Awesome-omni-skill sync-agents
Synchronize GitHub Copilot instructions, custom agents, and skills into detected AI coding agent configurations in this repository. Use when asked to mirror .github/copilot-instructions.md, .github/instructions, .github/agents, or .github/skills into Claude, Codex, Cursor, Gemini, Windsurf, and related tooling.
git clone https://github.com/diegosouzapw/awesome-omni-skill
T=$(mktemp -d) && git clone --depth=1 https://github.com/diegosouzapw/awesome-omni-skill "$T" && mkdir -p ~/.claude/skills && cp -r "$T/skills/data-ai/sync-agents-noahjenkins" ~/.claude/skills/diegosouzapw-awesome-omni-skill-sync-agents-2270a7 && rm -rf "$T"
skills/data-ai/sync-agents-noahjenkins/SKILL.mdSync 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 (Copilot / source of truth):
— global instructions.github/copilot-instructions.md
— path-scoped instructions.github/instructions/*.instructions.md
— custom agent profiles.github/agents/*.agent.md
— agent skills (open standard).github/skills/<name>/SKILL.md
Your Task
Execute the following steps in order, creating or overwriting files as needed. Never delete
.github/copilot-instructions.md or anything inside .github/.
Step 1 — Read the source of truth
Read
.github/copilot-instructions.md and store its full content.
If the file does not exist, stop and tell the user:
"No
found. Please create this file first — it is the source of truth for all AI agent instructions.".github/copilot-instructions.md
Also scan and store the contents of:
— any.github/instructions/
files. For each, record the filename,*.instructions.md
frontmatter glob (if present), and body content.applyTo
— any.github/agents/
files. For each, record the full filename, the complete YAML frontmatter block (*.agent.md
,name
,description
,model
,tools
, etc.), and the body prompt.handoffs
— any subdirectory containing a.github/skills/
file. For each skill, record the directory name, the completeSKILL.md
frontmatter (SKILL.md
,name
,description
,user-invokable
, etc.), the body content, and note any additional bundled asset files (scripts, examples, references) present in that directory.disable-model-invocation
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 | at repo root or directory | Full sync |
| Gemini CLI | at repo root or directory | Full sync |
| OpenAI Codex | at repo root or directory | Full sync |
| OpenCode | directory or | Full sync |
| Cursor | directory or file | Full sync |
| Windsurf | directory or file | Full sync |
| Cline | file or directory | Full sync |
| Roo Code | directory or file | Full sync |
| Kilo Code | directory | Full sync |
| JetBrains Junie | directory | Full sync |
| Zed | file at repo root | Instructions only |
| Augment Code | directory | Full sync |
If no agents are detected, tell the user:
"No agent configuration directories or files were detected in this repository. To enable syncing, initialize at least one agent's configuration (e.g., create a
file, aCLAUDE.mddirectory, etc.) and re-run this command.".cursor/
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 (
or the specific instructions file).github/copilot-instructions.md - 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: .github/copilot-instructions.md ║ ║ To update: run /sync-agents ║ ╚══════════════════════════════════════════════════════════════╝ --> # Claude Code Instructions > **Source of truth:** [`.github/copilot-instructions.md`](.github/copilot-instructions.md) > > This file is automatically generated by the `/sync-agents` command. > To propagate changes, update `.github/copilot-instructions.md` and re-run that command. --- [INSERT FULL CONTENT OF .github/copilot-instructions.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: .github/copilot-instructions.md ║ ║ This file uses Gemini CLI's native @import syntax so it ║ ║ always loads the latest copilot instructions at runtime. ║ ║ Run /sync-agents to update path-specific files. ║ ╚══════════════════════════════════════════════════════════════╝ --> # Gemini CLI Instructions > **Source of truth:** `.github/copilot-instructions.md` > This file dynamically imports from Copilot's instruction file using Gemini's > `@file` syntax. Any edits to `.github/copilot-instructions.md` are reflected > here automatically — no manual sync required. @.github/copilot-instructions.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 — OpenAI Codex
Global instructions → AGENTS.md
<!-- ╔══════════════════════════════════════════════════════════════╗ ║ AUTO-SYNCED — DO NOT EDIT THIS FILE DIRECTLY ║ ║ Source of truth: .github/copilot-instructions.md ║ ║ To update: run /sync-agents ║ ╚══════════════════════════════════════════════════════════════╝ --> # Codex Agent Instructions > **Source of truth:** [`.github/copilot-instructions.md`](.github/copilot-instructions.md) > > This file is automatically generated by the `/sync-agents` command. > To propagate changes, update `.github/copilot-instructions.md` and re-run that command. --- [INSERT FULL CONTENT OF .github/copilot-instructions.md HERE]
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.
3D — OpenCode
OpenCode reuses
at the repo root (same as Codex). If AGENTS.md
AGENTS.md was already
written in Step 3C, OpenCode will automatically pick it up — no additional file needed.
If OpenCode is detected but Codex is NOT, write
AGENTS.md using the same format as Step 3C.
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 .github/copilot-instructions.md — do not edit directly # Source of truth: .github/copilot-instructions.md | Re-sync: /sync-agents description: "Global project instructions synced from GitHub Copilot" alwaysApply: true --- [INSERT FULL CONTENT OF .github/copilot-instructions.md HERE]
Path-scoped instructions → .cursor/rules/<name>.mdc
For each
.github/instructions/<name>.instructions.md, create .cursor/rules/<name>.mdc:
- Set
from the filename or first headingdescription - Set
from theglobs
frontmatter value (if present)applyTo - Set
(the glob handles activation)alwaysApply: false - 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 .github/copilot-instructions.md — do not edit directly --> <!-- Source of truth: .github/copilot-instructions.md | Re-sync: /sync-agents --> # Global Project Instructions [INSERT FULL CONTENT OF .github/copilot-instructions.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 .github/copilot-instructions.md — do not edit directly --> <!-- Source of truth: .github/copilot-instructions.md | Re-sync: /sync-agents --> # Global Project Instructions [INSERT FULL CONTENT OF .github/copilot-instructions.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 .github/copilot-instructions.md — do not edit directly --> <!-- Source of truth: .github/copilot-instructions.md | Re-sync: /sync-agents --> # Global Project Instructions [INSERT FULL CONTENT OF .github/copilot-instructions.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 .github/copilot-instructions.md — do not edit directly --> <!-- Source of truth: .github/copilot-instructions.md | Re-sync: /sync-agents --> # Global Project Instructions [INSERT FULL CONTENT OF .github/copilot-instructions.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 .github/copilot-instructions.md — do not edit directly --> <!-- Source of truth: .github/copilot-instructions.md | Re-sync: /sync-agents --> # Project Guidelines [INSERT FULL CONTENT OF .github/copilot-instructions.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 .github/copilot-instructions.md — do not edit directly # Source of truth: .github/copilot-instructions.md | Re-sync: /sync-agents [INSERT FULL CONTENT OF .github/copilot-instructions.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 .github/copilot-instructions.md — do not edit directly # Source of truth: .github/copilot-instructions.md | Re-sync: /sync-agents always_apply: true --- # Global Project Instructions [INSERT FULL CONTENT OF .github/copilot-instructions.md HERE]
Path-scoped instructions → .augment/rules/<name>.md
For each
.github/instructions/<name>.instructions.md, create .augment/rules/<name>.md:
- Set
andalways_apply: false
in the YAML frontmatteragent_requested: true - Include a
derived from the filename ordescription
valueapplyTo - 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 | , , , | , , , | |
Normalization rules:
- If model is target-native, keep unchanged.
- If model is non-native but recognized, rewrite to target default mapped value.
- If model is missing, inject target default mapped value.
- 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:
- Create
(drop the.claude/agents/<name>.md
infix).agent - Copy YAML frontmatter and body, and apply Step 3M model normalization policy using the Claude target map.
- 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 |
|---|---|
, , , | Keep as-is |
, , , | |
| Missing/unknown | |
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). The format is
identical across Copilot, Claude Code, and Codex — only the storage path differs.
| Tool | Skills path | Supported? |
|---|---|---|
| GitHub Copilot (source) | | ✅ Source |
| Claude Code | | ✅ If detected |
| OpenAI Codex / OpenCode | | ✅ If detected |
| All others | — | ❌ No skills concept |
For each
.github/skills/<skill-name>/ directory:
If Claude Code is active:
- Create
directory.claude/skills/<skill-name>/ - Write
verbatim with a YAML sync commentSKILL.md - Copy any bundled asset files as-is
If Codex or OpenCode is active:
- Create
directory.agents/skills/<skill-name>/ - Write
verbatim with a YAML sync commentSKILL.md - Copy any bundled asset files as-is
Important —
field constraint: The name
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: .github/ (copilot-instructions.md, agents/, skills/, instructions/) 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 .github/ update.
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 | | Markdown | Source of truth |
| Claude Code | | Markdown | Full copy |
| Gemini CLI | | Markdown + | Live import |
| OpenAI Codex | | Markdown | Full copy |
| OpenCode | (shared w/ Codex) | Markdown | Full copy |
| Cursor | | MDC (Markdown + YAML) | Translated |
| Windsurf | | Markdown | Full copy |
| Cline | | Markdown | Full copy |
| Roo Code | | Markdown | Full copy |
| Kilo Code | | Markdown | Full copy |
| JetBrains Junie | | Markdown (single file) | Concatenated |
| Zed | | Plain text | Full copy |
| Augment Code | | Markdown + YAML | Translated |
Custom Agents
| Agent | Path | Notes |
|---|---|---|
| GitHub Copilot | | Source of truth |
| Claude Code | | Copied at sync; normalized for Claude |
| All others | ❌ Not supported | No sub-agent profile concept |
Skills (Open Standard)
| Agent | Path | Notes |
|---|---|---|
| GitHub Copilot | | Source of truth |
| Claude Code | | Verbatim copy |
| OpenAI Codex / OpenCode | | 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 +
: https://google-gemini.github.io/gemini-cli/docs/cli/gemini-md.html@import - 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