for technical founders, with an editor or without one

Show what you shipped, in a script your engineers can check.

Ask Claude Code, in your own repo, to show what you built. It writes the script from your commits and notes, saves it to ReadyToEdit through the MCP and puts your screen recordings on its lines. Play it as a Footage preview and post it yourself, or send one link to your editor.

$ git log --oneline

  1. a91c3f2chore: bump deps
  2. 7be41d0stream export rows instead of buffering the file
  3. c2d17f9progress: drop the total
  4. 4e0b8aaremove export queue worker
  5. e58a1b0fix flaky billing test
  6. 19fd6c2wip
  7. d4c9e37fix
  8. 6a2f0e1docs: export limits

drive / release-2.4 /

  • memory-profile.png
  • notes/streaming-tradeoffs.md
  • export-run-2.3.mov

release video · draft 3

  1. This release deletes the part that held the whole file in memory.7be41d0
  2. Peak memory went from 3.1 gigabytes to 38 megabytes.memory-profile.png
  3. What it cost: the progress bar can no longer show a total.streaming-tradeoffs.md

Numbers match each source to the line that cites it. Grey: read and left out.

Fig. 1ExampleA week of commits and the release folder, and the three script lines they became, each ending in its reference.

Release 2.4, on the record

  1. GeneratedMon 09:14

  2. Changes requestedMon 16:02

    “The opening skips the two years we spent buying hardware. Say that part.”

  3. In reviewTue 08:31

  4. ApprovedTue 11:47

How the week changes

The same week, without the briefing call.

--- today

+++ with ReadyToEdit

@@ -1,3 +1,4 @@ your release week

  1. 1

    Today: You ship every week and post maybe once a month, always about something three commits ago.

  2. 2

    Today: When you do post, the brief is a Slack paragraph, a Drive folder and a call to explain what changed.

  3. 3

    Today: The first cut shows a wrong flag, and the revision round goes to an engineer fixing it.

  4. 1

    With ReadyToEdit: At the end of the week you ask Claude Code, in the repo, to show what you built. The script it saves is about this release.

  5. 2

    With ReadyToEdit: You play it as a Footage preview, your screen recording under each line, and rewrite the line that does not hold up.

  6. 3

    With ReadyToEdit: The wrong flag is caught in the script, before the first cut, and no revision round goes to fixing it.

  7. 4

    With ReadyToEdit: Posting it yourself, the preview is the plan. With an editor, they get one link: the script, a clip for each beat, the original files. No account, no call.

On Light

Claude Code turns what you shipped into the script.

Connect the ReadyToEdit MCP to Claude Code once. In your repo, ask it to show what you built: it reads the commits and notes it can see, writes the lines, uploads the screen recording you made and puts it under the lines it shows. ReadyToEdit gives it your audience and style, and turns the script into a Footage preview and an editor link.

  • The preview is the result

    Cutting or posting it yourself? The Footage preview plays your recordings in script order, so you see the video before you cut it.

  • An editor when you want one

    The same script becomes a read-only link with a clip and range on each line. No account on their side.

  • Your code stays with you

    Claude Code reads the repo on your machine. ReadyToEdit receives the script and the files it saves.

Start with show-what-you-built, one of the ReadyToEdit skills for Claude, Codex and Cursor. Light includes about 10 footage hours of playback a month.

Claude Code, in ipsum/placeholder-api

You: show what you built this week

Read 8 commits and notes/streaming-tradeoffs.md

Assistant: The one worth a video is 7be41d0, “stream export rows instead of buffering the file”: exports stream now, so a large export no longer fills memory. Want the demo recording on it?

You: yes, export-demo-2.4.mov is on my desktop

Uploaded export-demo-2.4.mov to ReadyToEdit

Assistant: Saved “Exports stream now” to ReadyToEdit, five lines. Line 2 plays export-demo-2.4.mov 0:41–0:52. Line 4 has no clip yet and says so. The Footage preview and the editor link are below.

Works with Claude Code, Codex, Cursor and other assistants that sign in to an MCP server.
Fig. 2ExampleOne chat in the repo, from the week’s commits to a Footage preview.

If an editor cuts it

One link for your editor. No account, no call.

The approved script, beat by beat, each beat with the clip placed on it and a player for that range only; or the whole script in one piece. Originals open where they live in your Google Drive, a Download handoff button gives them a text file for their own notes, and you can revoke or replace the link.

The read-only editor handoff for the example Release 2.4 video: five script beats, three with a clip and two marked as having none, beside the evidence the script cites and the list of original files.
  1. Whole script: the approved text in one piece, to read straight through.
  2. Each beat names its clip. Press play and only that beat’s range runs.
  3. A beat with nothing behind it stays visible and empty. No clip is invented for it.
Fig. 3ExampleThe handoff page as your editor opens it, for the 2.4 release. No sign-in on their side.

Every line, annotated

Full plans

Read the script the way you read git blame.

Each line of the draft carries the commit, file or timecode it was written from, so an engineer can check it as fast as they read it. A line with nothing behind it is marked, and you decide what to do with it.

release video · draft 2 · Ipsum Analytics

  • commit
  • recording
  • screenshot
  • note
  • reviewer
  1. 1

    Two years ago somebody asked why our CSV export dies at a million rows.source 1

  2. −

    Struck by the reviewer: This release is about why the fix removed code instead of adding workers.

    struck ·Changes requested, Mon 16:02: “The opening skips the two years we spent buying hardware. Say that part.”

  3. 2

    Added by the reviewer: We added workers. We added a queue. We paid for a bigger box.

    edited Tue 08:31 ·Written by the reviewer. The revision is kept with the script.

  4. 3

    On 2.3, an export of 2.4 million rows gave up at sixty seconds.source 2

  5. 4

    This release deletes the part that held the whole file in memory.source 3

  6. 5

    Peak memory went from 3.1 gigabytes to 38 megabytes.source 4

  7. 6

    The same export now finishes in 41 seconds.

    no source ·Nothing in the material shows the 2.4 run. Record it, cite it or cut the line.

  8. 7

    What it cost: the progress bar can no longer show a total.source 5

Sources

  1. 1

    support-thread.pngscreenshot

    A support thread titled “CSV export fails on anything over a million rows”: 38 replies, reopened four times. Names removed.
  2. 2

    export-run-2.3.mov0:04–0:12

    A terminal. curl asks the 2.3 API for 2,400,000 rows and gets 504 Gateway Timeout after real 1m0.214s.
  3. 3

    7be41d0Priya N. · Thu 14:22

    stream export rows instead of buffering the fileRows go to the response as they are read. The in-memory buffer and the export queue worker are removed.
  4. 4

    memory-profile.pngscreenshot

    Peak memory per export: 3.1 GB on 2.3, 38 MB on 2.4.
  5. 5

    notes/streaming-tradeoffs.mdengineering note

    “The progress bar can no longer show a total: we do not know the row count until the last row is sent.”
In reviewTue 08:31 · opening revised; line 6 still has no source
Fig. 4ExampleThe 2.4 release video script after one revision round, each line beside the commit, recording, screenshot or note it was drafted from.

What it reads, exactly

Commits, recordings and notes, down to the field.

GitHub commits
Commit history, through a GitHub App on the repositories you choose: each commit’s message, SHA, author and time. New pushes arrive as they happen.notDiffs, pull requests, issues or the code itself.
Recordings, clips
mp4, mov and webm. It watches and transcribes them, so a line can cite a timecode and you can search the footage for a moment. Full plansnotA description can miss what is on screen. The timecode is printed so you can look.
Screenshots
png, jpg and webp, described, so a claim can point at the image that shows it. Full plans
Docs and notes
Markdown, PDF and other documents: engineering notes, specs, the support thread you saved.
Past posts
What you already posted on X, LinkedIn and YouTube (Instagram and TikTok too), so a draft knows what you have said before. For a draft it posted to Bluesky or Mastodon: the likes, replies and reposts it gets afterwards.notScheduling, or replying to anyone for you.
Google Drive
A folder or single files, imported by reference. The handoff links your editor to the originals.

One commit, as it is kept

sha
7be41d0c93a1f5e8b2d64f0a9c7e13b5d8f2a604
message
stream export rows instead of buffering the file

Rows go to the response as they are read. The in-memory buffer and the export queue worker are removed.
author
Priya N.
time
2026-09-10 14:22
repo
ipsum/placeholder-api

ExampleEvery field it keeps from a commit, for the one cited in Fig. 4.

How it finds things

Full plans

It looks everything up before it writes.

Under the draft is retrieval-augmented generation over your own project. Before it writes a line or answers a question, it searches the commits, recordings, screenshots and notes you connected, and each claim keeps what it used and why. Each line in Fig. 4 names what it came from; line 6 has nothing, and is marked.

What it stops writing

Correct it the same way enough times and it stops.

It learns from the corrections you actually make, not from a guess at your taste. Cut the same kind of line out of enough drafts and it stops writing it.

Each habit is written down on What I’ve learned with the edits behind it, where you can correct it or throw it out. There is one list per workspace, and every project in it shares that list. On Light, your own Claude or Codex reads the list through the ReadyToEdit MCP before it writes; on Full plans, ReadyToEdit’s drafts start from it too.

  • exclamation

    Exports finally stream!!

    “You’d rather state a thing than exclaim it.”

    5 supporting · 1 against

  • hype words

    A blazing-fast, next-level export path

    “You cut the marketing words out of what I write.”

    6 supporting · 0 against

  • opening

    In today’s data-driven world, exports matter more than ever. Two years ago somebody asked…

    “My openings do not survive. You write your own.”

    3 supporting · 0 against

ExampleStruck text is a correction you made before posting. The quotes are as What I’ve learned prints them.

What it will not do

Known limitations.

  • Read your recordings on Light. Claude Code names the clip and range for each line; Pro and Team watch and transcribe every recording.
  • GitHub only. There is no GitLab, Bitbucket, Linear, Notion or Frame.io connector.
  • It does not edit, cut, render or generate video. A person does the edit.
  • It posts a draft’s text to Bluesky or Mastodon and nowhere else, after you connect once per project and confirm the exact thread. For X, LinkedIn, YouTube and the rest, you paste it yourself and press Mark as posted.
  • No score, and no prediction of how a video will perform.
  • Frame descriptions can miss what is on screen. Every cited moment carries its timecode so you can look.
  • One set of learned habits per workspace, shared by every project and channel in it.
How it compares with a Claude or ChatGPT project

Light

Your own AI, through the MCP

$20/month

Starts with a 14-day trial, no card.

About 10 footage hours of playback a month (2,250 credits), up to 3 projects.

Start your free trial

Creates your workspace and starts the trial.

Several recordings a week

Pro

Adds ReadyToEdit’s own AI

$149/month

About 120 half-hour recordings a month (25,000 credits), up to 3 projects.

Start your free trial

Starts with the Light trial. Choose Pro any day.

Team

Adds ReadyToEdit’s own AI

$299/month

About 288 half-hour recordings a month (60,000 credits), up to 10 projects.

Start your free trial

Starts with the Light trial. Choose Team any day.

Light is ReadyToEdit’s production workspace, with your own Claude, Codex or Cursor as the writer, connected through the ReadyToEdit MCP. Pro and Team add ReadyToEdit’s own AI.

After the trial, the workspace is read-only until a plan is chosen: everything in it stays, and links you sent keep working. Choose a plan in Settings, during the trial or after it, and cancel there at any time.

Questions

Does it read our code?
ReadyToEdit reads commit messages, never the code. Claude Code reads what you let it in your own repo, on your machine, and ReadyToEdit receives only the script and the files it saves. Every field ReadyToEdit keeps from a commit is listed under “What it reads, exactly”, above.
Our commit messages are “fix” and “wip”. Does that still work?
Not on their own. A commit message is thin evidence. Add the screen recording, a screenshot or a short note from the engineer and the draft has something to cite. A line with nothing behind it is shown without a source, for you to back up or cut.
Why not Claude Code on its own?
It writes a good script. What it does not have is the rest: your screen recordings playing under its lines as a Footage preview, the record of what you and your reviewer changed, and a read-only link your editor can cut from. Connect ReadyToEdit and Claude Code saves the script there, with the clips on its lines.
Do I need an editor?
No. If you cut or post it yourself, the script and its Footage preview are the finished plan: play it, fix the weak line, record what is missing. When someone else cuts, they open one read-only link with no account.
Does it edit or post the video?
No. A person edits the video and you publish it. The one thing it posts itself is a draft’s text, to Bluesky or Mastodon, after you confirm it.

Connect Claude Code. Show what you built.

Connect the ReadyToEdit MCP, add the last screen recording you made, and ask Claude Code in your repo to show what you built. A new workspace starts a 14-day Light trial with no card.