What did you ship this week, and which prompts got you there?

We're open sourcing Agent Retro, a plugin that turns your agent history into an animated weekly recap: eight stats, four coaching tips, and three things to try.

It reads local Claude Code and Codex logs, with optional ChatGPT and Claude data exports. You'll see top projects, token usage, repeated requests, and how your tone changes between agents.

Over time, retros compare weeks and rotate in fresh stats.

Identify patterns

If you keep asking for the same review, Agent Retro might suggest turning it into a reusable command. If clearer briefs seem to help, it can offer a line to try in your next prompt.

It looks at your replies, like “great” or “that's not what I meant,” for clues about what worked. Those clues don't prove why something went well, so treat the tips as things to try. At least two of the four focus on habits that already seem to be helping.

We designed Agent Retro to give you insight into how you work with agents, and to make your AI workflows better over time.

Your data

Stats are calculated locally, and scripts redact common secret and personal-information patterns before your agent sends the text to its model provider to write the deck.

Agent Retro produces a local HTML deck, optionally publishable with a private-page tool. Temporary raw prompt files are deleted after the run, and a redacted digest stays on your machine.

Make your first retro

Free, MIT-licensed, and built for Claude Code and Codex (including the ChatGPT desktop app). Requires macOS or Linux with python3.

In Claude Code:

For Codex, in your terminal:

Then ask Codex to “build my Agent Retro.”

Or ask your agent to install it:

Install the Agent Retro plugin from <https://github.com/summationai/agent-retro>

Install the Agent Retro plugin from <https://github.com/summationai/agent-retro>

Install the Agent Retro plugin from <https://github.com/summationai/agent-retro>

Install the Agent Retro plugin from <https://github.com/summationai/agent-retro>