[JacobStephens2]

Summary

The Architect run's conclusion now reaches GitHub through the Pass seam it already holds for its gates. A plan is viewed once, and its Edit is built from that view.

 trait pass::Outside
   open_issues / closed_issues / ready_issue / apply / step
+  issue(url)      # state, labels, createdAt (github::issue)
+  interrupted()   # interrupt::requested

 run
   gates(&mut on_github)
-  review(worktree, …)
-    conclude(final_message, origin, started, base)
+  review(&mut on_github, worktree, …)
+    conclude(outside, final_message, origin, started, base)
       mark_plan_ready
-        progress::step(published)
-        check_plan            # github::issue        (read 1)
-        interrupt::requested
-        progress::step(marking)
-        Edit::read(plan).apply()  # issue_labels  (read 2)
+        outside.step(published)
+        labels = check_plan   # outside.issue        (only read)
+        outside.interrupted()
+        outside.step(marking)
+        outside.apply(Edit::of(plan, labels, …))
       label_idea
-        Edit::read(idea).apply()
+        outside.interrupted(); outside.step(labelling)
+        outside.apply(Edit::of(idea, outside.issue(idea).labels, …))

• Edit::read is deleted, so building an Edit no longer does I/O.
• The Pickup run gets the two new operations through the shared adapter and doesn't use them.
• The Architecture review's starting line still goes through progress::step.

Evidence

• Before: the plan's checks and the idea labelling ran only end to end, against the fake gh (5 tests in tests/architect_run.rs).
  After: 13 new unit tests in src/architect.rs drive conclude with pass::InMemory and assert on the ending or cause and on the call record:

  plan ready           → [Step(published), View(8), Step(marking), Edit(-needs-triage, +ready-for-agent +architect-plan, others kept)]
  plan refused         → [Step(published), View(8)]  for closed / older / ready-for-human / needs-info / wontfix
  other repository     → [Step(published)]           no view
  interrupt (plan)     → [Step(published), View(8)]  no Edit
  interrupt (idea)     → []
  idea triaged         → needs-triage put back, Architect-Idea in another case not added twice
  idea view/Edit fails → "could not label the Architect idea #8: …"
  no final line        → []
  

• The five end-to-end tests named in the issue are deleted. The other 79 Architect run tests pass unchanged.
• cargo test (every suite), cargo clippy --all-targets and cargo fmt --check all pass.

Merge Danger

Door: two-way

This is an internal refactor. Users see the same progress lines, causes, labels, stdout, exit codes and Run notifications.

Blast Radius: Architect-runs

A plan now takes one fewer gh request. Its Edit is built from the labels viewed for the checks. A label someone adds between that view and the Edit is overwritten. Before, the Edit read the labels a second time, so it kept such a label.

Unaddressed findings

Standards

• Divergent Change, the Pass seam now serves the gates and the conclusion: skipped. The issue chose this on purpose ("The same Pass seam, not a new one").
• run still calls interrupt::requested() and review still calls progress::step directly: skipped. That check comes before the review and is outside the conclusion. The issue says the review's starting line stays where it is.
• Data Clumps, origin, started and base passed together: skipped. This is a judgement call outside this issue's scope, and the parameters were there before.
• Duplicated Call::Edit test helper in architect.rs and pickup.rs: skipped. It is a small test-only helper. A shared helper in pass::in_memory can come with a later seam change.
• viewed(bool, i64, …) test helper parameters aren't self-describing: skipped. Its doc comment explains them, and the gates' InMemory::issue(n, open, labels) uses the same style.

Spec

• a_plan_created_in_the_second_the_run_started_passes is beyond the listed cases: kept. It covers the spec's "created since the run started (to the second)".

Closes #386

🤖 Generated with Claude Code