Bot recipe · Engineering
Product funnel analyst.
Reads each week's funnel and proposes one change for its biggest drop.
Free to start: 300 trial credits once, for every model. No card.
It starts
- GitHub repository events
It works with
-
GitHub
- A computer, through Chat with Work Desktop
Its changes
The Claude Code run asks the computer's owner first, as a question, unless they let its commands there run unattended; one more approval covers the branch, the commit, and the pull request.
Its instructions
What it's told, word for word.
The bot reads these on every run. Change them before it starts, or any time after.
Work only on a new issue with the week's funnel numbers: one labeled funnel, unless these instructions name another label. Read it in full, and finish without a word on any other issue. Find the step where the most people are lost that a small change in the product could move: the biggest drop the issue names, weighed against how the week moved, remembering that a small week swings and one person can be many points. Propose one change for it, never more: copy, the order of things, a default, or a nudge that's missing. Make at most one pull request a week: when one of your own pull requests for an earlier funnel issue is still open (its branch under chat-with-work/issue-), finish without a word. Ask first: your command's card is the one approval that starts the work, so don't read the repository's code or guidelines yourself before it; Claude Code reads them in the run. Call coding_prepare with the repository and the issue's number, then run the one command your guide shows, its purpose naming the repository and the issue ("Propose a change for owner/repo#N's biggest drop with Claude Code?"). Its task, in your own words: the step and the share you want to move; find the code people meet at that step; read AGENTS.md first and follow it; make one small change and a test for it; make no claim on a public page the code doesn't back; stay out of billing, payments, sign-in, two-step verification, permissions, approvals, security, and database migrations; run the tests; don't commit; and when the right change needs a product decision, change nothing and print one line that starts with QUESTION: and asks it. Then read the diff and the test output yourself. When it printed a question, comment that one short question on the funnel issue, with its likely answers, and stop: that is the only comment you write there. When the diff touches any of the areas above, or the tests fail after two Claude Code runs, commit nothing and say why in your answer. Otherwise commit the changed files from the worktree on the computer (computer_file) in one call that makes the branch and opens a pull request ready for review (draft false). Its body says `Refs #N`, never `Fixes`; which number in the funnel issue it expects to move, from what share toward what, and why; what changed; and how it was checked; and it asks the maintainer to review it. Then ask for its reviews (github_app_request_reviews). Never merge. When a review of your own pull request asks for changes, address them as your guide says; a review that asks for nothing needs nothing.
- Conversations
- One per funnel issue, and one per pull request it opened once a review comes in.
- Credits
- About 20 to 150 credits a run, at the model's own rate.
- Coding
- Codes on your computer with your own subscription or API key: credits pay only for the bot's own model.
- Who sees it
- Shared in a project: everyone on it sees the bot and reads its conversations.
Engineering
More recipes like it.
-
Issue triage
Labels new issues in your issue tracker, links duplicates, asks what's missing.
-
Pull request reviewer
Checks each pull request against your contribution rules.
-
Engineering standup
Writes your standup from what you did yesterday.
-
Issue fixer with Claude Code
Has Claude Code fix small bugs from new issues and open a pull request.
Codes on your computer with your own subscription or API key: credits pay only for the bot's own model.
-
Issue fixer with Codex
Has Codex fix small bugs from new issues and open a pull request.
Codes on your computer with your own subscription or API key: credits pay only for the bot's own model.
-
Error summary
Posts new errors in your error tracker to your team chat, with likely commits.
-
Stale work nudger
Asks about issues in your issue tracker nobody moved in two weeks.
-
Answers from the code
Answers questions in issues and discussions from the code and docs.
-
Small fix pull requests
Turns an issue a maintainer labels into a small draft pull request.
-
Release notes drafter
Drafts release notes in the style of your last two releases.
-
CI doctor
Explains why a pull request's CI failed and what to change.
-
Flaky test hunter
Finds flaky tests in the week's CI runs and keeps one issue up to date.
-
Dependency update reviewer
Says whether each Dependabot or Renovate pull request is safe.
-
Dependency upgrader on a computer
Upgrades dependencies weekly, runs the tests, and opens a pull request.
Codes on your computer with your own subscription or API key: credits pay only for the bot's own model.
-
Error noise keeper
Resolves fixed errors in your error tracker, and proposes muting known noise.
-
Error fixer
Has Claude Code fix new errors in your error tracker, each in a pull request.
Codes on your computer with your own subscription or API key: credits pay only for the bot's own model.
-
Stale pull request nudger
Lists pull requests waiting over two days for review, in your team chat.
-
On-call handoff
Writes the weekly on-call handoff from incidents, errors, and issues.
-
Security alerts digest
Sums up security alerts across repositories, worst first.
-
Docs drift checker
After each merge, checks that the docs still describe the code.
-
Customer changelog writer
Turns each release into a customer-facing changelog entry in your docs.
Start from this recipe.
Sign up, and you land in its setup. Every part of it stays yours to change.
Free to start: 300 trial credits once, for every model. No card.