Sync Containers

Sync Containers — Propagate Config to Docker

After updating skills, CLAUDE.md, AGENTS.md, hooks, or other agent config, propagate those changes to all claude-sandbox Docker containers so they stay in sync with the host.

Usage

/sync-containers

Optionally target specific containers: /sync-containers claude-sandbox-hq claude-sandbox-botpress

Process

1. Discover containers

List all claude-sandbox containers (running or stopped):

docker ps -a --filter "name=claude-sandbox" --format "{{.Names}}\t{{.Status}}"

Present the list to the user and confirm which containers to sync. Default: all claude-sandbox containers.

2. Determine what to sync

The following paths are synced from host to container:

3. Sync each container

For each target container, launch a background subagent (run_in_background: true) that:

1. Starts the container if not running (track whether we started it so we can stop it after)

2. Copies files using docker cp:

   # Skills and commands
   docker cp ~/.claude/skills/. <container>:/home/claude/.claude/skills/
   docker cp ~/.claude/commands/. <container>:/home/claude/.claude/commands/ 2>/dev/null || true

   # CLAUDE.md (resolve symlinks)
   cp -L ~/.claude/CLAUDE.md /tmp/claude-sync-CLAUDE.md 2>/dev/null && \
     docker cp /tmp/claude-sync-CLAUDE.md <container>:/home/claude/.claude/CLAUDE.md && \
     rm /tmp/claude-sync-CLAUDE.md

   # Settings
   docker cp ~/.claude/settings.json <container>:/home/claude/.claude/settings.json
   docker cp ~/.claude/settings.local.json <container>:/home/claude/.claude/settings.local.json 2>/dev/null || true

   # Agents
   docker cp ~/.agents/AGENTS.md <container>:/home/claude/.agents/AGENTS.md 2>/dev/null || true
   docker cp ~/.agents/skills/. <container>:/home/claude/.agents/skills/ 2>/dev/null || true
   docker cp ~/.agents/hooks/. <container>:/home/claude/.agents/hooks/ 2>/dev/null || true
   docker cp ~/.agents/.claude/. <container>:/home/claude/.agents/.claude/ 2>/dev/null || true

3. Rewrites paths for container environment:

   docker exec <container> bash -c "
     for f in /home/claude/.claude/settings.json /home/claude/.claude/CLAUDE.md; do
       [ -f \"\$f\" ] || continue
       sed -i 's|~/.claude/|/home/claude/.claude/|g' \"\$f\"
       sed -i 's|~/.agents/|/home/claude/.agents/|g' \"\$f\"
       sed -i 's|$HOME/.claude/|/home/claude/.claude/|g' \"\$f\"
       sed -i 's|$HOME/.agents/|/home/claude/.agents/|g' \"\$f\"
     done
   "

4. Sets permissions:

   docker exec <container> bash -c "
     chown -R claude:claude /home/claude/.claude/skills/ 2>/dev/null || true
     chown -R claude:claude /home/claude/.claude/commands/ 2>/dev/null || true
     [ -f /home/claude/.claude/CLAUDE.md ] && chown claude:claude /home/claude/.claude/CLAUDE.md
     [ -f /home/claude/.claude/settings.local.json ] && chown claude:claude /home/claude/.claude/settings.local.json
     chown -R claude:claude /home/claude/.agents/AGENTS.md 2>/dev/null || true
     chown -R claude:claude /home/claude/.agents/skills/ 2>/dev/null || true
     chown -R claude:claude /home/claude/.agents/.claude/ 2>/dev/null || true

     # Lock hooks and settings (root-owned, read-only)
     chown root:root /home/claude/.claude/settings.json
     chmod 444 /home/claude/.claude/settings.json
     if [ -d /home/claude/.agents/hooks ]; then
       chown -R root:root /home/claude/.agents/hooks/
       chmod 555 /home/claude/.agents/hooks/
       find /home/claude/.agents/hooks/ -type f -exec chmod 444 {} +
     fi
     if [ -d /home/claude/.claude/hooks ]; then
       chown -R root:root /home/claude/.claude/hooks/
       chmod 555 /home/claude/.claude/hooks/
       find /home/claude/.claude/hooks/ -type f -exec chmod 444 {} +
     fi
   "

5. Stops the container if we started it (restore to previous state)

4. Report

## Container Sync Results

| Container | Status | Synced | Notes |
|-----------|--------|--------|-------|
| claude-sandbox-hq | was running | Yes | — |
| claude-sandbox-botpress | was stopped → started → synced → stopped | Yes | — |
| claude-sandbox | was stopped → started → synced → stopped | Yes | — |

Files synced: skills/, commands/, CLAUDE.md, settings.json, AGENTS.md, hooks/

Integration with Other Skills

This skill should be run after:

• /update-skills — when skills are modified based on session analysis
• Manual edits to ~/.claude/skills/, ~/.agents/, or CLAUDE.md
• The self-improvement agent in /address-pr-comments (Step 8) makes approved changes

The /update-skills skill should suggest running /sync-containers after implementing approved changes if Docker containers exist.