A free kit for one agent, and ClearTrace for agents that hand work to each other.
1. Where we left off
In the last edition I explained why I built ClearTrace: every meaningful task should close with a change capsule, a short record of what changed, why, what was tested, and what is still open. This edition is about why that record matters more every month, and the two ways you can start keeping it today. One of them is free and open source.
2. Code got cheap
I watch Nate B. Jones every day while my agents work. In his recent conversation with Vijaye Raji, OpenAI's CTO of Applications, one idea stayed with me: now that code is cheap, more of it gets thrown away. Agents try several versions, teams keep one, and the rest gets cleaned up. Engineers get faster, and the wait moves to review, testing and deciding what to keep. (the conversation)
If the code is disposable, it stops being the asset. The asset is the decision:
- what was tried,
- why it was kept or thrown away,
- which tests ran, and what they actually returned,
- what the agent did not verify,
- what is still open for the next person, or the next agent.
Git keeps the diff. Everything else lives in a chat window that scrolls away. A week later nobody can say whether "all tests pass" meant the full suite, one file, or nothing.
3. Option one: the free kit (one agent, one repo)
I published change-capsules, a small open-source kit (MIT): https://github.com/ognaranjo/change-capsules
- Your agent writes one Markdown file per finished task into
.capsules/, committed with the work: why, what changed, which tests ran with the real results, evidence, risks and follow-ups. - Before a new task, the agent reads the open capsules, so unfinished work isn't forgotten.
- Rules for Codex (
AGENTS.md) and Claude Code (CLAUDE.md), plus a small checker. - No account, no service, no network, no dependencies.
If you work with one agent in one repository, this may be all you need. It's my contribution back to the community I learn from every day.
4. Option two: ClearTrace (several agents handing work to each other)
My own setup outgrew a folder of files. I'm a solo developer, but I don't run one agent. One agent writes code, another tests the build, and another reads what users report and turns it into work. They run on different machines and in different repositories. The hard part is no longer writing the code. It's the handoff: the testing agent has to know what the coding agent changed and why, and the coding agent has to know what the tester found.
That's what ClearTrace does. The agents write to one shared record through a CLI, an API and an MCP server, while they work:
- tickets, findings and test runs, with the evidence;
- change capsules for every closeout, with the commits and the files across repositories;
- what's still open and what's waiting for you.
A different agent can verify the work. The next agent, on another machine or a teammate's, starts from the record, not from zero. You review in the portal, or you just ask your agent: "what changed, what's open, what's waiting for me?"
It works the same for one developer with three machines and for a small team where every person runs their own agents. I'm its first customer: ClearTrace is built with Codex and Claude, and every change is recorded in ClearTrace itself.
Two minutes on the method: https://youtu.be/DmjcCO-Z9rw
5. Founding beta
For readers of this newsletter: 90 days free, no card, on any self-serve plan. Invites are personal. To get one, message me on LinkedIn or request an invite through the contact page, and tell me which agents you run and on how many repositories. I'll send you your own link. If you subscribe after the free period, the founding discount (30% off for as long as you stay) applies while places last.
6. What I'd like from you
Try the kit on one repository this week and tell me what a capsule from your agent looks like, good or bad. Reply here or open an issue on GitHub. If you run several agents, I'd value 20 minutes with you to watch how you hand work between them.
References
- Nate B. Jones with Vijaye Raji, OpenAI: https://www.youtube.com/watch?v=S6SEgMooZpE
- change-capsules (MIT): https://github.com/ognaranjo/change-capsules
- ClearTrace: https://cleartrace.dev/?utm_source=devto&utm_medium=article&utm_campaign=nl4
- Edition 3, "From AI Workflow to a Shared Record: Why I Built ClearTrace"
Want to try ClearTrace free for 90 days? Message me on LinkedIn or request an invite.
Top comments (0)