Run workstream_from_video

To file a GitHub issue with Claude Code from a recording, run the workstream_from_video skill. It reads what your agent receives from the recording and files one GitHub issue against your code, then you follow the work on Agent Workstreams.

File a GitHub issue with Claude Code

From the recording's Give to your agent panel, copy the prompt and paste it into your agent:

The agent then:

1. waits for the transcript if the recording is new,
2. reads the timed transcript, every mark and the image of each mark,
3. searches your repository for where each ask lands,
4. files one GitHub issue with a row per ask, and starts the work.

A recording never becomes more than one issue, however many asks it holds.

What the issue contains

Each ask gets a row: when you said it, the mark you drew with it, your words, and one to three file and line anchors. If the agent cannot find where an ask lands, it writes "anchor: none found" instead of guessing.

For example:

An issue that already exists

If someone already filed an issue from a recording, run:

The agent reads the recording linked from the issue and posts the same table as one comment, so any ask the written ticket dropped is restored.

Someone else's recording

Open their share link and use its Give this to your agent panel. The prompt it copies includes --token, which lets your agent read that one recording.

Your own tracker

add_issue files one or more issues (up to ten per run) in your own tracker, grounded in your code, and queues each on its own workstream board.

Many recordings, many repos

To file recordings later without picking the repo yourself, send them to the change request queue. A watcher agent you leave running files each one in the repo it belongs to.