# Recipes

Copy-paste agent workflows for the feedback → context → iteration loop.

These assume you've [connected the MCP server](/docs/mcp). Paste a prompt into your agent, or schedule it to run on a cron. Each one drives the `feedback_*` tools end to end.

## Nightly triage

Pull the last day of new feedback, enrich it, and route it — status, priority, tags, and a one-line summary per item.

```text
Use the iterateTo MCP. List all feedback with status "new" from the last 24
hours. For each item: read it in full, then set a priority (high/medium/low),
add 1–3 tags (e.g. area + severity), and add a one-line internal note
summarizing the root cause. Move obvious duplicates to "archived" with a note
linking the canonical item. Finish with a table of what you changed.
```

## Feedback → GitHub issue

Turn actionable bug reports into issues, with the page, environment, and stable selectors already filled in.

```text
Use the iterateTo MCP. Find high-priority "problem" feedback with the tag
"needs-issue" that has no note starting with "gh:". For each, open a GitHub
issue in owner/repo: title = the summary, body = the message, page URL,
browser/OS, and the element's data-feedback-id selector. Then add an iterateTo
note "gh: <issue-url>" and set status to "in_progress" so it isn't filed twice.
```

## Close the loop

When a fix ships, tell the people who reported it — using the contact emails they left.

```text
Use the iterateTo MCP. List "resolved" feedback updated in the last 7 days that
has at least one contact email. For each, draft a short, friendly reply telling
the reporter the issue they raised is now fixed, referencing their original
message. Show me the drafts grouped by recipient before anything is sent.
```

> **Note:** Run triage on a schedule (nightly cron) and the close-the-loop recipe after each deploy to keep the queue moving without manual work.