Grok BotSoftwareCode reviewAI score · 73/100AI-assisted edit · no obvious prompt-injection riskGitHub

Engineer Bot Template: Customer Feedback to Pull Request

Take confirmed customer feedback, reproduce it on the live product, scope and implement a fix via cloud agent, verify, and open a PR with before/after proof

Some experienceStepsWorkflowPromptTip
Who
Engineers or small teams who need to turn confirmed customer feedback into code fixes quickly
You get
One PR per feedback item with reproduction proof; pauses and escalates to human if a product decision is required
You need

Constraints

This template comes from stream observation notes for configuring a bot that handles customer feedback. Before use, prepare: a confirmed feedback source (e.g. ticket board BOARD), a target code repository (REPO), an accessible live product URL (URL), and a verification skill (VERIFY SKILL). Replace the placeholders in the template with actual values, then paste it as the bot's system prompt or role description. Humans retain two approval gates: choosing which feedback to act on, and deciding whether to allow autopilot merge to production. The source does not provide specific bot platform configuration steps or pre-built integrations.

What they did

Source notes record that Shub demonstrated a feedback-to-ship loop on stream using a bot named ProtoBot. On day 3, an autopilot fix introduced a bad SQL query and brought production down, leading to the addition of a /verify gate.

How it went

Author report (author_report): the full feedback→ship→deploy loop can happen in hours; however, autopilot merge once caused a production incident, so verification skill passing is now required before merge.

Worth knowing

1. This template is reconstructed from third-party stream observation notes, not an official verbatim release; edit and verify before use. 2. The source does not provide specific bot platform configuration, cloud agent invocation details, or verification skill implementation. 3. This site has not tested the workflow (reproduction=not_tested); validate after setup. 4. Humans must retain feedback selection and merge approval rights; do not fully automate. 5. Variable names, JSON field names, code, and runtime prompts remain in original language; surrounding instructions are translated consistently.

Steps

  1. Create the bot and paste the role description

    Create a new bot in your bot platform and paste the provided role description into the system prompt or role configuration field. Fill in all placeholders: {NAME} (bot name), {BOARD} (confirmed feedback source), {REPO} (target repository), {URL} (product verification URL), {VERIFY SKILL} (verification skill name).

    Setup guidance · Source text。Bot system prompt or role description field

    You are {NAME}. You turn confirmed feedback from {BOARD} into PRs on
    {REPO}. For each item I approve: reproduce it on {URL} with your own
    account; scope the smallest change that resolves it; run a cloud agent
    to implement it; verify on {URL} again with {VERIFY SKILL}; open a PR
    with before/after evidence.
    
    If the feedback implies a product decision (new feature, changed
    behaviour users rely on), stop and ask me instead of guessing.
    
    We are live in production: never merge without {VERIFY SKILL} passing.
  2. Configure input sources and verification environment

    Ensure the bot can access {BOARD} to read confirmed feedback, can access {URL} with its own account to reproduce issues, and has {VERIFY SKILL} configured for post-fix verification. The source does not specify connection details; configure according to your platform's capabilities.

  3. Set approval gates and boundary rules

    Confirm human approval items: the bot cannot decide which feedback to act on (human chooses), and cannot merge to production (gated on VERIFY SKILL passing). Unconfirmed feedback must go to triage, not this bot.

  4. Trigger execution and monitor outputs

    When a confirmed ticket arrives or the human issues "Pull the most recent customer feedback", the bot executes: pull feedback → reproduce → scope minimal fix → run cloud agent → verify again → open PR with before/after evidence. If feedback implies a product decision (new feature, changed behavior users rely on), the bot stops and asks the human.

Prompt you can copy

Setup guidance · Editorial suggestion。Hand to setup assistant; confirm inputs and permissions first

This block is the prompt. Ads are never placed here.

Please use the following role description template to configure a customer-feedback bot. Fill in the placeholders and set approval gates. Confirm you have ready: a confirmed feedback source (BOARD), a target repository (REPO), a live product URL (URL), and a verification skill (VERIFY SKILL).

Replace all placeholders with actual values before pasting; always keep the VERIFY SKILL gate for production

Source

feedback-to-pr · unicodef1wn · GitHub · 2026-09-18

Comments & feedback

Tried this method? Share results, problems or improvements. Comments appear after moderation. Please avoid sensitive information.

Loading…

    Related