Previously, publishing involved:

1. Write content in org-mode
2. Run Emacs batch export manually
3. rsync files to remote server
4. Verify deployment succeeded

Each step required context switching and manual verification.

With Claude Code, the workflow becomes:

1. Describe what you want to publish
2. Agent handles everything: file creation, export, deployment, verification

The agent autonomously:

• Creates or edits org files
• Runs Emacs publish commands
• Verifies deployment via version.json endpoint
• Tracks work in beads issue tracker
• Commits and pushes changes

The breakthrough was eliminating rsync entirely. Using Emacs TRAMP, org-mode can publish directly to remote servers via SSH:

(setq org-publish-project-alist
      `(("wal-sh-remote"
         :base-directory ,base-dir
         :publishing-directory "/ssh:jwalsh@www.wal.sh:/var/www/wal.sh/"
         ...)))

This means the agent can publish without any file transfer step - TRAMP handles SSH transparently.

Created version.json endpoint for automated deployment verification:

{
  "sha": "a1ed6c5",
  "timestamp": "2025-12-21T17:45:00Z",
  "status": "deployed"
}

The agent can fetch this after publishing to confirm the deployment succeeded.

Using beads/bd for distributed, git-backed issue tracking designed for AI agents:

bd new feature "Add CSS style prototypes" -p P2
bd close www.wal.sh-e4b
bd sync

This gives the agent a way to track multi-session work and maintain context.

With ~200k token context, the agent can:

• Hold entire publishing configuration in memory
• Reference previous changes across the session
• Maintain coherent multi-file edits

Opus 4.5 excels at chaining tools together:

1. Read project-config.el to understand publishing setup
2. Create new org file for content
3. Run gmake publish
4. Fetch version.json to verify
5. Create git commit with proper trailer
6. Push and sync beads

When publishing fails (FreeBSD path issues, missing dependencies), the agent:

• Reads error logs
• Identifies root cause
• Applies fix
• Retries operation

• Mermaid CLI doesn't work (puppeteer issues)
• Solution: Disable mermaid in batch mode

• GitHub Actions can't SSH to deploy server
• Solution: Use local/agent publishing, disable CI auto-trigger

• Agent loses context between sessions
• Solution: Detailed beads issues + conversation summaries