My latest project: SparkDrop — A content pipeline I built for me and my agents to run together.
How it works
Agents run a daily process:
I use a Claude Cowork scheduled daily task, which runs a Skill that instructs Claude to do the following:
- Pitches ideas (sparks) based on their training & inputs, pushes into SparkDrop.
- Finds the Sparks I green-lit for development, writes those drafts for various platforms and formats (article, LinkedIn, tweet, YouTube script, etc.)
- Reads my comments/feedback on drafts that I want rewritten. Rewrites those.
- Writes my newsletter based on the drafts I marked as sources for this week.
- Identifies learnings that can be folded back into our training (patterns in my feedback, new interests, audience trends, industry trends, etc.).
My daily process:
- Review new ideas (sparks). Approve, reject, or give feedback.
- Review new drafts. Approve, reject, or give feedback.
- 1-click scheduling of drafts. They auto-slot into my schedule.
- Review Claude's learnings, approve for integrating into our training.
Agent is trained on:
- Everything I do is logged in my file system (collected by OpenClaw agents daily):
- Everything I publish (YouTube, podcasts, tweets, newsletters)
- All my GitHub & Claude Code activity
All my journals, notes, etc.
- A radar scan of tweets from key people and companies I follow in my industry.
- Stored training in SparkDrop (voice, source-specific, platform-specific, context-specific).
- My comments on drafts in SparkDrop help inform new learnings to integrate into training.
Just finished building (2-week project, Claude Code powered), now actively using and tweaking daily.