[JacobStephens2]

Summary

An Architect run whose Architecture review found no Strong candidate now ends as a success on the issue the review named, instead of failing with "published no plan".

 conclude(final line of the review)
   "Architecture review plan: <url>"
     check the plan, swap needs-triage for ready-for-agent
     stdout: <url>   stderr: plan <url> is ready for an agent          exit 0
   "Architecture review idea: <url>"
-    fail: the Architecture review published no plan                   exit 1
+    no gh call, no label change, no dispatch
+    stdout: <url>   stderr: no Strong candidate: the Architecture
+                            review filed the idea <url>               exit 0
   "Architecture review already filed: <url>"
-    fail: the Architecture review published no plan                   exit 1
+    no gh call, nothing filed or changed
+    stdout: <url>   stderr: no Strong candidate: <url> already covers
+                            the Architecture review's top
+                            recommendation, so it filed nothing       exit 0
   anything else
     fail: the Architecture review ended without the final line its prompt asks for

architect::run returns an Outcome (PlanReady, IdeaFiled, AlreadyFiled) holding the issue it ended on. Its Display is stderr's last line and its URL goes to stdout. The README has a new "No Strong candidate" section, and thirdshift help describes both outcomes. --plan-only stays required until #232.

Seam under test: the thirdshift binary, through tests/architect_run.rs with the fake claude and gh.

Evidence

• Before: a_review_that_files_an_idea_prints_its_url_and_changes_no_label failed: exit 1, with the Architecture review published no plan: it filed the idea https://github.com/acme/widgets/issues/8. The already-filed test failed the same way.
  After: both pass. The idea test asserts exit 0, the idea's URL alone on stdout, the last stderr line, the label still needs-triage, and one session. The already-filed test asserts the fake GitHub's state is identical before and after.

cargo test --test architect_run   21 passed; 0 failed
cargo test                        27 test binaries, all ok
cargo clippy --all-targets        clean
cargo fmt --check                 clean

Unaddressed findings

Standards

• Outcome mirrors the private Report (Duplicated Code / Repeated Switches). Report is what the review said and Outcome is what the Architect run did. They diverge in Architect run dispatches its plan as a Spec run or a Run #232, when a plan goes on to a dispatched run.
• Outcome as a { kind, url } struct to drop url(). The enum keeps each case's meaning on its variant, and url() is one match arm.
• conclude as a mild Mysterious Name. No clearer name came up; its doc comment says what it ends and on what.
• The two stderr sentences are spelled out in the code, README and tests (Shotgun Surgery). This is how the repo treats every user-facing line; the reviewer also advised against acting on it.

Spec

• thirdshift doesn't check the idea issue or the already-filed issue (open, in this repository, created during this run). The Ticket asks for the URL printed with nothing changed, and the parent Spec requires checks only before dispatch. A check would add a gh call and failure modes nobody specified. Worth a decision by the Day shift.
• No mid-run progress line for the idea path, as the plan path has with "published the plan". The outcome line already names the idea and the case; a second line would repeat it.
• The one-line architect summary in the README and help usage still says "publish a plan". That is the command's purpose; the "No Strong candidate" section covers the other outcomes.
• The help text and the missing-final-line message changed, though the Ticket names only the README. Both would have been wrong otherwise: the old message said "naming its plan".

Merge Danger

Door: two-way

Reverting restores the failure on a no-plan report. No data or labels are touched on these paths.

Blast Radius: small

Only thirdshift architect --plan-only. A script that treated exit 0 as "a plan is ready" now also gets exit 0 for an idea issue or an already-filed issue; stderr's last line tells them apart. The missing-final-line failure message is reworded.

Closes #231

🤖 Generated with Claude Code