Progress gets buried across Claude Code and Codex sessions. Keep it in one place.
New sessions, forks, machines and accounts spread a project's progress thin, until no single session knows where it stands. Sessions push what they achieved and what comes next to the project, so the next one, or a teammate's, starts from the project record instead of from scratch.
Free to start. Copy it from the Keeper Marketplace into your workspace.
Claude Code
Laptop · Tuesday
Achieved: export endpoint
Codex
Work desktop · Wednesday
Achieved: tests for totals
Claude Code
Priya · Thursday
Picks up the next step
Invoice export
Active- Next
- Add the totals row to the CSV
- Handoff
- handoff.md, revised Wednesday
- History
- 2 pushes, 1 edit by hand
Who it's for
Running many Claude Code and Codex projects
Several projects in flight, each worked across many sessions. See where they all stand without reopening a single session.
Switching tools and machines
Laptop and desktop, work and personal accounts, Claude Code and Codex: the project record follows you.
Teams sharing agent work
Colleagues see where each project stands and continue it from their own sessions.
Sessions on the same work report to one project
Each session knows only its own part of the work, and no other session can see it. Agent Projects keeps where the project stands outside any session, in a record you and your team can read.
Push a checkpoint
Ask your agent to push this session's progress. It records what was achieved, the next step and any status change, and shows you before it saves.
The newest push wins
The latest checkpoint sets what the project shows. History keeps every push with who made it, next to edits made by hand.
Recognised by its work
Projects remember their repository, branch, folders, plan file and key files, so an agent files progress on the right project and asks before starting a new one.
A living handoff
One handoff document per project: where it stands, the backlog, decisions and how to check the work. A fresh session reads it first and revises it in place.
How to run it
Setup, run, check, build and deploy steps for each repository, plus the gotchas, kept current by the sessions that use them.
Edit it yourself
Change the next step or the status without an agent. Your edit shows in History, so the next session sees it too.
This branch matches Invoice export by its plan file. I'll record:
- Achieved
- Tests for the CSV totals, all passing.
- Next
- Add the totals row to the CSV
- Status
- Active, unchanged
OK to push?
Switch sessions, machines and agents without starting over
No need to find the exact session that did the work, or explain the project again. Any session can start from the record.
Copy a prompt for a fresh session
Start a project here, Pick up in a session, Update from a session: each button copies a prompt with the workspace, app and project filled in.
Claude Code today, Codex tomorrow
Any session with the MechaBee plugins can read and push, on any machine and with any of your accounts.
Checks it's in the right place
A picked-up session compares its repository and branch with the project's, and tells you before it changes anything.
You approve what it records
Every prompt has the agent show what it will push and wait for your OK.
See every running project at once
Group projects by domain, line them up on a timeline, and let an agent suggest the order.
Timeline
Projects as bars by domain, with due dates, the next step on each bar and arrows between dependent projects. A week list on phones.
Dependencies
Record that one project needs or blocks another, with the reason.
Priorities review
An agent reads every running project and writes a dated review that ranks them, with a one-line reason each.
Domains and team
Group projects by product, client or business, and share a domain with colleagues. Members see only the domains they're given.
Statuses that mean something
Active, Waiting, Blocked, Paused, Done or Archived. An archived project takes no more pushes until you reopen it.
Aim higher: set goals your agent sessions work toward
Goals is a companion Keeper app for work bigger than one project. State the outcome you want; agent sessions plan it and do the work in Agent Projects, Issue Tracker, Content Roadmap, your repositories and folders, and the goal keeps track.
A plan you approve
An agent drafts the plan and the reasons behind it. Nothing starts until you accept it, and re-plans come back to you too.
Points to the real work
The goal links to records in your other apps, repositories and folders instead of copying them.
Your queue
Questions agents ask you and anything waiting on you, in one list across every goal.
Ready to test
When the work is done, an agent writes a validation report and the goal tells you it's ready for you.
Goals is optional. Agent Projects works on its own; add Goals when you want a layer above your projects.
Make it yours
Change it by describing what you need
Agent Projects is a Keeper app, so it isn't fixed. Ask the assistant in Keeper, or Claude Code or Codex with the App Author plugin, for the field, view or step your team is missing. You review the change before it goes live.
Your data, your workspace
Records live in your MechaBee workspace, versioned, and you can export them.
Reviewed before it's live
Every change to the app is checked and shown to you before your team sees it.
- “Add a Customer field to projects and group the list by it.”
- “Add a Review status between Active and Done.”
- “Ask for the hours spent with every push and total them per project.”
- “Show a weekly page of everything that was pushed.”
Getting started
- 1
Open Keeper
Sign up and open Keeper. Your account starts with free credit.
- 2
Copy Agent Projects
Pick Agent Projects in the Marketplace and copy it into your workspace. It's yours from then on.
- 3
Connect your agent
Install the Workspace Content plugin below in Claude Code or Codex, sign in once, then ask your session to push its progress.
Install the Workspace Content plugin. Your client asks you to sign in to MechaBee the first time a session uses it.
Claude Code
claude plugin marketplace add MechaBee/keeper-authoring-plugins
claude plugin install keeper-workspace-content@mechabeeThen start a new Claude Code session so the plugin is loaded.
Codex
codex plugin marketplace add MechaBee/keeper-authoring-plugins
codex plugin add keeper-workspace-content@mechabeeThen start a new Codex task so the plugin is loaded.
Questions
Which agents can push to it?
Does it read my sessions or my code?
What if two sessions push to the same project?
Can my team use it too?
What isn't included yet?
Stop rebuilding context every session
Copy Agent Projects into your workspace, connect Claude Code or Codex, and push your first checkpoint.