Skip to main content
Maestro includes a built-in feedback system that uses AI to help you craft well-structured GitHub issues. Instead of filling out a form, you have a short conversation - the AI asks clarifying questions, checks for duplicates, and submits a polished issue on your behalf.

​
Prerequisites

  • GitHub CLI (gh) must be installed
  • gh must be signed in to GitHub. If it is not, the Feedback modal offers Log in to GitHub, which runs gh auth login (one-time code plus browser) in a terminal inside Maestro and continues into the chat as soon as it finishes. Check Again re-checks if you signed in somewhere else.
Before the chat starts, Maestro also checks that GitHub will accept an issue from your gh account, so a refusal shows up front instead of after you write the report. The check sends GitHub an empty issue, which it authorizes and then rejects for having no title: nothing is created. If GitHub refuses the account (for example, an organization has not approved the GitHub CLI, or the token lacks a scope), the modal names the account gh is signed in as and says what to fix, offering Log in to GitHub when signing in again can fix it. If filing still fails later because the gh login expired or was revoked, the same Log in to GitHub button appears beside the error and files the issue again once you are signed in.

​
Sending Feedback

​
1. Open the Feedback Modal

Click the Feedback button in the bottom-left corner of the sidebar, next to New Agent. Feedback and New Agent buttons You can also open it via Quick Actions (Cmd+K / Ctrl+K) → “Send Feedback”.

​
2. Which Account Runs the Chat

Maestro picks the account for you. It looks at the accounts your own agents run as (Claude Code, Codex, or OpenCode, including any per-agent CLAUDE_CONFIG_DIR or CODEX_HOME) and uses the first one that is installed and signed in. The provider’s default login and accounts on SSH remotes are tried last. If the first account cannot answer (for example, its login expired), the chat sends your message to the next account instead of failing. The account that answered is remembered for next time. The Running as menu at the very bottom of the window, under the message box, shows the account in use. Pick another one there to override it.

​
3. Describe Your Issue

Tell the AI what’s on your mind - a bug you hit, a feature you want, or general feedback. The AI classifies your input and asks targeted follow-up questions:
  • Bug reports - What happened? What was expected? Steps to reproduce?
  • Feature requests - Use case? Desired outcome? Why it matters?
A confidence bar at the top tracks how well the AI understands your issue. As you answer questions, it fills toward 100%. Feedback conversation Optional attachments:
  • Drag and drop up to 5 screenshots (PNG, JPG, GIF, or WebP - 10 MB each)
  • Check Include support package to attach debug information

​
Live Diagnostics

The feedback agent runs on your own machine, so it can look at the problem instead of asking you to describe it. When the chat opens, Maestro reads its own live state and hands it to the agent: every agent that is busy, errored, paused on a usage limit, retrying, holding queued messages, or running Auto Run, the auto-resume settings, and the recent warnings and errors from the log (renderer console errors included). If that already shows what you are about to describe, such as an Auto Run still parked after its limit reset, the agent says so in its first reply instead of asking. Paths and secrets are redacted the same way the support package redacts them. While it thinks, it may check your Maestro version, read the tail of the debug log, or query the running app with maestro-cli (status, doctor, list agents, and other read-only verbs). What it finds is folded into the issue as context you would otherwise have to gather by hand. Two guarantees:
  • It is read-only. The agent is launched in the provider’s CLI-enforced read-only sandbox (--permission-mode plan for Claude Code, --sandbox read-only for Codex, --agent plan for OpenCode). It cannot write files, install anything, or change a setting. Commands that mutate Maestro are off limits.
  • Nothing is hidden. Every command it runs is listed under the thinking indicator while the turn is in flight, so you can see exactly what was inspected.
The agent is told to stay inside Maestro’s own logs and configuration. Your source files, git history, and personal documents are out of scope.

​
4. Review and Submit

You never have to finish the interview. As soon as you have said anything, a Submit now button sits beside the progress bar. It files the issue with whatever has been gathered so far, even while the agent is still thinking: your own words fill any field the agent has not, and the issue notes it was submitted early. Your screenshots, performance trace, and support package (if checked) go with it. Once understanding reaches 80%, the button turns into a green Submit Feedback. The AI presents a structured summary of what it will submit. Review the summary, tweak anything by continuing the conversation, then click Submit. Every filed issue carries a Live Diagnosis section when something in the app looked wrong at the time: each affected agent’s provider, state, error type, Auto Run progress, and pause or retry status, plus the auto-resume settings. It never includes agent names, error text, or paths. Submit feedback

​
Closing Without Losing Anything

Closing the feedback window never throws away a conversation. If you have work in progress, the X button, Escape, and clicking the backdrop all park it: the agent keeps working in the background and the sidebar Feedback button carries a draft indicator. Click it again to pick up where you left off. Minimize does exactly the same thing. To actually throw the conversation away, use the trash button in the window’s header. It asks for confirmation first, then stops the agent and clears the draft.

​
Duplicate Detection

As you describe your issue, Maestro searches existing GitHub issues in the background. If similar issues are found, an inline card appears listing them. You can:
  • Subscribe to an existing issue - Your context is added as a comment with a +1 reaction
  • Create a new issue anyway - Bypasses the match and files a fresh issue
This prevents duplicate reports while still capturing your unique context.

​
After Submission

Once submitted, you’ll see a confirmation with:
  • A direct link to the new GitHub issue
  • A copy-to-clipboard button for the issue URL
  • An option to open the issue in your browser

​
From the Command Line

Agents and scripts can file the same issue through maestro-cli, with no modal:
submit stops when it finds likely duplicates; pass --force to file anyway. maestro-cli open feedback opens the modal. Full flags are in the CLI reference.