Skip to content

Lead slices tasks around where the user works, not the CLI first - #221

Merged
jamescrosswell merged 3 commits into
mainfrom
docs/ui-first-slices
Sep 27, 2026
Merged

jamescrosswell merged 3 commits into
mainfrom
docs/ui-first-slices

Conversation

@jamescrosswell

@jamescrosswell jamescrosswell commented Sep 27, 2026 •

Copy link
Copy Markdown
Contributor

Summary

The #34 and #174 breakdowns each shipped a CLI-only first task (#203, #213 → #220) and put the dashboard command a task or three later, so the reviewer ended up approving plumbing the dashboard and agents call. The #174 breakdown cited "#34's pattern" for its order.

  • roles/lead.md: a slice is judged by the vision's user, from where the vision says they work. A command, API or script the product's own code or agents call is a layer, so it ships inside the slice that puts it in front of the user. Kept product-agnostic: for a product whose users do live on the CLI, the vision says so and CLI slices are still fine.
  • Acceptance criteria are end results only. a-team try shows them as the reviewer's checklist, so each item is something checked by using the product, never how it's built (classes, APIs, internal commands, config keys, file formats). Those go under Tests. I can stop a run from the command line, and it stays stopped #213's criteria listed CLI commands, dispatch.hold and exit codes.
  • roles/dev.md: the PR body says what the user can now do or see, and any choice that changes that beyond the task. How it was built goes in the commit messages. a-team stop ends a run and holds the role until you resume it #220's summary and notes were about which script stop lives in and the order of its writes.
  • docs/vision.md: Where features land already ruled out a command with only a CLI form; this makes that hold for each slice, not just the finished feature.

Worth a look

🤖 Generated with Claude Code

jamescrosswell and others added 3 commits September 27, 2026 22:15
The #34 and #174 breakdowns both shipped a CLI-only first task (#203, #213)
and put the dashboard command a task or three later. The reviewer then had
to approve PRs like #220 for plumbing the dashboard and agents call, which
unit tests cover better than a review does.

The Lead's test was "a user can do or see something", and running a CLI
command passes it. The vision already said a command can't have only a CLI
form, but read as a statement about the finished feature, not each slice.

- roles/lead.md: judge a slice by the vision's user, where the vision says
  they work; a command, API or script the product's own code or agents call
  is a layer; internal commands go under Tests, not acceptance criteria.
- docs/vision.md: the no-CLI-only rule holds per slice.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
a-team try shows a task's acceptance criteria as the reviewer's checklist,
so each item has to be something they check by using the product. #213's
listed CLI commands, a config key and exit codes.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
#220's summary and notes were about which script stop lives in and the
order of its writes. The reviewer judges the result by trying it; the
implementation story belongs in the commit messages.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@jamescrosswell
jamescrosswell marked this pull request as ready for review September 27, 2026 09:33
@jamescrosswell
jamescrosswell merged commit ae00fd3 into main Sep 27, 2026
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant