On this page
Open the issue in a browser. Copy the text. Switch to a terminal, cd into the right folder, start the agent, paste. You have probably done this dance a hundred times.
And the context was already there all along. The bug report, the steps to reproduce, your teammates' comments: it all lives in the GitHub issue. DevWorkspace Pro sends it straight to your agent, so you can skip the copy and paste.
In short:
- Connect with the GitHub App or your
ghlogin. - Send to AI on an issue starts your default agent in the project folder with the full issue and all comments.
- Send failure to AI on a red Actions run sends the failed job, step and the end of its log.
- Not ready to hand it off? Send the issue to Focus Pad as a task instead.
Connect GitHub
Open a project, then open the Git tab. The Issues and Actions sub-tabs work for every project with a GitHub repository, in any stack.
There are two ways to connect:
| Method | Use it when |
|---|---|
| GitHub App | You want live Actions updates and run notifications |
| GitHub CLI | You are logged in to gh and do not want to install the GitHub App |
DevWorkspace Pro finds the repository from the GitHub URL in the project settings. If that is empty, it uses the origin remote, then the first other remote that points to GitHub.
Your token stays in the main process of the app. The app's pages never get it. GitHub Integration has the details.
Send an issue to your AI tool
- In Git › Issues, open the issue.
- Click Send to AI. The dialog shows the full prompt.
- Optional: type more in Additional details, for example "Write a failing test first." The app adds it to the end of the prompt.
- Click Send, or press ⌘+Enter (Ctrl+Enter on Windows and Linux).
Email notifications not sent for new orders #137
Send to AI · acme-shop
acme/shop · issue #137
Email notifications not sent for new orders
github.com/acme/shop/issues/137
Description, plus 2 comments with author and date
Your default AI tool (claude, codex or opencode) starts in a terminal in the project root folder. The prompt includes:
- The repository and the issue number
- The title and the link
- The description
- All comments, with their author and date
The terminal tab gets a title such as Issue #42: Fix login redirect, so you can spot it in the status bar later.
Find the right issue quickly
The issue list has filters for Open, Closed, Assigned to me and Mentioned, plus Label and Assignee. The search field accepts GitHub search syntax, such as author:octocat or no:assignee.
| Key | Action |
|---|---|
| j or ↓ | Next issue |
| k or ↑ | Previous issue |
| Enter | Open the selected issue |
| / | Go to search |
| o | Open the issue on GitHub |
Send a failed CI run to your AI tool
A red CI run is often quicker to fix than an issue. The trouble is the log: it is long, and the failed step hides in it. In Git › Actions:
- Select the failed run. The log of the first failed job opens at the failed step, scrolled to the first error.
- Click Send failure to AI.
- Optional: add details, then click Send.
The prompt contains the repository, the workflow, the run number and attempt, the branch, the commit, the run link, the failed job, the failed step and the end of that step's log.
Need the context somewhere else? Copy failure for AI puts the same prompt on your clipboard.
Once the agent pushes a fix, the new run shows up in the run list. With the GitHub App, the list updates live and you get a notification when the run completes. For a flaky job, click Re-run failed jobs.
Not for the agent yet? Track it as a task
Some issues are not a job for an agent right now. Click Send to Focus Pad on the issue to add it to your Focus Pad checklist. When you complete that task, Focus Pad asks if you want to close the issue on GitHub.
Later, you can send the task from Focus Pad to your AI tool with the sparkle icon.
Why full context gives better results
An AI agent only knows what you tell it. A one-line prompt like "fix the login bug" leaves it guessing. Issue comments often hold the reproduction steps, the error message and decisions your team already made. Give the agent all of that, and it starts from the same facts you have.
It also saves you some task switching. Every manual step (browser, copy, terminal, folder, paste) is a small interruption. Research on interrupted work found that people make up for it by working faster, at the cost of more stress and time pressure (Mark et al., 2008).
Choose the AI tool and flags
Send to AI always uses your default tool. Set it in Preferences → Focus Pad:
| Setting | What it does |
|---|---|
| AI CLI Tool | claude (default), opencode or codex |
| AI CLI Tool Arguments | Extra flags for the default tool, for example --dangerously-skip-permissions |
The tool must be installed and on your PATH. See AI Sessions.
Restarted before the agent finished?
Sessions you start with Send to AI are saved for resume. If you quit the app before the agent finishes, click the to resume button in the status bar after a restart. See Resume Claude Code sessions.
References
- Mark, G., Gudith, D., & Klocke, U. (2008). The cost of interrupted work: More speed and stress. Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, 107–110. https://doi.org/10.1145/1357054.1357072
FAQ
Frequently asked questions
What does the prompt for an issue contain?
The repository, the issue number, title, link and description, and all comments with their author and date. You can add more text in Additional details.
Can I send a failed GitHub Actions run to the agent?
Yes. Send failure to AI makes a prompt with the repository, workflow, run number and attempt, branch, commit, run link, failed job, failed step and the end of the log of the failed step.
Do I need the GitHub App?
No. You can connect with the GitHub App or with your GitHub CLI login. Live Actions updates and run notifications need the GitHub App.
Where is my GitHub token stored?
In the main process of the app. With the GitHub App, the token is encrypted with the system keychain. With the GitHub CLI, the app does not keep a copy and gets the token from gh each time.