How do I install the Import Config skill?

Run npx skillmds@latest add rachidchabane/import-config in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Import Config safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Import Config?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Import Config free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Import Config?

RachidChabane (@rachidchabane) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Import Config by rachidchabane — Agent Skill

Name: Import Config
Author: RachidChabane

Import Config

Reverse-engineer an existing hand-maintained AI config (Claude Code / Copilot / OpenCode / VS Code) into an agentsync config/, then prove nothing is lost. Use to migrate onto agentsync without rebuilding config by hand.

RachidChabane 3fb0cdf 2 files · 6.0 KB Updated Sep 17, 2026

File contents

Import an existing config into agentsync

Turning a live, hand-built setup into a config/ is judgment (dedup, decide what's core vs passthrough, find skill sources) made safe by a deterministic diff-proof. Do not switch the user's real config until diff-prove.sh is clean.

1. Read the live config (resolve symlinks to the real source)

Claude — ~/.claude/settings.json, ~/.claude/CLAUDE.md (follow @imports), MCP via claude mcp list / ~/.claude.json, skills in ~/.claude/skills/ (note where each symlinks to — that's its source).
Copilot — ~/.copilot/{settings.json, mcp-config.json, copilot-instructions.md}, ~/.copilot/skills/.
OpenCode — ~/.config/opencode/opencode.json.
VS Code — Code/User/settings.json (instructions + hook locations).

2. Synthesize `config/` (the judgment)

instructions.md — the shared instruction text (merge per-tool variants into one).
skills.json — tier per skill (from skillOverrides / disabledSkills / permission.skill). Where a skill dir symlinks to a real source, record source (+ subpath) so agentsync manages it; for git-backed skill repos use the git URL.
mcp.json — the union of MCP servers across tools, deduped by identity (same server defined per-tool collapses to one). Mark secret-bearing ones with auth.
profile.json — the harnesses in use.
overrides.json — everything agentsync's four core concerns don't cover but the live config sets: plugins/marketplaces, status line, model effort, custom notification hooks (under hooks, merged additively), editor prefs. This catches what would otherwise be lost.

3. Prove nothing is lost (deterministic — required)

bash skills/import-config/diff-prove.sh

It applies config/ into a sandbox seeded with the live files and diffs the result against the originals. Iterate config/ until it prints CLEAN. A DIFF means a key or hook the live config has isn't reproduced — usually something to add to overrides.json. Don't proceed on any unexplained diff.

4. Only then switch

With a clean diff-proof, the real agentsync apply is safe — every change it would make is already accounted for. Keep the old config repo as rollback until you've run a session in each harness. agentsync uninstall reverts agentsync's changes if needed.

---name: import-configdescription: Reverse-engineer an existing hand-maintained AI config (Claude Code / Copilot / OpenCode / VS Code) into an agentsync config/, then prove nothing is lost. Use to migrate onto agentsync without rebuilding config by hand.---# Import an existing config into agentsyncTurning a live, hand-built setup into a `config/` is **judgment** (dedup, decide what'score vs passthrough, find skill sources) made **safe** by a deterministic diff-proof. Donot switch the user's real config until `diff-prove.sh` is clean.## 1. Read the live config (resolve symlinks to the real source)- **Claude** — `~/.claude/settings.json`, `~/.claude/CLAUDE.md` (follow `@imports`), MCP via `claude mcp list` / `~/.claude.json`, skills in `~/.claude/skills/` (note where each symlinks to — that's its source).- **Copilot** — `~/.copilot/{settings.json, mcp-config.json, copilot-instructions.md}`, `~/.copilot/skills/`.- **OpenCode** — `~/.config/opencode/opencode.json`.- **VS Code** — `Code/User/settings.json` (instructions + hook locations).## 2. Synthesize `config/` (the judgment)- `instructions.md` — the shared instruction text (merge per-tool variants into one).- `skills.json` — tier per skill (from `skillOverrides` / `disabledSkills` / `permission.skill`). Where a skill dir symlinks to a real source, record `source` (+ `subpath`) so agentsync manages it; for git-backed skill repos use the git URL.- `mcp.json` — the union of MCP servers across tools, **deduped** by identity (same server defined per-tool collapses to one). Mark secret-bearing ones with `auth`.- `profile.json` — the harnesses in use.- `overrides.json` — everything agentsync's four core concerns don't cover but the live config sets: plugins/marketplaces, status line, model effort, custom notification hooks (under `hooks`, merged additively), editor prefs. This catches what would otherwise be lost.## 3. Prove nothing is lost (deterministic — required)```bash skills/import-config/diff-prove.sh```It applies `config/` into a sandbox seeded with the live files and diffs the resultagainst the originals. **Iterate `config/` until it prints CLEAN.** A `DIFF` means a keyor hook the live config has isn't reproduced — usually something to add to`overrides.json`. Don't proceed on any unexplained diff.## 4. Only then switchWith a clean diff-proof, the real `agentsync apply` is safe — every change it would makeis already accounted for. Keep the old config repo as rollback until you've run a sessionin each harness. `agentsync uninstall` reverts agentsync's changes if needed.

Frequently asked questions