You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Consider publishing specs and plans as pull requests instead of uploading them
to the issue, so reviewers can view planning artifacts as part of a PR and
hold discussion there.
Problems with the current mechanism
Specs and plans are published to the issue with patchmill set-spec and patchmill set-plan, which attach them to the issue as comments/attachments.
Clumsy to view: artifacts appear inline in issue comment bodies. Long
documents buried in a comment thread are hard to navigate, and each revision
adds another comment rather than updating a single current view.
No checkout: the artifacts are not on a branch. A reviewer cannot check
them out locally, open them in an editor, or run tooling against them.
No real review: issue comments do not support line-level review
comments, suggestions, or diff views between revisions. Feedback on a
specific section means quoting it in a new comment.
No review state: there is no approval mechanism comparable to PR
reviews; approval is tracked indirectly through labels.
What a PR would give us
The spec and plan live as files on a branch, viewable in the PR files view
with full Markdown rendering.
Line-level review comments and threaded discussion on the exact passage
under debate.
Reviewers can check out the branch locally.
Revision history through commits; changes between rounds are visible as
diffs.
Native approval state via PR reviews, which could replace or complement the
current label-based approval.
Questions to explore
How does a planning PR relate to the issue and to the later implementation
PR? One PR that starts with planning artifacts and grows implementation
commits, or separate PRs?
What happens to patchmill set-spec / set-plan and the existing
attachment envelope/checksum validation?
How do unattended runs create and update these PRs without human input?
This issue is the design and coordination umbrella. Implementation is split into independently reviewable sub-issues with explicit dependencies. The umbrella is marked agent-unsuitable; agents should claim only an unblocked child issue.
Summary
Consider publishing specs and plans as pull requests instead of uploading them
to the issue, so reviewers can view planning artifacts as part of a PR and
hold discussion there.
Problems with the current mechanism
Specs and plans are published to the issue with
patchmill set-specandpatchmill set-plan, which attach them to the issue as comments/attachments.documents buried in a comment thread are hard to navigate, and each revision
adds another comment rather than updating a single current view.
them out locally, open them in an editor, or run tooling against them.
comments, suggestions, or diff views between revisions. Feedback on a
specific section means quoting it in a new comment.
reviews; approval is tracked indirectly through labels.
What a PR would give us
with full Markdown rendering.
under debate.
diffs.
current label-based approval.
Questions to explore
PR? One PR that starts with planning artifacts and grows implementation
commits, or separate PRs?
patchmill set-spec/set-planand the existingattachment envelope/checksum validation?
(show-me visualizations, which would get a natural home as committed HTML
artifacts on the branch)?
Delivery overview
This issue is the design and coordination umbrella. Implementation is split into independently reviewable sub-issues with explicit dependencies. The umbrella is marked
agent-unsuitable; agents should claim only an unblocked child issue.Dependency order
#184is the first delivery slice. After it closes,#185,#186, and#187can proceed in parallel. They converge on#188, followed by#189and#190.GitHub native sub-issue and blocked-by relationships are authoritative. The
blockedlabels mirror those dependencies for Patchmill triage.Acceptance criteria
PRs.
above.