ln-004-agent-sync
Sync skills (symlinks) and MCP settings from Claude to Gemini CLI and Codex CLI
- License 许可
- MIT
- 最近更新
- 2026-11-10
- Stars
- ★ 12
- Forks
- 3
01 前言 Front Matter
- name
- ln-004-agent-sync
- license
- MIT
- description
- Sync skills (symlinks) and MCP settings from Claude to Gemini CLI and Codex CLI
03 SKILL.md 正文
Paths: File paths (
shared/,references/,../ln-*) are relative to skills repo root. If not found at CWD, locate this SKILL.md directory and go up one level for repo root.
Agent Sync (Standalone Utility)
Type: Standalone Utility Category: 0XX Shared
Synchronizes skills and MCP server configurations from Claude Code (source of truth) to Gemini CLI and Codex CLI. Creates symlinks for skills, copies/converts MCP settings.
When to Use This Skill
- After adding/removing MCP servers in Claude Code settings
- After installing new plugins in Claude Code
- First-time setup of Gemini CLI or Codex CLI alongside Claude Code
- Periodic sync to keep all agents aligned
Input Parameters
| Parameter | Required | Default | Description |
|---|---|---|---|
| targets | No | both | gemini, codex, or both |
| mode | No | full | skills (symlinks only), mcp (MCP only), or full (both) |
| dry_run | No | false | Show planned actions without executing |
Workflow
Detect OS → Discover Configs → Sync Skills → Sync MCP → Report
Phase 0: OS Detection
| Check | Result | Impact |
|---|---|---|
uname or platform |
win32 / darwin / linux | Junction vs symlink |
| Home directory | $HOME or $USERPROFILE |
Config paths |
Paths by OS:
| Agent | Windows | macOS / Linux |
|---|---|---|
| Claude | %USERPROFILE%\.claude\settings.json |
~/.claude/settings.json |
| Gemini | %USERPROFILE%\.gemini\settings.json |
~/.gemini/settings.json |
| Codex | %USERPROFILE%\.codex\config.toml |
~/.codex/config.toml |
Phase 1: Discover Current State
-
Read Claude settings:
- Parse
~/.claude/settings.json→ extractmcpServersblock - If no
mcpServers→ WARN "No MCP servers configured in Claude", skip MCP sync
- Parse
-
Read target configs (if exist):
- Gemini: Parse
~/.gemini/settings.json→ extract existingmcpServers - Codex: Parse
~/.codex/config.toml→ extract existing[mcp_servers.*]
- Gemini: Parse
-
Detect installed plugins:
- Glob
~/.claude/plugins/*/plugin.json→ list plugin directories - Also check if skills repo itself is a plugin source
- Glob
-
Check existing symlinks:
~/.gemini/skills→ exists? points where?~/.codex/skills→ exists? points where?
-
Show current state:
Current State: | Agent | Skills | MCP Servers | Config Exists | |-------|--------|-------------|---------------| | Claude (source) | 2 plugins | 4 servers | yes | | Gemini | no link | 2 servers | yes | | Codex | → ~/.claude/plugins | 4 servers | yes |
Phase 2: Sync Skills (symlinks/junctions)
FOR EACH target IN (gemini, codex) WHERE target in targets:
-
Determine link path:
- Gemini:
~/.gemini/skills - Codex:
~/.codex/skills
- Gemini:
-
Check if already linked correctly:
- IF link exists AND points to correct source → SKIP (already synced)
- IF link exists AND points to wrong source → WARN, ask user before replacing
- IF regular directory (not link) exists → WARN "Target is a real directory, not a link. Skip to avoid data loss."
-
Determine source:
- IF single plugin: link directly to plugin dir
- IF multiple plugins: ask user which plugin to share
-
Create link:
OS Command Windows cmd /c mklink /J "{target_path}" "{source_path}"macOS / Linux ln -s "{source_path}" "{target_path}" -
Verify: Check link exists and resolves correctly
Phase 3: Sync MCP Settings
Source: ~/.claude/settings.json → mcpServers block
3a: Claude → Gemini (JSON → JSON)
- Read Claude
mcpServersas JSON object - Read Gemini
settings.json(or create{}if missing) - Merge strategy: Claude servers override Gemini servers by key name. Gemini-only servers preserved.
- Write updated
settings.json
Conversion: None needed — identical format.
3b: Claude → Codex (JSON → TOML)
-
Read Claude
mcpServersas JSON object -
Read Codex
config.toml(or create empty if missing) -
Convert each server:
Claude JSON field Codex TOML field Notes commandcommandSame argsargsJSON array → TOML array env[mcp_servers.{name}.env]Nested table type: "http"+urlurlHTTP transport type: "sse"+urlurlSSE transport Example conversion:
"context7": { "command": "npx", "args": ["-y", "@upstash/context7-mcp"] }[mcp_servers.context7] command = "npx" args = ["-y", "@upstash/context7-mcp"] -
Merge strategy: Claude servers override. Codex-only servers preserved.
-
Write updated
config.toml
Unsupported conversions (preserve as-is in Codex):
bearer_token_env_var— no Claude equivalentenabled_tools/disabled_tools— no Claude equivalent
Phase 4: Report
Sync Complete:
| Action | Target | Status |
|--------|--------|--------|
| Skills symlink | Gemini | Created → ~/.claude/plugins/... |
| Skills symlink | Codex | Already linked |
| MCP sync | Gemini | 4 servers synced (2 new, 2 updated) |
| MCP sync | Codex | 4 servers synced (1 new, 3 updated) |
Critical Rules
- Claude = source of truth. Never write TO Claude settings. Read-only source.
- Non-destructive merge. Target-only servers/settings preserved. Only Claude servers added/updated.
- No data loss. If target is a real directory (not symlink) — warn and skip, never delete.
- Backup before write. Before modifying any config file, create
.bakcopy. - Dry run first. If
dry_run=true, show all planned actions without executing. - Ask on conflict. If symlink points to different source — ask user, don't auto-replace.
Anti-Patterns
- Writing TO Claude settings from Gemini/Codex (reverse sync)
- Deleting target-only MCP servers during sync
- Creating symlinks inside symlinks (circular)
- Modifying config files without backup
Definition of Done
| # | Criterion |
|---|---|
| 1 | Claude settings read successfully |
| 2 | Skills symlinks created/verified for each target |
| 3 | MCP settings synced with format conversion (JSON→TOML for Codex) |
| 4 | Backup files created before any config modification |
| 5 | Report shown with all actions and warnings |
Version: 1.0.0 Last Updated: 2026-02-15
04 同仓库更多 Skills
web-access
Use Treg for public web or platform data only when existing search, browser and connected tools cannot cover the task, with explicit cost and data boundaries.
treg
Use the Treg CLI or MCP catalog to discover, price and call authorized external tools with team-scoped credentials and a user-approved budget.
recruit
Turn a job description into an evidence-linked candidate shortlist through Treg, with job-related ranking and optional authorized business contact enrichment.
rea-growth
Analyze a company’s public growth and content through Treg, distinguish observations from hypotheses, and produce an evidence-linked playbook within an approved budget.
make-ugc
Create original AI UGC-style product videos through Treg: research hooks, approve a licensed presenter and voice, generate clips, add captions and report costs.
lead-signals
Find and qualify public buying signals through Treg, rank fitting business accounts and optionally enrich authorized business contact information.