Do user request

Besides fixing specific Bugbot findings, you can ask the bot to perform general code changes in the repository: add tests, refactor a function, implement a small feature, update docs, etc. This is called do user request. The same permission and workflow setup as Autofix apply: only org members or the repo owner can trigger it, and the workflow must grant contents: write.

This page explains how to use it and how it differs from autofix.

How it works

When you comment on an issue or pull request, the action first runs intent detection (OpenCode Plan): it decides whether your comment is:

• A fix request — “fix it”, “fix all”, etc. → Autofix runs (fix specific findings).
• A do request — “add a test for X”, “refactor this”, “implement feature Y”, etc. → Do user request runs (general code change).
• Neither — e.g. a question → Think runs (answer only, no file changes).

So you don’t choose a “mode”; you just write what you want. If the agent classifies it as a do request and you have permission, the action runs the Build agent with your request, then runs the same verify commands as for autofix and commits and pushes.

How to ask

Write a comment on the issue or on the PR (or, for PRs, you can reply in a review thread if the context makes sense). OpenCode receives your sanitized comment (trimmed, length-limited, escaped for the prompt) and the repo context.

Example phrases:

• “add a unit test for the login function”
• “refactor this to use async/await”
• “add a README section for installation”
• “implement the missing validation in the form”
• “add error handling for the API call”
• “fix the typo in the docstring”

You can be brief or detailed. The Build agent will apply the changes in the OpenCode workspace; the action then runs verify commands (from bugbot-fix-verify-commands) and, if they pass, commits and pushes.

Permissions and workflow

• Who can trigger: Same as Autofix: organization members (for org repos) or the repository owner (for user repos). Others get a Think response only.
• Workflow: The workflow that runs on issue_comment or pull_request_review_comment must grant contents: write so the action can push.
• Branch: On issue comment, the action resolves the branch from an open PR that references the issue (same as autofix). On PR comment or PR review comment, it uses the PR’s head branch.

Commit message

After a successful do-user-request run, the action commits with a message like:

• chore(#123): apply user request when the issue number is known, or
• chore: apply user request when there is no issue context.

So you can distinguish these commits from autofix commits (fix(#N): bugbot autofix - resolve ...) in the history.

Verify commands

Do user request uses the same verify commands as autofix: bugbot-fix-verify-commands. If any command fails, the action does not commit. See Configuration.

When to use autofix vs do user request

Intent is inferred by OpenCode from the comment text and the list of unresolved findings; you don’t need to tag or label the comment.

Next steps

• Autofix — Fix specific Bugbot findings from a comment.
• Configuration — bugbot-fix-verify-commands and OpenCode inputs.
• Examples — Comment examples and workflows.