Delivery clarity for teams that build with AI agents
Know what your AI agents changed, why, and what proves it.
Your agents report each task to ClearTrace through the project kit, CLI, API or MCP. You see the change, the reason, where it landed, the tests and evidence, and what is still open.
- Works with Codex and Claude Code
- Read-only app for work records
- No document text stored
Assign appears to do nothing for weekend jobs
- Dispatcher“Assign does nothing for weekend jobs”Sent from the portal's feedback box
- Claude CodeTriaged against the codeNot reproduced · ticket opened
- CodexFound the causeReassign disabled with no message · screenshots
- Claude CodeFixed in the Web PortalCommit bbb1b6e · 27 tests passed
- CodexRetested and closed6 acceptance checks passed
See it work
ClearTrace in six and a half minutes.
The method behind it, one real ticket handled by Claude Code and Codex, the questions the record answers, plans and releases, and the founding beta.
The problem
Implementation can move faster than authoritative context.
Agents now change many files across several repositories in one session. The work gets done. What gets lost is the record: why it changed, what it touched, and what proved it.
- 01
One room, not the building
An agent working in one repository does not know the contract the other repositories depend on.
- 02
Tickets before, nothing after
Teams plan the work in a ticket. Few record what actually happened once it is done.
- 03
"Tests pass" is not proof
A sentence in a chat proves little. A test run with its scope and result does.
What you can answer
Six questions about every change your agents close out.
AI creates speed. Speed without visibility creates doubt. ClearTrace keeps the answers to these questions in one place, task by task.
- 01
What changed?
Files, commits and the reviewed diff.
- 02
Why?
The ticket, finding or reason the agent recorded. A reason is optional.
- 03
Where?
Solution, project, repository, service, table or flow.
- 04
What proved it?
The test run or scenario that checked it.
- 05
Which evidence?
Test output, screenshots and logs linked to the change.
- 06
What is still open?
Doubts, risks and decisions that remain.
How it works
Three steps, then it runs with every task.
Your agents do the writing. You agree the map and read the record.
- 01
Connect
Create a workspace and an API key. Your agent installs the project kit and scans each repository, with a preview first.
- 02
Agree the map
You and your agent declare databases, environments, services and flows. That map becomes the truth for your architecture.
- 03
Close out every task
The agent records the change, the reason, the proof and the doubts. You read it in the app.
Built in practice
It comes from real products, not from a slide.
ClearTrace is the workflow I used to build my own products, turned into a tool.
- Where the need appeared
WhoPlays
Years building one product across mobile, backend, data and integrations showed the limit: AI could write each piece, but it did not know the whole system.
- Where the method took shape
Reconcilia AI
A bootstrap control plane, shared contracts and change capsules let Codex and Claude work across several repositories. Its agents now close out their work in ClearTrace.
- Built with the same method
ClearTrace
ClearTrace is built the same way. Beta feedback goes into its own Intake, and agents triage it into tickets.
The method
The developer is not replaced. The developer is promoted.
ClearTrace comes from From Developer to Architect, my newsletter on building software with AI agents. The architect owns the intent, the boundaries and the proof. The agent implements. ClearTrace keeps the record between them.
Beta
Start with one project.
Connect one repository, run the first scan, and let your agents close out their next tasks. You will see what changed, why, and what is still open.