BuildGist

What tools summarize what an AI coding agent did in a session?

Kayvan Zahiri ·

Start with git diff --stat and the agent's own last message, then pick a tool for what you need next. Session timeline tools such as promptarc let you browse every prompt, tool call and diff. Self-hosted summarizers such as Codesteward's Session Summarizer turn audit logs into structured reports. BuildGist is a Mac app that writes a plain-English explanation of each meaningful Claude Code, Cursor or Codex run: what changed, what it means and what to check. You can read one on the sample run.

Why the agent's last message is not enough

The agent's closing message says what it believes it did. Often that matches the code. Sometimes a step is missing, a test was skipped, or a change landed in a file you did not expect. The diff shows what changed, but not why, and not what the agent ran in the shell. A useful summary ties those together: the request, the files, the tests, and anything that needs a second look.

Do it by hand first

  1. Run git status and git diff --stat to see which files changed and by how much.
  2. Read the diff for the files that matter most: sign-in, payments, deleting data, migrations.
  3. Run the tests yourself rather than relying on "all tests pass" in the chat.
  4. Ask the agent: "List every file you changed and why, and anything you did not finish." Compare its answer with the diff.

This costs nothing and works with any agent. For more on the diff, see BuildGist or the git diff and reviewing AI code before you merge. It gets slow with several sessions a day.

Tools that help

These are tools people use for this, each described from its own page or repository as read on October 25, 2026. Check each one for what it supports today.

Pick by what you need. For the full record of every tool call, a timeline tool fits. If your team already runs an audit proxy and wants reports on its own servers, a self-hosted summarizer fits. To find and reopen old sessions, a session browser fits. For a short explanation you can read in a minute after each run, with what to check, that is what BuildGist adds.

Where BuildGist fits

After each meaningful run, BuildGist puts an explanation in your Inbox. Small changes stay quiet, and uncommitted work counts, so you do not need to commit or push. Each explanation has two parts:

On the public sample run, Claude Code added share links to a notes app. BuildGist recorded 3 files changed (+46 −2) and 7 tests passing. The explanation noted that the links do not expire yet, which the developer had asked to do later, and raised two concerns: the line that creates each share token is masked, so it cannot be checked that tokens are hard to guess (medium), and share tokens are not cleared when a note is deleted (low).

One Inbox covers all three agents. When you switch agents, Continue hands the task to the new chat so you do not explain it again.

Limits worth knowing: an explanation can include edits you made by hand between agent turns, and BuildGist cannot always tell who made an edit. Explanations are written by AI, so check important behavior against the code and tests. There is no monthly cap on explanations, but each account has a daily cost limit; past it, new explanations wait until the next day begins (00:00 UTC), and your agents keep working as usual.

Questions

Can Claude Code, Codex or Cursor summarize their own session?

You can ask any of them to, and it is a good habit. The summary reflects what the agent believes it did, so check it against git diff and a fresh test run.

Is reading the git diff enough?

It shows what changed, not why, and not what ran in the shell or which tests passed. For small changes it is often enough. For sign-in, payments or data changes, read a summary and the diff together.

Does BuildGist work on Windows, Linux or Intel Macs?

Not today. It runs on Macs with Apple silicon and macOS 12 or later, for people in the United States who are 18 or older.

What you need

Codex needs you to approve BuildGist once, in Codex. BuildGist shows the steps. Setup is in Getting started.

To explain a run, your computer sends BuildGist what that run needs, and BuildGist passes it to Anthropic. BuildGist does not keep this material. It does not send your whole repository, files the run did not change, or files named like secrets, such as .env files and keys. It keeps the explanation and short cited excerpts of changed code, at most 12 lines each. Secret masking is best-effort. Privacy has the full list.

Every account gets 7 days of BuildGist Pro free, with no card. The 7 days start at your first explanation or Continue handoff. Nobody is charged when the trial ends. After that, BuildGist Pro is $12 per month, plus any applicable tax shown at checkout. See pricing.

Try BuildGist Download for Mac

Apple silicon, macOS 12 or later. United States, 18 or older.

Related

Questions: kayvanandre@gmail.com.