonnet, haiku, opus	gpt-*, o*, codex-*, gemini-*	sonnet

Normalization rules:

1. If model is target-native, keep unchanged.
2. If model is non-native but recognized, rewrite to target default mapped value.
3. If model is missing, inject target default mapped value.
4. If model is present but unrecognized, rewrite to target default and preserve provenance with a comment.

For targets that do not support agent profile models in this sync flow, do not inject synthetic model fields. Record each one as not applicable in Step 6 reporting.

---

Step 4 — Sync custom agents (where supported)

Custom agent profiles (.github/agents/*.agent.md) can only be synced to agents that support sub-agent profiles. Currently, only Claude Code supports this.

If Claude Code is active:

For each .github/agents/<name>.agent.md:

1. Create .claude/agents/<name>.md (drop the .agent infix)
2. Copy YAML frontmatter and body, and apply Step 3M model normalization policy using the Claude target map.
3. Write the frontmatter, adding these YAML comments after ---:

   # AUTO-SYNCED from .github/agents/<name>.agent.md — do not edit directly
   

Source of truth: .github/agents/ | Re-sync: /sync-agents

4. Write the body prompt verbatim after the closing `---`

**Model normalization rules (Copilot → Claude):**

- If `model` is already Claude-native (starts with `claude-`, or is `sonnet`, `haiku`, or `opus`), keep it unchanged.
- If `model` is Copilot/OpenAI/Gemini specific (examples: `gpt-5`, `gpt-5.3-codex`, `o3`, `o4-mini`, `gpt-4.1`, `gemini-*`, `codex-*`), rewrite it to `sonnet`.
- If `model` is missing, add `model: sonnet` in the synced `.claude/agents/<name>.md` frontmatter.
- If `model` is present but unrecognized, rewrite to `sonnet` and add a YAML comment directly above it:
```yaml
# Normalized by /sync-agents from unsupported source model: <original-model>
model: sonnet

Default map reference:

Source model family	Synced Claude model
claude-*, sonnet, haiku, opus	Keep as-is
gpt-*, o*, codex-*, gemini-*	sonnet
Missing/unknown	sonnet

All other agents: No sub-agent profile concept exists. Skip agent syncing.

---

Step 5 — Sync agent skills (where supported)

Agent Skills use an open standard (SKILL.md with YAML frontmatter). Only tools with a documented skill/command path receive synced copies.

Tool	Skills path	Supported?
GitHub Copilot (source)	.github/skills/<name>/SKILL.md	✅ Source
Claude Code	.claude/skills/<name>/SKILL.md	✅ If detected
OpenCode	.opencode/commands/<name>.md	✅ Via Step 3D
OpenAI Codex	—	❌ No skills concept
All others	—	❌ No skills concept

For each .github/skills/<skill-name>/ directory:

If Claude Code is active:

1. Create .claude/skills/<skill-name>/ directory
2. Write SKILL.md verbatim with a YAML sync comment
3. Copy any bundled asset files as-is

OpenCode custom commands are handled in Step 3D. No additional action needed here.

Important — name field constraint: The name field in SKILL.md frontmatter must exactly match the parent directory name. Do not rename directories or alter the field.

Gemini CLI (if active): Append a note to GEMINI.md:

<!-- NOTE: Agent Skills are not supported by Gemini CLI. -->
<!-- Skills available in this repo: <comma-separated skill names> -->
<!-- See .github/skills/ for skill definitions usable by Copilot, Claude, and Codex. -->

All others: No skills concept. Skip.

---

Step 6 — Verify and report

After writing all files, list every file created or updated, grouped by agent:

✅ Sync complete.

🔍 Detected agents: <list>
📁 Source of truth: AGENTS.md (agents and skills from .github/)

Files written:

  [Agent Name]
    - <file path>                              (<description>)
    ...

  [Agent Name]
    - <file path>                              (<description>)
    ...

  SKIPPED (not detected in repo):
    - <agent name>: <detection signal not found>
    ...

To keep all agents in sync, run /sync-agents after any update to AGENTS.md or .github/.

Also note:

💡 AGENTS.md is the source of truth: GitHub Copilot, OpenAI Codex, OpenCode,
   Cursor, Gemini CLI, and VS Code read it natively — no sync needed for those tools.
   All other tools above received synced copies.

Also include a model normalization subsection:

🧠 Model normalization:

  Applied:
    - Claude Code: <N> agent file(s) normalized using target map

  Not applicable (no agent profile model field in this sync flow):
    - Gemini CLI
    - OpenAI Codex
    - OpenCode
    - Cursor
    - Windsurf
    - Cline
    - Roo Code
    - Kilo Code
    - JetBrains Junie
    - Zed
    - Augment Code

---

Sync Coverage Reference

Instructions

Agent	Primary File	Format	Sync Method
GitHub Copilot	AGENTS.md (read natively)	Markdown	Source of truth (no sync needed)
Claude Code	CLAUDE.md	Markdown	Full copy
Gemini CLI	GEMINI.md	Markdown + @imports	Live import
OpenAI Codex	AGENTS.md (read natively)	Markdown	Source of truth (no sync needed)
OpenCode	AGENTS.md (read natively)	Markdown	Source of truth (no sync needed)
Cursor	.cursor/rules/*.mdc	MDC (Markdown + YAML)	Translated
Windsurf	.windsurf/rules/*.md	Markdown	Full copy
Cline	.clinerules/*.md	Markdown	Full copy
Roo Code	.roo/rules/*.md	Markdown	Full copy
Kilo Code	.kilocode/rules/*.md	Markdown	Full copy
JetBrains Junie	.junie/guidelines.md	Markdown (single file)	Concatenated
Zed	.rules	Plain text	Full copy
Augment Code	.augment/rules/*.md	Markdown + YAML	Translated

Custom Agents

Agent	Path	Notes
GitHub Copilot	.github/agents/*.agent.md	Source of truth
Claude Code	.claude/agents/*.md	Copied at sync; model normalized for Claude
All others	❌ Not supported	No sub-agent profile concept

Skills (Open Standard)

Agent	Path	Notes
GitHub Copilot	.github/skills/<n>/SKILL.md	Source of truth
Claude Code	.claude/skills/<n>/SKILL.md	Verbatim copy
OpenAI Codex / OpenCode	.agents/skills/<n>/SKILL.md	Verbatim copy
All others	❌ Not supported	No skills concept

---

Docs references

• GitHub Copilot custom instructions: https://docs.github.com/copilot/customizing-copilot/adding-custom-instructions-for-github-copilot
• GitHub Copilot custom agents: https://docs.github.com/en/copilot/how-tos/use-copilot-agents/coding-agent/create-custom-agents
• GitHub Copilot agent skills: https://docs.github.com/en/copilot/how-tos/use-copilot-agents/coding-agent/create-skills
• Claude Code memory/instructions: https://docs.anthropic.com/en/docs/claude-code/memory
• Claude Code sub-agents: https://docs.anthropic.com/en/docs/claude-code/sub-agents
• Gemini CLI GEMINI.md + @import: https://google-gemini.github.io/gemini-cli/docs/cli/gemini-md.html
• OpenAI Codex AGENTS.md: https://developers.openai.com/codex/guides/agents-md/
• OpenAI Codex skills: https://developers.openai.com/codex/skills/
• OpenCode AGENTS.md + commands: https://opencode.ai/docs/customization
• Cursor rules: https://docs.cursor.com/context/rules-for-ai
• Windsurf rules: https://docs.windsurf.com/windsurf/customize#rules
• Cline custom instructions: https://docs.cline.bot/improving-your-workflow/cline-rules
• Roo Code custom instructions: https://docs.roocode.com/features/custom-instructions
• Kilo Code custom rules: https://kilo.ai/docs/features/custom-rules
• JetBrains Junie guidelines: https://www.jetbrains.com/help/junie/guidelines.html
• Zed rules: https://zed.dev/docs/ai/rules
• Augment Code rules: https://docs.augmentcode.com/using-augment/augment-rules
• VS Code custom agents (cross-tool format compatibility): https://code.visualstudio.com/docs/copilot/customization/custom-agents